The Value of Achieving Water Quality Improvements in the Rivers of the Metropolitan Region of Berlin and Brandenburg

WORKING PAPER ON MANAGEMENT IN ENVIRONMENTAL PLANNING

The value of achieving water quality improvements in the rivers of the metropolitan region of Berlin and Brandenburg

Jürgen Meyerhoff, Marco Boeri, Volkmar Hartje

032/2013

Working Paper on Management in Environmental Planning 32/2013 Arbeitspapiere zum Management in der Umweltplanung

Authors

Jürgen Meyerhoff * Technische Universität Berlin Institute for Landscape and Environmental Planning [email protected]

Marco Boeri Gibson Institute for Land, Food and Environment, School of Biological Sciences, Queen’s University, Belfast (UK), and UKCRC Centre of Excellence for Public Health (NI) Queen's University of Belfast [email protected]

Volkmar Hartje Technische Universität Berlin Institute for Landscape and Environmental Planning [email protected]

* corresponding author

ABSTRACT

The study reports willingness to pay estimates for improving the water quality of five river stretches in the metropolitan region of Berlin and Brandenburg. The region struggles, as many regions in Germany do, to achieve the objectives of the Water Framework Directive until 2015. One of the major problems concerning the quality of the surface waters is the high load of nutrients. Thus, currently it is investigated which combination of abatement measures, i.e. changing agriculture and extending current or building additional sewage treatment plants, would change the present water quality at the lowest costs. The spatially explicit choice experiment aims at eliciting to what extent people in the region of metropolitan Berlin and Brandenburg would benefit from improving water quality in various rivers of the Havel basin. Spatial heterogeneity is introduced by accounting respondents place of residence, i.e. whether they reside in the urban or the rural part of the metropolitan area. People in the region would significantly benefit from improved water quality but the benefits are not spread evenly across the region. Urban dweller would benefits to a much larger extent rais- ing the questions of who has the bear the cost of changing the quality in the federal system of Germany.

Keywords: Stated choice experiment, EU Water Framework Directive, policy acceptance costs, random parameter logit, spatial heterogeneity

- 1 -

1. Introduction

The European Water Framework Directive (WFD) that entered into force in the year 2000 has recently moved into a new phase of implementation. Generally, the WFD aims at achieving a good ecological status of all water bodies within the EU. Due to the EU timetable, administra- tive resources went in the first phases of the WFP implementation mainly to establishing inven- tories, monitoring networks, and developing first river basin management plans. Currently, i.e. in the phase between 2012 and 2015, administrations are asked to operationalize programmes of measures ensuring that the environmental objectives can be met. It has, however, become obvious that it is very unlikely to reach the above target for all water bodies by 2015. Thus, management plans and even objectives might have to be adjusted accordingly.

Adjustment will very likely involve balancing costs and benefits of the management actions. It has become clear within the last years that meeting the targets of the WFD will be very costly in some cases. This raises the question whether societies are willing to spend the necessary amounts of money to achieve in all water bodies a good ecological status. Although the economic analysis was from the beginning on at the core of the WFD implementation process, estimating the benefits associated with a good ecological status was not originally on the agenda. Researchers suggested to employ economic valuation when the need to define disproportional high measurement costs became evident (Brouwer [1]; Hanley et al. [2]). Meanwhile, several studies from across Europe determining the benefits of a good ecological status have been presented (e.g., Bliem and Getzner [3]; Brouwer et al. [4]; Glenk et al. [5]; Metcalfe et al. [6]). Overall, the studies indicate that people value a good ecological status positively and are willing to pay for changing water quality. A finding presented by some of those studies is that benefits from achieving a good quality are not distributed spatially evenly. Taking into account the spatial distribution of the benefits can thus provide crucial information for decision makers developing management plans. Spatial characteristics have, if at all, been incorporated in the analysis of stated preference data by accounting for the distance between respondents’ place of residence and the water basin in question. However, recently studies used the location of respondents’ place of residence to account for spatial variation and to better incorporate spatial complexity (Bateman [7]). For example, Brouwer et al. [4] analysed whether the place of residence in a river basin influences choices among water quality improvement alternatives. Due to spatial heterogeneity, the authors expect respondents to value changes in environmental good provision differently de- pending on where the change takes place. They found that respondents have preferences for acceptable levels of water quality in the entire basin, but are only willing to pay for good water quality in their own sub basin. The authors argue that aggregating WTP estimates from sub basins to the whole river basin without taking spatial dependence into account would underes- timate the welfare effects of improved water quality. Accounting as well for spatial characteristics, this time the quality of waterways in the vicinity of respondents’ residence, Tait et al. [8] found that the local water quality significantly affects WTP values for a river and stream preser- vation programme in Canterbury. They also found that not accounting for the spatial sensitivity of individual welfare would bias aggregate compensating surplus values.

