Unbundling household preferences for improved sanitation: A choice experiment from an urban settlement in Nicaragua

Journal of Environmental Management 218 (2018) 477e485

Contents lists available at ScienceDirect

Journal of Environmental Management journal homepage: www.elsevier.com/locate/jenvman

Research article

Unbundling household preferences for improved sanitation: A choice experiment from an urban settlement in Nicaragua
squez a, *, Jessica Alicea-Planas b William F. Va a b

Department of Economics, Fairfield University, 1073 North Benson Rd, Fairfield, CT 06824, USA Egan School of Nursing & Health Studies, Fairfield University, 1073 N. Benson Rd., Fairfield, CT 06824, USA

a r t i c l e i n f o

a b s t r a c t

Article history: Received 4 February 2017 Received in revised form 20 March 2018 Accepted 20 April 2018 Available online 27 April 2018

Many urban settlements in developing countries still lack access to sanitation services, which puts the environment and population health at risk. The lack of knowledge on household preferences for improved sanitation has been an impediment to extending conventional and onsite sanitation infrastructure. This study implemented a choice experiment to elicit households' willingness to pay for the disposal of different types of waste (i.e. wastewater, excreta, and rainwater) in an urban settlement in Nicaragua. Generalized multinomial logit models were estimated to account for heterogeneity among respondents in both choice behavior and preferences for specific attributes. Findings indicate that households are willing to pay a considerable amount of money for improved disposal of wastewater, excreta, and rainwater. However, households have stronger preferences for wastewater and excreta removal than for disposal of rainwater. © 2018 Elsevier Ltd. All rights reserved.

Keywords: Sanitation Choice experiment Willingness to pay Urban settlements Health risk Nicaragua

1. Introduction The lack of sanitation services can have negative environmental and health consequences, especially in urban settlements with high residential densities where the squalor and health risks associated with unimproved sanitation are particularly acute (McGranahan, 2015). Inappropriate disposal of wastewater and rainwater presents considerable health risks for urban settlements as stagnant water is often a breeding ground for mosquitoes that can transmit diseases (Sharma, 2014). Unimproved latrines may also put the population at an increased risk for illness as excreta could reach and contaminate underground water sources (Buttenheim, 2008; Graham and Polizzotto, 2013). Moreover, in urban settlements, space constraints may influence decisions to dig new latrines when old ones reach capacity. Under these circumstances, appropriate disposal of runoff and excreta may have considerable environmental and health benefits for households in urban settlements. Sanitation infrastructure has been quite effective in fighting
squez and Aksan, 2015; Zwane diseases in developing countries (Va and Kremer, 2007). Yet, the progress in expanding improved

* Corresponding author. E-mail addresses: wvasquez@fairfield.edu (W.F. V
asquez), jplanas@fairfield.edu (J. Alicea-Planas). https://doi.org/10.1016/j.jenvman.2018.04.085 0301-4797/© 2018 Elsevier Ltd. All rights reserved.

sanitation coverage in those countries has been slow (Van Minh and Nguyen-Viet, 2011). For example, out of 191 state members of the United Nations, only 95 met the sanitation target within the Millennium Development Goals framework (United Nations, 2015). As a point of comparison, 147 countries met the drinking water target. Lack of information on household preferences and willingness to pay for improved sanitation services can be an important impediment to the implementation of conventional and onsite sanitation infrastructure. Winters et al. (2014) argue that the demand for enhanced sanitation is less expressed than the demand for other services because citizens are embarrassed to talk about sanitation issues. Improved understanding of household preferences for sanitation services may help in prioritizing public investments in sanitation infrastructure by showing the relative importance of those services to citizens. This is particularly important in developing countries where the needs are multiple and resources are limited. Moreover, unbundling household preferences for appropriate disposal of different types of waste may facilitate selecting among conventional and onsite technologies to extend sanitation services (e.g. septic tanks, ecological latrines and biological gardens). Choice experiments have been shown to be an appropriate method to elicit household preferences for improved public services in developing countries (Bennet and Birol, 2010). Compared to

478


squez, J. Alicea-Planas / Journal of Environmental Management 218 (2018) 477e485 W.F. Va

other preference elicitation techniques (e.g. contingent valuation method), choice experiments are more suitable to estimate willingness to pay for different attributes of the public service in question (Birol et al., 2006; Hanley et al., 2001). In the case of sanitation, choice experiments could provide useful insights to design and provide sanitation services that respond to households' willingness to pay for disposing different types of waste (e.g. excreta, wastewater, and rainwater). Few studies have conducted choice experiments to investigate local preferences for improved sanitation in developing countries, and most of those studies have done so in areas where conventional sewerage systems or improved on-site technologies (e.g. absorption pits and septic tanks) already exist (e.g. Birol and Das, 2010; Genius et al., 2012; Woldemariam et al., 2016). Similar information would be useful for urban settlements where basic sanitation is nonexistent as residents may face considerable environmental and health risks. Ndunda and Mungatana (2013), for example, found that farmers living in informal settlements in Nairobi (Kenya) would be willing to pay for improved wastewater treatment primarily because it would increase the amount and quality of water available for irrigation. That study, however, focused on wastewater management thus neglecting other types of waste that may also jeopardize the environment and population health (e.g. excreta and rainwater). In this study, we have implemented a choice experiment to unbundle household preferences for improved sanitation in the form of proper disposal of wastewater, excreta, and rainwater in the urban settlement of Nueva Vida, Nicaragua. As many other urban settlements in developing nations, Nueva Vida is subject to environmental and health risks due to the lack of sanitation infrastructure. Respondents' choices were analyzed using generalized multinomial logit models to account for two types of respondent heterogeneity: 1) taste heterogeneity (i.e. respondents may value attributes differently), and 2) choice behavior heterogeneity (i.e. some respondents may show more random choice behavior than others). Survey results indicated that households are aware of the environmental and health consequences of current sanitation conditions. Findings also suggested that households are willing to pay a considerable amount of money for disposal of wastewater, excreta, and rainwater. Households, however, had stronger preferences for wastewater and excreta removal than for disposal of rainwater. The rest of this paper is organized as follows. The next section describes the urban settlement of Nueva Vida, with an emphasis on describing current sanitation conditions in that community. Section 3 explains the survey design including the choice experiment. Section 4 introduces the analytical framework and econometric approach followed to analyze responses to the choice experiment. Section 5 presents the estimation results. Section 6 concludes the paper with a discussion of the results and potential policy implications. 2. Study site This study was conducted in the urban settlement of Nueva Vida, located in the municipality of Ciudad Sandino at approximately 17 km west of the capital city of Managua, Nicaragua. This settlement was initially formed by households that were displaced by hurricane Mitch in 1998. According to Universidad Centroamericana (2016), the population is estimated at 8085 residents in 1724 households. This settlement currently occupies an area of 2.7 km squared, with a high population density of 2994 people per kilometer squared. The settlement is organized in five territorial zones locally referred to as etapas. Universidad Centroamericana (2016) produced an urban diagnostic of Nueva Vida, which portrayed precarious living conditions, particularly in terms of sanitation. The settlement has a drinking