Working Paper on Management in Environmental Planning 32/2013 - 2 -

Schaafsma et al. [9] not only explored the distance decay effect but also whether the direction in which the respondents’ residences are located relative to the environmental asset, lakes in their case study, influences stated WTP values. They argue that simply including a distance parameter as an additional socio-demographic variable or distance as an attribute in a CE may only partially address the spatial heterogeneity in WTP values. Direction was accounted for by including dummy variables for different directions in site-specific utility functions. The authors find that accounting for the direction’s effects in distance decay leads to improved model fit and significant differences in WTP values. Moreover, they find that the spatial pattern in WTP values reflects the availability in substitute sites quite well. Jørgensen et al. [10] account in their analysis for the travel time to both the river valued and to potential substitute sites. Their contingent valuation study is concerned with the demand for restoring the Odense River in Den- mark. They found the effect of spatially uneven distributed benefits from water quality improvements varies among users and non-users of the river Odense as the spatial availability of substitutes reduces demand among non-users for improvements in the Odense River. The present study adds to the literature on both benefits from changing water quality and the spatial distribution of benefits. The applied choice experiment is concerned with the benefits from achieving the good ecological status in a river system in the metropolitan area of Berlin and Brandenburg. This area is located within the Havel river basin in the East of Germany, itself a sub basin of the Elbe river basin. The rivers Spree, Dahme, and Havel, all are slow- flowing lowland rivers and partly comprise chains of linked lakes, make up the river system within the Havel basin. This basin is one of the five WFD-coordinating areas within the Elbe basin. Major obstacles preventing meeting the WFD objectives for these three rivers are struc- tural and morphological changes in the rivers and high nutrient loads from diffuse sources (agriculture, storm water). The latter leads, among others, again and again to algal blooms during the summer in some of the river stretches. Spatial heterogeneity was explicitly introduced by accounting for respondents place of residence, i.e., whether the respondent lives in the urban area, here the city of Berlin, or in the rural area, here counties around Berlin the federal state of Brandenburg. Moreover, the choice experiment was deigned in an spatially explicit manner as different river stretches are used as attributes. Respondents could therefore chose for which river stretch they prefer what kind of quality improvement. The paper proceeds as follows. In the next section we present the case study area and the survey design. Section 3 briefly introduces the choice experiment method and the econometric analysis used. Subsequently, the estimation results as well as the compensating variation for various quality improvement scenarios are presented before section 5 concludes.

2. Case study area and survey design

The three rivers, the Spree, the Dahme, and the Havel, characterise the river system in the metropolitan region of Berlin and Brandenburg. The river Spree, a left bank tributary of the river Havel, is approximately 400 kilometres long, has its source in the Lusatian Highlands (Lausitzer Bergland) and flows towards the city of Berlin where it merges in Spandau with the river Havel. The Dahme, a tributary of the Spree, is around 95 kilometres in length and expanding in some sections to a chain of lakes. The River Havel itself is a right tributary of the river Elbe with a

Working Paper on Management in Environmental Planning 32/2013 - 3 - length of 325 kilometres providing a link in the waterway connections between east and West Germany. Figure 1 shows the locations of the three rivers and indicates the present water quality using the classification system of the WFD.