water system in place to which most, if not all, households are connected. In contrast, Nueva Vida does not have a conventional sanitation system to dispose excreta, wastewater, and rainwater. Almost 60% of households use an unimproved latrine and almost 35% use a toilet, although few of them (if any) are connected to a septic tank that can be periodically emptied as those services are nonexistent for the community of Nueva Vida. Stagnant water is another sanitation issue given that Nueva Vida lacks a sewerage system. Although some households have a sort of artisanal absorption pit, a vast majority of households let wastewater flow onto the streets. Because the settlement is located on a relatively flat area, and most streets are not paved, wastewater tends to stagnate. This issue is even worse in the rainy season. Given the high population density of Nueva Vida, the settlement seems to be rapidly reaching its capacity for digging latrines. Inappropriate disposal of excreta represents a latent source of pollution for the aquifer from which Nueva Vida gets its water. Therefore, it is imperative to provide improved sanitation services that can dispose excreta in an appropriate manner. In addition, stagnant water currently puts the population health at risk as it facilitates the proliferation of bacteria, parasites, and mosquitoes. During our visits to Nueva Vida, we observed several barefooted children having direct contact with stagnant water. According to Universidad Centroamericana (2016), Nueva Vida's inhabitants rank sanitation services at the top of their many needs. Against this backdrop, a better understanding of willingness to pay for improved disposal of wastewater, excreta, and rainwater may be useful to prioritize public investments which, in turn, may have substantial environmental and health benefits. 3. Survey and choice experiment design An interdisciplinary team of researchers from Fairfield University (United States) and Universidad Centroamericana (Nicaragua) designed a household survey to investigate household behaviors, perceptions, and preferences related to water and sanitation in the community of Nueva Vida. The final survey questionnaire had a total of 38 questions (some of them with multiple parts) organized in six sections. First, respondents were asked about current conditions of water services. In the second section, respondents reported their behaviors regarding the use, storage, and treatment of tap water. The third section measured water users' satisfaction levels and inquired about perceptions of water quality. The fourth section included questions on the water-health nexus. The fifth section elicited household preferences for improved sanitation services using a choice experiment. Finally, the survey gathered respondents' sociodemographic information. For the survey implementation, Nueva Vida was stratified into five geographical zones and parcels in each stratum were randomly selected from a map that Universidad Centroamericana (2016) had generated as part of an urban diagnostic project in 2016. This map was the best available framework for sampling given that mailing addresses are not used in Nicaragua. A total of 419 households were randomly selected according to a geographically-stratified random sampling strategy, and 398 completed questionnaires were obtained via an in-person interview process implemented in July 2016. Interviewers were instructed to ask for the adult who usually pays the water bill. Local preferences for improved sanitation services were elicited through an unlabeled choice experiment. This experiment included three binary attributes and an increase in water bills to pay for improved services. We defined the binary attributes representing sanitation improvements in terms of the waste to be disposed rather than based on potential technological projects aimed at disposing those wastes. Those attributes consisted of disposal of


squez, J. Alicea-Planas / Journal of Environmental Management 218 (2018) 477e485 W.F. Va Table 1 Choice experiment design. Attributes

Levels

Indicators

Disposal of wastewater Disposal of excreta Disposal of rainwater Increase in monthly water bill

No; Yes No; Yes No; Yes 0; 50; 100; 150; 200

WASTEWATER (0/1) EXCRETA (0/1) RAINWATER (0/1) e

wastewater, excreta, and rainwater. The base level of each attribute was the status quo, and the improvement was the supply of disposal services. This functional approach allowed us to derive willingness-to-pay estimates for specific sanitation services that can be compared with the cost of any sanitation technology. A labeled choice experiment defined in terms of sanitation technologies would help eliciting preferences for technologies included in the experiment only. Hence, an unlabeled choice experiment defined in functional terms is more flexible to consider a variety of sanitation technologies. The cost attribute could take any of five levels from C$ 0 to C$ 200 (about US$ 7) in increments of C$ 50 (see Table 1). At the time of implementing the survey, the exchange rate was C$ 28.75 per US$ 1. The zero-cost level was included to reflect the possibility of receiving free services. There were non-governmental organizations interested in providing sanitation services at zero cost and volunteer groups willing to contribute financial resources and labor to install on-site technologies (e.g. ecological latrines and biological gardens). The ability to analyze free and priced services is an important advantage of choice experiments (Koo et al., 2010; Papies et al., 2011). A total of 12 choice tasks were selected following a D-efficiency experimental design (with priors for coefficients set at zero), each of which had two options that vary in attribute levels and the status quo alternative as a third option. The choice tasks were grouped in three sets so each respondent was asked to perform four choices.1 Using a script that was previously revised by residents and staff members of a local non-governmental organization, interviewers explained to respondents the choice task they were asked to perform (see Appendix A). First, respondents were reminded about current sanitation conditions in the community and corresponding health and environmental risks. Then, interviewers introduced the improved waste disposal attributes and the payment vehicle. Respondents were reminded about their budget constraint to imprint realism to the choice task. Each choice included visual aids (i.e. pictures taken in the community that related to the waste to be disposed) so respondents could clearly identify the attributes of each option (Mangham et al., 2009). Before performing the four choice tasks, respondents were presented with an example to practice choosing among options (see appendix A). Respondents were informed that the example was a practice round. The example was not included in the estimations but had two alternative purposes that were achieved by including a dominant option into the choice task (i.e. an option that was unequivocally better than the other two options). First, this example helped the respondent become familiar with the choice task. Along with the visual aids, this practice example was expected to reduce the cognitive burden that choice experiment may impose on respondents. Second, it allowed us to identify respondents who did not understand the choice task (i.e. respondents who chose an option other than the dominant one).

1 Given our experimental design, there were 40 options (2  2 x 2  5) that would amount to 780 pairs to be compared (40  39/2). Those are too many comparisons and, therefore, 12 pairs were selected. Then, three blocks of four choice tasks each were selected to avoid respondent fatigue.

479

Choice models are estimated using both the entire sample of respondents and a subsample that excludes those respondents who behaved irrationally by choosing a dominated alternative in the practice example. Lancsar and Louviere (2006) argue that deleting seemingly irrational respondents may reduce the statistical efficiency of the estimated choice models and may induce sample selection bias. However, some studies have found more conservative and arguably more accurate willingness-to-pay estimates by excluding respondents who chose a dominated alternative (e.g. ~ o et al., 2012). Campbell, 2007; Solin 4. Analytical framework and econometric modeling The Random Utility Model (RUM) provides a theoretical framework to analyze the results of choice experiments. In this model, the utility derived from an alternative depends on observed and unobserved attributes of the alternative and individual. While observed attributes are depicted by explanatory variables, unobserved ones are represented as random variables. Given that individuals are assumed to choose the alternative that gives them the highest utility, there is a monotonic relationship between the conditional indirect utility that the individual n derives from alternative i at time t (i.e. Vnit) and the probability of choosing alternative i over any other alternative j [i.e. Pr(i) ¼ Pr(Vnit > Vnjt)]. Consequently, individuals' choices can be analyzed using a multinomial model. A challenge for estimating choice models is that the association between utility and attributes can vary across individuals due to their heterogeneity. Prior studies have proposed different choice modeling strategies to account for the unobserved heterogeneity of respondents. For instance, under the assumption that most of the heterogeneity across respondents is due to taste differentials, McFadden and Train (2000) propose adding random coefficients to the average level of utility derived from a given attribute (i.e. the mixed logit model). On the other hand, Louviere et al. (1999, 2008) argue that respondents are heterogeneous because choice behavior is more random for some respondents than others. This implies that the scale of their errors is greater. Hence, Louviere et al. (1999) proposed allowing the scale parameter to vary across respondents so it proportionally modifies the average effect of all attributes (i.e. the so-called scale-heterogeneity multinomial logit model).2 In the context of this study, respondents could show taste heterogeneity by assigning different values to the disposal of a given waste (wastewater, excreta, or rainwater). Additionally, respondents could process the choice tasks differently, thus showing heterogeneity in choice behavior. Fiebig et al. (2010) proposes a generalized multinomial logit (GMNL) model that accounts for heterogeneity in both taste and choice behavior. This approach introduces random components for both types of heterogeneity into the indirect utility level V for individual n as follows:

Vnit ¼ ½sn b þ ghn þ ð1  gÞsn hn Xit þ εnit

(1)

sn ¼ expðdZn þ twn Þ

(2)

In equation (1), Xit is the vector of sanitation attributes and additional payment proposed in the choice experiment, sn is the

2 Alternatively, one could use conditional logit models with interactions between choice attributes and respondent characteristics to investigate heterogeneity among respondents. This approach, however, does not control for unobservable heterogeneity. The latent-class logit specification is another modeling approach commonly used to control for heterogeneity among respondents. While appealing for generating discrete types of consumers, it has been suggested that latent-class models understate respondents' heterogeneity in choice data (Allenby and Rosi, 1998; Elrod and Keane, 1995).

480


squez, J. Alicea-Planas / Journal of Environmental Management 218 (2018) 477e485 W.F. Va

scale parameter that varies across individuals to account for heterogeneity in choice behavior, and hn is a vector of normally distributed random variables that depict unobserved taste heterogeneity.3 b is the conformable vector of coefficients to be estimated in order to depict the average utility derived from each attribute. g is a (weight) estimable parameter restricted to be between zero and one. In equation (2), Zn is a vector of individual characteristics that can depict some observable heterogeneity in choice behavior, and wn is a random variable that follow a standard normal distribution included to model unobserved heterogeneity in choice behavior. d and t are parameters to be estimated. Note that the average utility derived from each attribute (b) is modified by the scale parameter (sn) that varies across respondents according to equation (2). In addition, a weighted average of random coefficients is included to model taste heterogeneity (g hn) and potential interactions between taste and choice behavior heterogeneity (g sn hn). The special case in which g equals one implies that taste and choice behavior heterogeneity are independent from each other. When g equals zero, taste heterogeneity is proportional to sn. Fiebig et al. (2010) GMNL model nests other approaches commonly used to analyze choice experiments. The mixed logit model can be derived from equation (1) if the scale parameter sn is normalized to one for all respondents. The scale-heterogeneity multinomial logit model is also a special case of the GMNL model in which the variance of random parameters hn equals zero. The GMNL model also nests the multinomial logit model [i.e. sn ¼ 1 and Var(hn) ¼ 0]. Therefore, their estimation approach allows for empirically selecting the model specification that best fits the results of a choice experiment. By including a payment variable among the attributes, one can estimate the marginal willingness to pay (also known as partworth) for moving from one attribute level to another. It is standard practice to estimate choice models in preference space (as specified in equations (1) and (2)), and then transform the estimated coefficients to monetary values. This can be achieved by multiplying those coefficients by the negative reciprocal of the coefficient corresponding to the payment variable (i.e. eb/bPAYMENT). However, this approach may result in unconventional, heavily skewed distributions for willingness-to-pay estimates particularly when values of the denominator are close to zero, which is possible under most typical distributions (Scarpa et al., 2008). Alternatively, choice models can be estimated in willingness-to-pay space by constraining the coefficient corresponding to the payment variable to negative one (Cameron, 1988; Train and Weeks, 2005). This approach allows for making distributional assumptions directly on the monetary values of each level of associated attributes, which provides willingness-to-pay estimates that are presumably more precise. For the data at hand, the traditional approach of estimating models in preference space yielded statistically insignificant willingness-to-pay estimates. Therefore, we use the user-written Stata command gmnl, which implements a maximum simulated likelihood approach (Gu et al., 2013), to estimate GMNL models (i.e. equations (1) and (2)) in willingness-to-pay space in order to analyze choice responses on the proposed sanitation improvements while accounting for potential heterogeneity in taste and choice behavior among respondents. Coefficients and random variables were modeled using normal distributions. Scarpa et al. (2008) and Balogh et al. (2016) present recent examples of choice models estimated in willingness-to-pay space.