In the present study the river system was divided into five stretches: the Lower Havel (a), the Upper Havel (b), the city stretch of the Spree (c), the Spree at Koepenick (d), and the Dahme section up to the lake Scharmützelsee (e). The Water quality of the river stretches is described using a water quality ladder. This ladder , as shown in Figure 2, provides information concerning both the meaning of the different water quality levels for recreational activities as well as for animal and plant species in the region and was developed using the five level classification system of the WFD (BMU [11]). The two lowest categories were merged, as both do not differ significantly with respect to their influence on recreational opportunities or ecological aspects such as species richness. Figure 1 shows the river sections and indicates the present water quality. The diamond each time marks the border between river stretches. The present water quality conditions differ from poor/moderate in the river stretches Lower Havel (a), Upper Havel (b), and the city stretch of the Spree (c) to moderate/good in the Spree-Köpenick stretch (d) and poor/good in the Dahme-Scharmützelsee stretch (e) (also Table 1).

Note: The diamonds mark the end / beginning of a river stretch.

Figure 1: Present water quality of river stretches in the region Berlin and Brandenburg

Starting from the present water quality different improvements are achievable for each river sections. Table 1 presents the present water quality according to the EU-WFD classification system for each stretch and reports the potential improvements according to the levels of the water quality ladder. Based on these potential improvements an efficient experimental design was generated. In this design the river sections are the attributes and the potential quality im-

Working Paper on Management in Environmental Planning 32/2013 - 4 - provements are the levels. Using a Bayesian D-efficient design (Ferrini and Scarpa [13]), 24 choice sets were created and assigned to two blocks of each time 12 choice sets. The priors used in the design process were derived from choice sets presented at focus groups and a pilot study. An example of a choice set is presented in Figure 2.

Table 1: Attributes and levels

Attribute Level

Present Quality Quality improvements Stretch 1 (a) Very Lower Havel Poor/Unsatisfactory Moderate Good Good Stretch 2 (b) Very Upper Havel Poor/Unsatisfactory Moderate Good Good Stretch 3 (c) City stretch Unsatisfactory/moderate Moderate Good ./. Spree Stretch 4 (d) Very Spree Köpenick Moderate/Good ./. Good Good Stretch 5 (e) Dahme- Moderate / Very Unsatisfactory/Good Good Scharmützelsee Good Good Cost in € 10 / 25 / 50 / 75 / 100 / 150

The survey data were collected in the core metropolitan region of Berlin and Brandenburg in 2011. This area has a radius of about 80 kilometres around the city centre of Berlin and covers nearly 90 % of the population of the two federal states Berlin and Brandenburg. For the pur- pose of this survey counties at the edge of Brandenburg were not included due to their remote- ness to the five river stretches. The interviews, conducted by a survey company, proceeded in two-stages. In the first stage a random sample of respondents living in the study region was contacted by phone. If individuals agreed to participate they were interviewed about, for example, their use of water bodies for recreational purposes in the study region and their perception of the water quality. Subsequently attitudinal statements were presented and socio- demographics requested. At the end of the phone interview people were asked whether they are willing to participate in a web-based survey concerned with water quality improvements in the region of Berlin and Brandenburg. Those who agreed were emailed a personalised link to the survey.

Working Paper on Management in Environmental Planning 32/2013 - 5 -

Figure 2 Example choice set

The web-based interview proceeded as follows. At the beginning respondents were introduced to the water quality ladder and, using the levels of this ladder, informed about the present water quality of five river stretches (Figure1). Next, they were informed that it is possible to improve the water quality by, for example, extending sewage treatment plants and by changing agricultural practises. However, as it would not be possible to finance the measure completely by public budgets, additional contribution by both private households and businesses would be essen- tial. Respondents were informed that both industry and private households are responsible for the present water quality. Subsequently, the payment vehicle was introduced. The focus groups clearly indicated that a surcharge to the water bill as it was used in other studies [3,4] would very likely have increased the number of protest responses. The reason for this is that a very controversial debate within Berlin takes place about whether the main water company, the Ber- liner Wasserbetriebe, should be run as a private or public company. Currently, it is a private company but many people are in favour of running it again as a public company. Many partici- pants of the focus groups conducted in three different places with the study region indicated that they are willing to pay for improving the current water quality but opposed to pay via a surcharge to their water bill. They were concerned that this would raise the profits of the company but not results in a higher water quality of the river and lakes in their region. Thus, as a payment vehicle a contribution to a fund responsible for implementing the measures needed to improve the water quality was introduced. Respondents were told that they would have to pay for 10 years. Subsequently, the choice sets were introduced, and respondents faced the twelve choice sets that were presented to each in a randomized order. The questionnaire concluded with attitudinal questions concerning the choice sets, among others.