3 Although other distributions have been proposed (e.g. triangular distribution), the normal distribution is most commonly used to depict taste heterogeneity in choice models (Fiebig et al., 2010).

Fig. 1. Top 10 illnesses perceived to result from stagnant water.

5. Survey and estimation results Survey results suggested that respondents were aware of the health consequences of current sanitation conditions in Nueva Vida. More than 95% of the respondents believed that stagnant wastewater and rainwater on the streets represented a risk for their health. Fig. 1 shows a list of illnesses that sampled households perceived to result from stagnant water. Households seem to be primarily concerned with the possibility that stagnant water can become a breeding ground for mosquitoes that transmit illnesses such as Dengue, Chikungunya, Zika, and Malaria. In the last years, the prevalence of those illnesses has been a public health issue for Nicaragua, as well as for other Latin American countries (Belli et al., 2015; Petersen et al., 2016). Respondents also believed that stagnant water can cause gastrointestinal illnesses such as diarrhea, vomiting, and stomachache. Respondents were asked to report their satisfaction with the accumulation of residential wastewater and rainwater (see Fig. 2). While a vast majority of respondents were dissatisfied with that sanitation issue, dissatisfaction levels were higher for residential wastewater than for rainwater. Rainwater can be more concerning than wastewater in terms of the amount of stagnant water. However, wastewater is a constant problem while rainwater is a periodic one. This may explain the particular discontent with the lack of infrastructure to dispose wastewater. Probit models estimated to identify the profile of respondents who showed some level of dissatisfaction with water accumulation from residential wastewater, rainwater, and both indicated that household income was the only significant factor related to respondents' dissatisfaction. More affluent respondents were more likely to be dissatisfied with current sanitation issues in the community. Other characteristics were found unrelated to dissatisfaction with stagnant water.4 Residents' beliefs regarding the health consequences of stagnant water, as well as their discontent with the sanitation issue, would indicate that there is a latent demand for improved sanitation in this settlement. The choice experiment implemented in this study can reveal that demand. Table 2 shows results from the generalized multinomial logit model estimated using a total of 1544 choices. Random effects depicting taste heterogeneity are statistically insignificant, and the weight parameterg is insignificant as well. In contrast, the coefficient t is positive and statistically significant. These results indicated that most of the heterogeneity among respondents is shown in their choice behavior presumably because they process the choice tasks differently (i.e. some respondents

4 Estimation results from probit models are not presented here but are available from the corresponding author upon request.

Fig. 2. Satisfaction with accumulation of wastewater and rainwater.

Table 2 Generalized multinomial logit model. Generalized Multinomial Logit Modela

WASTEWATER EXCRETA RAINWATER

t g Choice Tasks

Part-Worth Estimates (C$)

St. Dev. of Random Effects

106.07 (17.03)b 104.41 (22.77)b 52.79 (16.75)b 45.98 (2.76)b 0.22 (0.18) 1544

12.90 (10.22) 13.57 (10.97) 10.73 (8.60) e e

a

Standard errors (S.E.), reported in parenthesis, are clustered by respondent to account for potential correlation among choices made by the same individual. b These coefficients are significant at 1% level.

show more random choice behavior than others).5 Estimated coefficients of sanitation attributes are positive and statistically significant, indicating that respondents were willing to pay for sanitation improvements. The average monthly willingness to pay for wastewater disposal was C$ 106.07 (US$ 3.69) (95% CI: C$ 72.71e139.46), equivalent to 3.19% of the average household income. As a point of comparison, the average household pays a monthly water bill of C$ 92 (US$ 3.20). At C$ 104.41 (US$ 3.63) (95% CI: C$ 59.77e149.04), the willingness to pay for excreta disposal was statistically similar to the willingness to pay for wastewater disposal (Z ¼ 0.11, p ¼ 0.912). Respondents were also willing to pay for rainwater disposal. The average household would pay C$ 52.79 (US$ 1.84) for rainwater disposal (95% CI: C$ 19.96e85.62), equivalent to 1.59% of its income. The willingness to pay for rainwater disposal was statistically lower than the willingness to pay for disposal of wastewater (Z ¼ 22.14, p ¼ 0.000) and excreta (Z ¼ 3.48, p ¼ 0.001).6 Choice experiments may impose a substantial cognitive burden on respondents due to the complexity of the choice task (Hoyos, 2010). Hence, there could have been respondents who did not fully understand the choice task. Given that all respondents were given the same example with a dominant option, the choice responses to that example can be used to identify respondents who

5 Given the heterogeneity among respondents, the conditional logit model may yield biased results. In this study, WTP estimates derived from the conditional logit model indicated that respondents had stronger preferences for the disposal of rain water than for the disposal of residential wastewater (see Appendix B). After controlling for heterogeneity among respondents, we found the opposite. 6 We tested for interaction effects between WASTEWATER and EXCRETA, as some respondents could be aware of sanitation systems that can dispose both at the same time (e.g. drainage systems and septic tanks). The interaction term was found statistically insignificant, ruling out the possibility of perceived interdependence between wastewater and excreta disposal.