Table 2 shows the statistics for socio-demographics and the recreational use of rivers and lakes within the region Berlin-Brandenburg. The upper part reports the values for all people who participated in the phone interview while the lower part presents the figures for those who participated in the web-based CE study. Moreover, statistics are presented for urban and rural dwellers separately. Overall, 2301 phone interviews and 752 web-based interviews were com-

Working Paper on Management in Environmental Planning 32/2013 - 6 - pleted. Noteworthy is that the net household income is higher in the rural area (Brandenburg) than in the urban area (Berlin). Concerning the recreational use of the water bodies in the region, respondents stated that they visited on average 58 times water bodies during the twelve months prior to the interview. The major activity at the water bodies is walking while 35% went swimming and 7% angling. Accounting for whether a respondents is from the urban or rural area results in significantly different frequencies of recreational activities. People in the rural area made overall 20 visits more to water bodies than the urban dwellers during the twelve months prior to the interview.

Table 2: Socio-demographics whole sample vs. web-survey All Urban Rural mean Sd mean sd mean sd Phone survey N = 2181 N = 1085 N = 1096 Age (in years) 46.97 16.10 45.27 16.47 48.30 15.60 Gender (1=female) 0.49 0.50 0.50 0.50 0.50 0.50 Person per household 2.19 1.12 2.07 1.11 2.31 1.11 Net household income1 2625.29 1597.88 2491.05 1588.58 2756.95 1596.96 (Euro per month) Number of recreational 57.61 80.46 46.39 67.82 68.71 89.93 stays at water bodies Swimming (%) 0.34 0.47 0.33 0.47 0.35 0.47 Angling (%) 0.09 0.29 0.05 0.22 0.13 0.34 Web-based survey N = 752 N = 409 N = 343 Age (in years) 44.71 14.92 43.66 15.18 46.44 14.70 Gender (1 = female) 0.50 0.50 0.48 0.50 0.50 0.50 Person per household 2.26 1.18 2.08 1.12 2.44 1.23 Net household income2 2829.28 1428.87 2555.95 1374.59 3005.83 1454.33 (Euro per month) Number of recreational 55.74 77.03 46.62 69.58 65.69 82.73 stays at water bodies Swimming (%) 0.35 0.48 0.31 0.46 0.40 0.49 Angling (%) 0.07 0.26 0.04 0.19 0.11 0.31 Note: Due to missing responses 1) n = 1681 observations; 2) n = 616

Comparing both the phone and the web-based survey the most obvious difference is with respect to net household income which is on average around 200 € higher in the web-based sample. Also average age is about two years lower while other differences are rather small. Overall, we could not find statistically significant differences between both samples. We therefore conclude that no strong selection bias is present in the data.

3. Discrete choice experiments

Discrete Choice Experiments (CE) is a survey based technique to elicit preferences of citizens on non-market goods or services characterized by more attributes. Our study aims at identify- ing citizens’ preferences on different levels of water quality in different stretches of the river system in metropolitan area of Berlin and Brandenburg.

Working Paper on Management in Environmental Planning 32/2013 - 7 -

3.1 The econometric analysis In our empirical case we consider the situation in which a subject n has to choose between J alternatives of water improvements in the rivers of the metropolitan region of Berlin and Brandenburg for a sequence of T choice tasks. By assuming that respondents choose by maximizing their utility, we apply a standard random utility model (RUM – see [14,15]) in the analysis. Under this setting, the core assumption of CE is that choices are driven by the maximisation of respondents’ utility. The utility that each alternative brings to the respondents can be represented by the function:

UXnit ' nit nit (1) where n indicates the respondent, i the chosen alternative, t the choice occasion, X is a vector of attributes,  is a vector of parameters to be estimated and  is a random error term (unobserved by the researcher, often referred to as disturbance) assumed to be iid Gumbel. From the utility function in eq. 1 it is possible to represent the probability for individual n of choosing alternative i over any other alternative j in choice set by a multinomial logit (MNL) model [16] is:

(2)

where Vin= ’ Xni. Finally, as we are interested in differences of taste between urban and rural residents, we will estimate two sets of : one for people living in the urban area of Berlin and one for respondents living in rural areas. The MNL assumes independence of irrelevant alternatives (IIA) property which is the same as assuming that everybody in the sample have the same preferences for water quality changes. While in some cases this assumption may hold, a number of empirical studies have shown that there is often heterogeneity in the preferences that individuals hold for different attributes. The limitations of the MNL model in accommodating preference heterogeneity have given rise to a suite of models that fit under the umbrella of mixed logit (MXL) models (McFadden and Train [17]). MXL models can provide a flexible, theoretical and computationally practical econometric method for any discrete choice [17]. The central feature of MXL models is their ability to accommodate random taste variation, unrestricted substitution patterns and correlation in unobserved factors over time (Train [18]). Furthermore, MXL models allow researchers to incorporate in their analysis the correlation between different alternatives by means of error components. MXL models are generally shown to significantly improve model fit [19, 20], as well as provide greater insights into choice behaviour [17] and welfare estimation [21, 22, 23]. By applying the MXL model to both sets of betas (for rural and urban) we account for unobserved preference heterogeneity within each group.

In MXL models, the parameters are allowed to vary across respondents. If the values of the vector of estimated parameters were known with certainty for each respondent, then the probability of respondent n’s sequence of choices would be respectively given by:

Working Paper on Management in Environmental Planning 32/2013 - 8 -

(3)

where is the sequence of choices over the T choice occasions for respondent n. As it is clearly not possible to know the value of the parameters with certainty for each respondent, random variation is allowed to facilitate the heterogeneity across respondents in estimation. Under this condition, the unconditional choice probability is obtained by integrating the product of logit probabilities over the distribution of n:

(4)

We assume normal distributions for the non-monetary attributes but maintain a fixed cost attribute to alleviate problems with taking the ratio of two random parameters [24]. However, the coefficient varies across the two groups of respondents. The analyses were performed with Biogeme 2.2 [25,26], a new and more flexible version of Biogeme based on python. Models were estimated using the CFSQP algorithm [27] considering the repeated choice nature of the data. Since the choice probabilities in equations 4 has no closed form, it is estimated by maxi- mum simulated likelihood (MSL) with 1000 quasi-random draws via Latin-hypercube sampling [28].

3.2 Individual Posterior parameters and welfare analysis

Given that one of the main objectives of environmental studies is the assessment of users’ welfare, we compute the willingness to pay (WTP) for each attribute for each individual in the sample conditional to the pattern of choices observed. The conditional marginal WTPs can be computed using the estimator proposed by Greene et al [29]:

r 1 ˆ att, n L ˆ r y ,x r ˆ r nnn R  price, n EWTPˆ  (5) att, n 1 L ˆ r yx, R r nnn

r where L(.) is the posterior likelihood of the individual respondents and the β n are drawn from the multivariate normal computed at the MSL estimates βΩˆˆ, . Furthermore, we calculate the compensating variation (CV, also referred to as Consumer surplus) as described by Hanemann [30], in associated with 4 specific policy changes of inter- est in the study. To compute the CV from the MXL model’s estimates, it is necessary to obtain the individual-specific posterior estimates (Equation 5) and then compute the difference in log- sum for each individual between the baseline scenario (current situation) and the policy change scenario [31, 32, 18].

Working Paper on Management in Environmental Planning 32/2013 - 9 -

(6)

where CVn is the individual n’s compensating variation for a change from initial conditions V0n

(current situation) to the conditions under the program V1n (policy change scenario) and βprice is the cost parameter which represents the marginal utility of money. Once the CV is computed for each scenario at a given price, it is straightforward to retrieve the “Policy Acceptance Price” by testing different prices and selecting the one that leaves 50% of the sample with a positive CV.

4. Results

4.1 Estimation results

Table 3 reports the parameter estimates from two MNL models and the RPL model. Starting with the basic MNL model, results indicate that respondents are in favour of higher water quality levels in all river section. All water quality related attributes are statistically significant with a positive sign except for the change to the yellow quality level for stretch (a) (Lower Havel) and stretch (c) (city stretch Spree). The cost coefficient has the expected negative sign, implying that an option is less likely to be chosen when costs increase. The status quo constant (ASCsq) is positive suggesting that respondents are reluctant to move away from the status quo, i.e, to make a trade-off between money and an improved water quality. The performance of the basic MNL model is, however, rather poor. Thus, we move to the models that allow for spatial heterogeneity estimating the taste weights for both urban and rural dwellers separately.

Compared to the basic MNL model the performance of both models that allow for differences between urban and rural dwellers significantly improves as expected. The RPL considering the panel characteristic of the data performs best. The majority of parameter estimates is again statistically significant with a positive sign. Both models indicate differences between urban and rural dwellers. Generally, the size of the coefficients is larger in the group of urban dwellers compared to rural dwellers apart from the parameter for stretch (b) (Upper Havel). For this river stretch, rural people reveal stronger preferences. The size of the coefficient of the cost attribute is larger in the urban group showing higher sensitivity toward costs among the rural people. Moreover, the status quo specific constant (ASCsq) is positive in both groups, but of different size. While on average respondents in both groups are reluctant to move away from the status quo this is less pronounced among urban dwellers. Another difference is that rural dwellers do not reveal a preference for improving the water quality of the city stretch of the Spree, the river stretch in the city of Berlin. As Figure 1 shows improvements in this river section are likely to mainly benefit people living in the city when we assume that people from Brandenburg, the rural dwellers, probably do not go to Berlin for water based recreational activities.

In the RPL model, again the size of the mean parameter of the statistically significant attributes is larger within the group of people living in the city than among those living in rural areas confirming that urban dwellers have in the present case stronger preferences for water quality

Working Paper on Management in Environmental Planning 32/2013 - 10 - improvements. The only exception is again the river stretch (b) (Upper Havel). Generally, the mean parameter estimates in the RPL model are larger in magnitude compared to the MNL model suggesting that the former model has higher scale, i.e. lower error variance. Moreover, all standard deviations are statistically significant suggesting that beyond the heterogeneity captured by the characteristic of being an urban or rural inhabitant water quality improvements are not valued uniformly. Accounting for unobserved heterogeneity, thus, leads to a significant improvement in model fit compared to the two group MNL model, showing the importance of taking into account the panel nature of the CE and unobserved heterogeneity. The sizes of the status quo parameters align again in the RPL model mirroring the substantial share of respondents who do not prefer to leave the current situation.

4.2 Welfare analysis

Our welfare analysis is performed on our best model which is the RPL model including heterogeneity between urban and rural citizens. Table 4 reports those WTP estimates. The welfare analysis for the RPL model is based on retrieved posterior individual parameters conditional to the choices that each individual faced in the survey (see Section 3.2). Positive WTP values suggest desirable attribute levels, whilst negative WTP values suggest undesirable attribute levels, taking into account that there are trade-offs between quality improvements and costs. Generally, lower quality improvements result in lower WTP estimates, in some cases they even result in negative values. Respondents do not seem to see changes to a moderate quality level, for example, as a significant improvement. An exception is stretch (e) (Dahme-Scharmützelsee) in the southeast of Berlin. Changing the part of that river stretch that is currently of a poor quality already leads to significant WTP values. On average respondents are willing to pay around €43 (urban dwellers) and around €30 (rural dwellers) for this change.

Working Paper on Management in Environmental Planning 32/2013 - 11 -

Table 3: results from MNL and RPL models – comparing rural and urban

MNL MNL RPL Urban Rural Urban Rural Par. s.e. Par. s.e. Par. s.e. Par s.e. SD s.e. Par s.e. SD s.e. Stretch 1: moderate 0.005 0.068 0.013 0.091 0.001 0.102 ‐0.023 0.155 0.778* 0.174 ‐0.304 0.209 1.44* 0.205 Stretch 1: good 0.650* 0.056 0.711* 0.076 0.586* 0.085 1.51* 0.163 1.42* 0.224 1.26* 0.181 1.15* 0.261 Stretch 1: very good 0.781* 0.051 0.817* 0.068 0.750* 0.077 1.26* 0.169 2.07* 0.164 1* 0.197 2.22* 0.216

Stretch 2: moderate 0.157* 0.060 0.125 0.080 0.219* 0.091 0.064 0.154 0.625* 0.306 0.262 0.194 1.32* 0.255 Stretch 2: good 0.747* 0.056 0.713* 0.076 0.821* 0.085 1.00* 0.161 2.08* 0.166 1.06* 0.21 2.57* 0.217 Stretch 2: very good 0.879* 0.050 0.802* 0.066 1.000* 0.078 1.25* 0.169 2.64* 0.191 1.78* 0.245 3.18* 0.234