did not understand the choice experiment. Results indicate that 28.2% of the respondents did not choose the dominant option in the task example. Under the assumption that those respondents did not understand the choice task, they were excluded from the analysis. A number of probit models were estimated to identify the profile of respondents who did not choose the dominant option in the task example. Results indicated that observable characteristics (income, sex, age, education, household size, home ownership, etc.) were not related to that choice response. Table 3 shows estimation results obtained using the subsample of respondents who chose the dominant option in the task example. Model 1 assumes that scale heterogeneity is totally due to unobservable characteristics. Model 2 relaxes this assumption by including household income (INCOME) and size (HHSIZE) to control for observable heterogeneity in respondents' choice behaviors. Random effects depicting taste heterogeneity (h) were statistically significant in both models. The parameters t were statistically significant as well. These results indicate that this subsample of respondents is heterogeneous in both taste and choice behavior. Moreover, the comparison of these results with estimates presented in Table 2 suggests that respondents who did not understand the choice tasks (i.e. seemingly irrational respondents) may have more random choice behaviors than rational respondents. Thus, including (seemingly) irrational respondents may increase choice behavior heterogeneity to the extent that taste heterogeneity becomes relatively insignificant. Results also indicate that, for the subsample of rational respondents, there was a significant interaction between taste and choice behavior heterogeneity as the g coefficient was statistically less than one at a 1% significance level (Z ¼ 12.25) and statistically different than zero (see Table 3). This suggests that the variance of taste heterogeneity increased among respondents with more random choice behavior.7 Estimation results in Table 3 also demonstrated significant willingness to pay for disposal of wastewater, excreta, and rainwater (see Table 3). Based on Model 1, which yielded more conservative estimates, respondents who seem to have understood the choice task would pay an average of C$ 93.60 (US$ 3.26) for

7 Hess and Rose (2012) argue that it is empirically difficult to disentangle choice behavior heterogeneity from taste heterogeneity. Moreover, they argue that random scale models, including the GMNL model, can be a better fit for some datasets because those models provide for more flexible distributions, rather than due to their ability to separate scale and taste heterogeneity. This implies that the GMNL model may not be appropriate to identify types of heterogeneity and, consequently, these results should be interpreted with caution. Nevertheless, the GMNL model is suitable to estimate respondents' willingness to pay for observed attributes, which is the primary objective of this study.

Table 3 Using subsample of respondents who seemingly understood the choice task example.a. Generalized Multinomial Logit Model

WASTEWATER EXCRETA RAINWATER INCOME HHSIZE

t g Choice Tasks a b

Generalized Multinomial Logit Model

Part-Worth Estimates (C$)

St. Dev. of Random Effects

Part-Worth Estimates (C$)

St. Dev. of Random Effects

93.60 (12.74)b 80.01 (10.17)b 69.25 (15.16)b e e 27.63 (1.06)b 0.18 (0.07)b 973

13.09 (4.97)b 16.33 (5.99)b 12.86 (4.78)b e e e e

100.14 (5.62)b 95.35 (5.96)b 74.41 (4.59)b 0.60 (0.09)b 0.20 (0.07)b 3.94 (0.09)b 0.09 (0.02)b 932

29.96 (5.52)b 29.53 (6.20)b 26.47 (5.02)b e e e e

Standard errors (S.E.), reported in parenthesis, are clustered by respondent to account for potential correlation among choices made by the same individual. These coefficients are significant at 1% level.

wastewater disposal and C$ 80.01 (US$ 2.78) for excreta disposal, a differential that is statistically insignificant (Z ¼ 0.42, p ¼ 0.675). Those willingness-to-pay estimates are equivalent to 2.8% and 2.4% of the average household income, respectively. They were also willing to pay an average of C$ 69.25 (US$ 2.41) for rainwater disposal, which is equivalent to 2.1% of the average income. The willingness-to-pay differential between rainwater and wastewater disposal is statistically significant (Z ¼ 3.38, p ¼ 0.001), as is the differential with respect to excreta disposal (Z ¼ 2.81, p ¼ 0.005). ~ o et al., Consistent with prior studies (e.g. Campbell, 2007; Solin 2012), the subsample of respondents who understood the choice task showed a more conservative willingness to pay for wastewater and excreta disposal compared to estimates based on the full sample. In contrast, the willingness to pay for rainwater disposal is higher for the subsample of respondents who understood the ~ o et al. choice task than for the full sample of respondents. As Solin (2012) argue, respondents who did not understand the choice task may have different preferences that generate extreme or anomalous values, which may bias willingness-to-pay estimates.8 Model 1 does not control for observable heterogeneity in choice behavior (i.e. d coefficients in equation (2) are assumed to be equal to zero). To identify observable characteristics that could explain some of the heterogeneity in choice behavior, we included variables for household income and size (INCOME and HHSIZE, respectively) into Model 2. Findings indicate that both income and household size can explain some of the heterogeneity in choice behavior, although some heterogeneity remains unobservable as indicated by a statistically significant t coefficient.9 The scale parameter sn decreases with both household income and size. Model 2 provides estimates comparable to results obtained from Model 1 in terms of sign and significance. That is, households were willing to pay a similar amount for disposal of wastewater and excreta, and a lower amount for rainwater (see Table 3). The differential between willingness to pay for disposal of wastewater and excreta reported in Table 3 was statistically insignificant (Z ¼ 1.38, p ¼ 0.167). In contrast, the willingness to pay for rainwater disposal was significantly lower compared to wastewater disposal (Z ¼ 10.55, p ¼ 0.000) and excreta disposal (Z ¼ 2.40, p ¼ 0.016).

8 As a reviewer noted, choices made by respondents who did not understand the choice task may not reflect their true preferences. This would also produce biased willingness-to-pay estimates. 9 The significance of the t coefficients in Tables 2 and 3 also indicate that there is choice behavior heterogeneity even after controlling for taste heterogeneity. Under these circumstances, modeling approaches that do not control for choice behavior heterogeneity may yield biased results (e.g. Silano and Ortúzar, 2005). For example, WTP estimates obtained through mixed logit models are approximately four times greater than those presented here. We also estimated a latent-class logit model with two classes. WTP estimates were insignificant for the first group. For the second group, estimates of the WTP for excreta and rainwater disposal are comparable to the results presented in Table 3, and about 60% of the WTP for wastewater disposal. However, respondents' characteristics (i.e. income and household size) were insignificant to distinguish between classes. Based on those findings, generalized multinomial logit models are deemed more appropriate for the data at hand than modeling approaches that do not control for scale heterogeneity.

Table 4 Conditional logit model including interaction terms with Household income.a. Part-Worth Estimates (C$) WASTEWATER WASTEWATER x INCOME EXCRETA EXCRETA x INCOME RAINWATER RAINWATER x INCOME

65.48 (15.46)b 7.77 (1.94)b 68.70 (17.81)b 7.84 (2.08)b 95.08 (33.16)b 0.23 (3.87)

a Standard errors (S.E.), reported in parenthesis, are clustered by respondent to account for potential correlation among choices made by the same individual. b These coefficients are significant at 1% level.

To further investigate the heterogeneity among respondents, we estimated a conditional logit model with interaction terms between income and the attributes included in the choice experiment (see Table 4). Results indicate that respondents' willingness to pay for disposal of wastewater and excreta increases with income. Interaction terms suggest that respondents would pay an additional amount of approximately C$ 7.80 (US$ 0.27) for wastewater and rainwater disposal when household income increases by C$ 1000 (see Fig. 3). In contrast, the willingness to pay for rainwater disposal does not vary with income as indicated by the insignificant coefficient of the interaction term between RAINWATER and INCOME. The implicit assumption behind the conditional logit model presented in Table 4 is that interaction terms with income depicted the heterogeneity among respondents. Yet, there could still be some unobserved heterogeneity that is not explained by differentials in income. Therefore, these results should be interpreted with caution. 6. Discussion and conclusions This paper has investigated local preferences for sanitation improvements in the urban settlement of Nueva Vida, Nicaragua. Currently, unimproved latrines and stagnant wastewater and rainwater represent substantial environmental and public health risks for the community. Survey results indicated that respondents were aware of those health risks and were discontent with current sanitation conditions in the settlement. A choice experiment was implemented to unbundle preferences for improved sanitation in terms of willingness to pay for disposal of residential wastewater, excreta, and rainwater. After controlling for taste and choice behavior heterogeneity, estimates indicated that the average respondent was willing to pay between 2.8% and 3.2% of the household income for disposal of residential wastewater (depending on the sample of respondents used to estimate that willingness to pay). Compared to the average water bill of C$ 92 (US$ 3.20), households would be willing to pay at least the same amount for wastewater disposal. Willingness-to-pay estimates were statistically similar for disposal of excreta (2.4%e3.14% of average income), suggesting that those sanitation services were equally preferred by respondents. On the other hand, respondents deemed rainwater disposal as less important, demonstrated by a

Fig. 3. Part-Worth Estimates by Household Income Levels (with 95% confidence intervals)a. (a) Part-Worth of Disposal of Wastewater. (b) Part-Worth of Disposal of Excreta. a Income was measured in intervals of 1000 Cordobas (i.e. 1 ¼ 1 to 1000 Cordobas; 2 ¼ 1001 to 2000 Cordobas; …. ; 10 ¼ 9001 to 10,000 Cordobas; 11 ¼ more than 10,000).

relatively lower willingness to pay for this service (1.6%e2.2% of the average income). These estimates validated that there is a latent demand for improved sanitation in the community of Nueva Vida. The willingness to pay for rainwater disposal is about 50e75% of what they would pay for disposal of wastewater and excreta. While appropriate disposal of wastewater and excreta can be achieved with private investments (e.g. installing biological gardens and ecological latrines at home), rainwater disposal requires collective action. McGranahan (2015) contends that the need for collective action is one of the primary challenges that urban settlements face in extending sanitation services. It can be argued that the willingness to pay for rainwater disposal is relatively low because households perceived that other community members will not pay for that service. This result could also reflect that households are willing to pay more for disposing waste that they produce than for rainwater which they did not produce. Interestingly, the primary health concerns identified by respondents (Zika, Dengue and Chikungunya) are strongly associated with stagnant water, which often has accumulated after rainfall. The finding that households are willing to pay less for rainwater disposal than for disposal of wastewater and excreta deserves further attention. Conventional sewerage systems may be difficult and costly to implement particularly in urban settlements with high levels of poverty, as income poverty may depress the demand for sanitation (Gross and Günther, 2014). As McGranahan (2015) argues, low cost

alternatives will be necessary if universal provision of sanitation services is to be achieved. Investing in infrastructure that people are willing to pay for may be crucial for long-term sustainability, especially in developing countries where resources are limited. Given that the choice experiment was designed in functional terms, willingness-to-pay estimates presented here can be compared to the cost of providing sanitation services under different alternatives (i.e. conventional sewerage systems and onsite technologies). For example, biological gardens represent an alternative for residential wastewater disposal. Installing a biological garden would cost about C$ 2300 (about US$ 75), which could be paid in approximately 29 months if households borrow that money at a monthly rate of 1% and monthly payments are set at the conservative willingness-to-pay estimate of C$ 93.60.10 Excreta disposal could be achieved by installing an ecological latrine that requires minimal maintenance. However, at a cost of C$ 15,500 or US$ 500 ~ ez Soto, 2010), the estimated willingness to pay for excreta (Nun removal would be insufficient to pay for an ecological latrine at current interest rates. If willingness-to-pay projections are aggregated over the 1724

10 The Central Bank of Nicaragua reports an annual interest rate of 11.89% (equivalent to a monthly rate of about 1%) for long-term loans given by commercial banks in August 2017 (see http://www.bcn.gob.ni/estadisticas/monetario_ financiero/financiero/tasas_interes/ponderadas/index.php).