Stretch 3: moderate ‐0.051 0.051 ‐0.042 0.068 ‐0.065 0.077 ‐0.246 0.135 ‐1.29* 0.178 ‐0.458* 0.16 ‐0.95* 0.168 Stretch 3: good 0.132* 0.047 0.248* 0.062 ‐0.012 0.072 0.439* 0.133 1.56* 0.144 ‐0.099* 0.157 1.45* 0.159

Stretch 4: good 0.441* 0.066 0.531* 0.091 0.320* 0.097 1.05* 0.178 ‐1.53* 0.228 0.719* 0.213 1.51* 0.276 Stretch 4: very good 0.566* 0.037 0.625* 0.049 0.495* 0.056 1.12* 0.139 ‐1.86* 0.131 0.648* 0.183 ‐2.55* 0.202

Stretch 5: moderate 0.334* 0.064 0.394* 0.089 0.257* 0.095 0.764* 0.164 1.15* 0.157 0.509* 0.19 1.05* 0.235 Stretch 5: good 0.541* 0.068 0.629* 0.095 0.428* 0.099 1.32* 0.189 1.15* 0.163 0.919* 0.23 1.81* 0.204 Stretch 5: very good 0.669* 0.046 0.722* 0.061 0.597* 0.071 1.32* 0.189 ‐1.78* 0.162 0.691* 0.198 ‐2.3* 0.28

ASCsq 1.210* 0.091 0.992* 0.123 1.44* 0.137 1.14* 0.221 1.48* 0.273 Cost ‐0.008* 0.001 ‐0.009* 0.001 ‐0.006* 0.001 ‐0.021* 0.002 ‐0.020* 0.002 LogL (Model) -9055.677 -8975.965 -6739.227 K 15 30 56 * Significant at least at 5% level.

Working Paper on Management in Environmental Planning 32/2013 - 12 -

The change to a good water quality, as it is the target of the EU-WFD, is valued positively throughout the study area. However, we observe that rural dwellers are not particularly in favor of paying for quality improvements in the river stretch (c) (Stadtspree), the part of the river Spree that lies entirely within the city area of Berlin. The negative sign indicates that rural inhabitants on average do not prefer improvements in this stretch. Taking into account the low likelihood that people from the rural areas will go to Berlin for water-based recreation this finding supports the validity of the results. Improvements to very good quality are rated higher as changes to a good quality only in the case of stretch (b) (Upper Havel) and stretch (e) (Dahme-Scharmützelsee). In the other cases respondents do not place an additional value to achieving a very good quality. Regarding the spatial heterogeneity, we find that except for stretch (b) (Upper Havel) urban dwellers have a higher WTP for improving water quality. This is despite the fact that on average rural dwellers have higher household incomes and do use water bodies more often than urban dwellers. A reason for this could be that rural dwellers have more substitute sites such as lakes in their surroundings available and thus recreational activities do not depend as much on the river stretches as urban dwellers do.

Table 4: Mean and median conditional posterior WTP for quality improvements

Attribute RPL Urban Rural Mean Median Mean Median Stretch a: poor/moderate -> moderate 0.02 -1.91 -15.20 -19.40 Stretch a: poor/moderate -> good 74.69 79.53 66.19 68.35 Stretch a: poor/moderate -> very good 62.88 69.39 52.49 34.21 Stretch b: poor/moderate -> moderate 4.01 2.62 13.73 8.74 Stretch b: poor/moderate -> good 51.94 42.68 59.63 55.84 Stretch b: poor/moderate -> very good 64.84 53.14 86.20 61.13 Stretch c: poor/moderate -> moderate -12.67 -12.43 -22.39 -22.49 Stretch c: poor/moderate -> good 22.72 19.58 -6.79 -11.86 Stretch d: moderate/good -> good 16.22 5.40 14.52 3.00 Stretch d: moderate/good -> very good 44.65 36.24 24.80 18.09 Stretch e: poor/good -> moderate 42.78 43.86 30.24 29.75 Stretch e: poor/good -> good 61.70 60.54 40.82 27.47 Stretch e: poor/good -> very good 60.04 51.42 36.54 33.99