484


squez, J. Alicea-Planas / Journal of Environmental Management 218 (2018) 477e485 W.F. Va

households estimated to live in Nueva Vida, over the course of 1 month at least C$ 161,366 (US$ 5613) could be collected from wastewater disposal, C$ 137,937 (US$ 4798) from the disposal of excreta, and C$ 91,010 (US$ 3166) from rainwater disposal. That is, a monthly total of C$ 390,313 (US$ 13,577) could be utilized to cover the costs of installing and maintaining a conventional drainage system. The national water and sanitation company (ENACAL) could use this estimate to assess whether it is viable to implement a conventional sanitation system in this community. ENACAL might also consider that improved sanitation services can alleviate poverty by reducing health costs and missed work days due to illness, as well as increasing worker productivity, gender equality and environmental sustainability (Van Minh and Nguyen-Viet, 2011). Our estimates can include some of those benefits given that respondents associated some illnesses with current sanitation conditions in their community. Yet, it would be a logical extension to this study to measure specific benefits related to improved sanitation such as reductions in the cost of illnesses. Those analyses would provide a lower bound of the willingness to pay for improved sanitation. As a last caveat, it should be mentioned that the willingness-topay estimates are large in comparison to the average income reported by respondents. This is consistent with considerable dissatisfaction and health concerns associated with their current sanitation conditions. Yet, it is possible that willingness-to-pay estimates are upwardly biased due to the hypothetical nature of choice experiments. We followed best practices in designing the choice experiment to reduce hypothetical bias (e.g. we reminded respondents about their budget constraints), and estimated GMNL models with different subsamples to provide conservative estimates. However, just to proceed with caution, our estimates can be considered an upper bound of households' willingness to pay for sanitation services.

Example

Option 1

Acknowledgements This work was supported by a research grant from Fairfield University's Interdisciplinary Health Studies Scholars Program. The authors are grateful to Shauna Dresel and Alex Ferrante for their research assistance. Many thanks also to Profs. Romer Altamirano Guerrero, Alberto Javier Solorzano Saravia, and Roberto Carlos Ibarra at Universidad Centroamericana (Nicaragua), as well as the UCA students who participated as interviewers, for their valuable insights and hard work in the field. Appendix A. Example of Choice Task Think about how the majority of homes in Nueva Vida do not have a way to discard their waste water. Improper handling of residential water and rain water allows for puddling in the yards and streets. In addition, the improper disposal of feces may jeopardize the environment and health of the people living in Nueva Vida. There are ways to solve these problems. Certain solutions would eliminate only the residential water (dirty water from bathing and washing clothes and dishes), only the feces, only the puddling caused by rain water or a combination of these issues. Next, I will ask you to choose between options that could solve one, two or even all three of the problems mentioned before. A final option is that the community stays the way it is-no change. To finance some of these options, your household would have to pay a monthly installment that would be included in your water bill. Please keep in mind your family income when you choose between the three options. The cost for the option you choose would lead to that money not being available for other household expenses like food, clothes, etc. I will start with an example of the options so you can choose one of them, and then I will give you the opportunity to choose between other options.

Option 2

Current Condition

No

No

Disposal of Residential Wastewater

Si Disposal of Feces

Si

Si No

Disposal of Rain Water

Si Monthly Maintanance Cost Your Choice:

C$ 0 ,

No C$ 200 ,

No C$ 0 ,


squez, J. Alicea-Planas / Journal of Environmental Management 218 (2018) 477e485 W.F. Va

Appendix B. Conditional Logit Models.a

Full Sample of Respondents

Subsample of Seemingly Rational Respondents

WASTEWATER EXCRETA RAINWATER

67.59 (0.89)b 84.93 (2.07)b 81.52 (1.40)b

75.77 (2.55)b 77.63 (3.45)b 98.54 (1.79)b

Choice Tasks

1544

973

a

Standard errors (S.E.), reported in parenthesis, are clustered by respondent to account for potential correlation among choices made by the same individual. b These coefficients are significant at 1% level.

References Allenby, G.M., Rossi, P., 1998. Marketing models of consumer heterogeneity. J. Econ. 89 (1e2), 57e78.
k e
si, D., Gorton, M., Popp, J., Lengyel, P., 2016. Consumer willingness to Balogh, P., Be pay for traditional food products. Food Policy 61, 176e184. https://doi.org/ 10.1016/j.foodpol.2016.03.005. Belli, A., Arostegui, J., Garcia, J., Aguilar, C., Lugo, E., Lopez, D., Valle, S., Lopez, M., Harris, E., Coloma, J., 2015. Introduction and establishment of Aedes albopictus (Diptera: Culicidae) in Managua, Nicaragua. J. Med. Entomol. 52 (4), 713e718. https://doi.org/10.1093/jme/tjv049. Bennet, J., Birol, E., 2010. Choice Experiments in Developing Countries: Implementation, Challenges and Policy Implications. Edward Elgar, Northampton, MA. Birol, E., Das, S., 2010. Estimating the value of improved wastewater treatment: the case of river ganga, India. J. Environ. Manage. 91, 2163e2171. https://doi.org/ 10.1007/s10311-016-0551-9. Birol, E., Karousakis, K., Koundouri, P., 2006. Using economic valuation techniques to inform water resources management: a survey and critical appraisal of available techniques and an application. Sci. Total Environ. 365 (1e3), 105e122. https:// doi.org/10.1016/j.scitotenv.2006.02.032. Buttenheim, A.M., 2008. The sanitation environment in urban slums: implications for child health. Popul. Environ. 30 (1), 26e47. https://doi.org/10.1007/s11111008-0074-9. Cameron, T.A., 1988. A new paradigm for valuing non-market goods using referendum data: maximum likelihood estimation by censored logistic regression. J. Environ. Econ. Manage. 15 (3), 355e379. Campbell, D., 2007. Willingness to pay for rural landscape improvements: combining mixed logit and random-effects models. J. Agric. Econ. 58 (3), 467e483. Elrod, T., Keane, P., 1995. A factor-analytic probit model for representing the market structure in panel data. J. Mark. Res. 32 (1), 1e16. Fiebig, D.G., Keane, M.P., Louviere, J., Wasi, N., 2010. The generalized multinomial logit model: accounting for scale and coefficient heterogeneity. Mark. Sci. 29 (3), 393e421. Genius, M., Menegaki, A.N., Tsagarakis, K.P., 2012. Assessing preferences for wastewater treatment in a rural area using choice experiments. Water Resour. Res. 48, W04501 https://doi.org/10.1029/2011WR010727. Graham, J.P., Polizzotto, M.L., 2013. Pit latrines and their impacts on groundwater quality: a systematic review. Environ. Health Perspect. 121 (5), 521e530. https://doi.org/10.1289/ehp.1206028. Gross, E., Günther, I., 2014. Why do households invest in sanitation in rural Benin: health, wealth, or prestige? Water Resour. Res. 50, 8314e8329. https://doi.org/ 10.1002/2014WR015899. Gu, Y., Hole, A.R., Knox, S., 2013. Fitting the generalized multinomial logit model in Stata. Stata J. 13 (3), 382e397. Hanley, N., Mourato, S., Wright, R.E., 2001. Choice modelling approaches: a superior alternative for environmental valuation? J. Econ. Surv. 15 (3), 435e462. Hess, S., Rose, J.M., 2012. Can scale and coefficient heterogeneity Be separated in random coefficient models? Transportation 39, 1225e1239. Hoyos, D., 2010. The state of the art of environmental valuation with discrete choice experiments. Ecol. Econ. 69, 1595e1603. https://doi.org/10.1016/

485

j.ecolecon.2010.04.011. Koo, T.T.R., Wu, C., Dwyer, L., 2010. Ground travel mode choices of air arrivals at regional destinations: the significance of tourism attributes and destimation contexts. Res. Transp. Econ. 26 (1), 44e53. Lancsar, E., Louviere, J., 2006. Deleting ‘irrational’ responses from discrete choice experiments: a case of investigating or imposing preferences? Health Econo. 15, 797e811. https://doi.org/10.1002/hec.1104. Louviere, J.J., Meyer, R.J., Bunch, D.S., Carson, R., Dellaert, B., Hanemann, W.M., Hensher, D., Irwin, J., 1999. Combining sources of preference data for modelling complex decision processes. Mark. Lett. 10 (3), 205e217. Louviere, J.J., Street, D., Burgess, L., Wasi, N., Islam, T., Marley, A.A.J., 2008. Modeling the choices of individuals decision makers by combining efficient choice experiment designs with extra preference information. J. Choice Model. 1 (1), 128e163. Mangham, L.J., Hanson, K., McPake, B., 2009. How to do (or not to Do)…Designing a discrete choice experiment for application in a low-income country. Health Policy Plan. 24, 151e158. https://doi.org/10.1093/heapol/czn04. McFadden, D., Train, K., 2000. Mixed MNL models for discrete response. J. Appl. Econ. 15 (5), 447e470. https://doi.org/10.1002/1099-1255(200009/10). McGranahan, G., 2015. Realizing the right to sanitation in deprived urban communities: meeting the challenges of collective action, coproduction, affordability, and housing tenure. World Dev. 68, 242e253 (accessed 17.03.02). http:// www.sciencedirect.com/science/article/pii/S0305750X14004069. Ndunda, E.N., Mungatana, E.D., 2013. Evaluating the welfare effects of improved wastewater treatment using a discrete choice experiment. J. Environ. Manage. 123, 49e57. ~ ez Soto, O., 2010. Inodoro Ecolo
gico Popular. Centro para la Promocio
n, la Nun
n y el Desarrollo Rural y Social (CIPRES), Managua, Nicaragua. Investigacio €mert, N., 2011. Music for Free? How free ad-Funded Papies, D., Eggers, F., Wlo downloads affect consumer choice. J. Acad. Mark. Sci. 39 (5), 777e794. Petersen, E., Wilson, M.E., Touch, S., McCloskey, B., Mwaba, P., Bates, M., Dar, O., Mattes, F., Kidd, M., Ippolito, G., Azhar, E., Zumla, A., 2016. Rapid spread of Zika Virus in the americas e implications for public health preparedness for mass gatherings at the 2016 Brazil Olympic games. Int. J. Infect. Dis. 44, 11e15 (accessed 17.03.02). http://www.sciencedirect.com/science/article/pii/ S1201971216000217. Scarpa, R., Thiene, M., Train, K., 2008. Utility in willingness to pay space: a tool to address confounding random scale effects in destination choice to the alps. Am. J. Agric. Econ. 90 (4), 994e1010. https://doi.org/10.1111/j.14678276.2008.01155.x. Sharma, V.P., 2014. Water, mosquitoes and malaria. In: Singh, P.P., Sharma, V. (Eds.), Water and Health. Springer, India. Silano, M., Ortúzar, J., 2005. Willingness-to-Pay estimation with mixed logit models: some new evidence. Environ. Plann. A 37, 525e550. ~ o, M., Farizo, B.A., V
Solin azquez, M.X., Prada, A., 2012. Generating electricity with forest biomass: consistency and payment timeframe effects in choice experiments. Energy Policy 41, 798e806. Train, K., Weeks, M., 2005. In: Alberini, Anna, Scarpa, Riccardo (Eds.), Discrete Choice Models in Preference Space and Willingness-to-pay Space. Chapter in Applications of Simulation Methods in Environmental and Resource Economics. Springer Publisher, Dordrecht, The Netherlands.
stico Urbano: Barrio Nueva Universidad Centroamericana, 2016. Síntesis de Diagno Vida en Ciudad Sandino. Universidad Centroamericana, Managua, Nicaragua. United Nations, 2015. The Millennium Development Goals Report e 2015. United Nations. New York. http://www.un.org/millenniumgoals/2015_MDG_Report/ pdf/MDG%202015%20rev%20%28July%201%29.pdf (accessed 17.03.02). Van Minh, H., Nguyen-Viet, H., 2011. Economic aspects of sanitation in developing countries. Environ. Health Insights 5, 63e70.
squez, W.F., Aksan, A., 2015. Water, sanitation, and diarrhea incidence among Va children: evidence from Guatemala. Water Policy 17, 932e945. https://doi.org/ 10.2166/wp.2015.211. Winters, M.S., Karim, A.G., Martawardaya, B., 2014. Public service provision under conditions of insufficient citizen demand: insights from the urban sanitation sector in Indonesia. World Dev. 60, 31e42. Woldemariam, G., Seyoum, A., Ketema, M., 2016. Residents' willingness to pay for improved liquid waste treatment in urban Ethiopia: results of choice experiment in addis ababa. J. Environ. Plan. Manage. 59 (1), 163e181. https://doi.org/ 10.1080/09640568.2014.996284. Zwane, A.P., Kremer, M., 2007. What works in fighting diarrheal diseases in developing Countries? A critical review. World Bank. Res. Obs. 22 (1), 1e24. https://doi.org/10.1093/wbro/lkm002.