Next, we compute the compensating variation for four different scenarios. The scenarios are identified in Table 5. The first scenario “always the best” simply has the quality levels always at the highest possible level. All quality levels are at “very good” except for the river section (c) (Stadtspree) for which the quality cannot be improved beyond a good status. The second scenario reflects the objective of the EU-WFD that aims at achieving a good status for all water bodies. The remaining two scenarios improve each time two of the river stretches and leave all remaining stretches unchanged. The difference is that in Scenario 3 both river stretches with improved water quality are located in the West of Berlin while the two stretch-

Working Paper on Management in Environmental Planning 32/2013 - 13 - es in Scenario 4 are located in the East of Berlin. Note also that the current water quality is partly higher in the two stretches located in the East of Berlin (Figure 1).

Table 5: Scenarios for calculating compensating variation measures

Section in Map Scenario 1 Scenario 2 Scenario 3 Scenario 4 “Simply the “EU-WFD” “West of Ber- “East of best” lin” Berlin” Stretch 1 (a) Very good Good Good No change

Stretch 2 (b) Very good Good Good No change

Stretch 3 (c) Good Good No change No change

Stretch 4 (d) Very good Good No change Good

Stretch 5 (e) Very good Good No change Good

The mean and median CV for all four scenarios and for each cost level are reported separately for urban and rural dwellers in Table 6. The last row of this table displays the policy acceptance cost, calculated as the amount of money that leaves half of the sample with a positive consumer surplus, i.e. benefits from water quality improvements are for one half still positive at these costs. Overall, the distribution of the CV estimates is negatively skewed. This is due to the fact a significant share of respondents is very sensitive to costs and thus has a high propensity to chose the zero-price option. Taking into account the positive ASCsq when calculating the CV results thus in a mean value lower than the median value. Moreo- ver, as expected, the CV is decreasing with increasing costs reflecting the declining likelihood that options with higher costs have been chosen. Among the scenarios, “Simply the best”, not surprisingly, results in the highest welfare measures. Even at costs of €150 per year the median CV is positive for both rural and urban dwellers. Changing to the highest quality is strongly supported by respondents’ preferences. However, for rural dwellers the mean CV for this scenario at a cost of €150 is already negative. Implementing the EU-WFD scenario would result in significantly lower CV values (Scenar- io 2). In this case for urban dwellers the median CV would already be negative at costs of €150 per year. The policy acceptance costs for the urban dwellers are at 120 €, well below the highest level of the monetary attribute. The CV measures for the two remaining scenarios show that improvements in the river stretches (a) (Lower Havel) and (b) (Upper Havel) located in the West of Berlin would result in higher benefits to society than improvements in the two stretches in the East of Berlin. That said, the welfare associated with changes becomes negative for both rural and urban dwellers (€75 per year) at costs of €50 per year respectively €75 per year. With regard to the changes described in Scenario 4, on average neither urban nor rural dwellers are willing to trade-off money for a change to a good quality in these two river stretches if costs are €25 per year or higher. The policy acceptance cost among urban dwellers is €18 per year while for rural dwellers not positive figure for the policy acceptance costs exists. As Figure 1 shows these two stretches partly have already a good quality.

Working Paper on Management in Environmental Planning 32/2013 - 14 -

Table 6: Mean (Median) compensating variation and ‘Policy Acceptance Cost’ for different scenarios of water quality improvements

Scenario’s Scenario 1 Scenario 2 Scenario 3 Scenario 4 cost “Simply the “EU-WFD” “West of Berlin” “East of Berlin” best” Urban Rural Urban Rural Urban Rural Urban Rural 25 175.60 92.11 147.70 73.24 47.08 24.69 -1.62 -45.79 (221.70) (140.70) (180.10) (97.95) (49.84) (32.54) (-6.55) (-53.91) 50 156.60 67.11 122.70 48.24 22.08 -0.31 -26.62 -70.79 (196.70) (115.70) (155.10) (72.95) (24.84) (7.54) (-31.55) (-78.91) 75 125.60 42.11 97.72 23.24 -2.92 -25.31 -51.62 -95.79 (171.70) (90.72) (130.10) (47.95) (-0.16) (-17.46) (-56.55) (-103.90) 100 100.60 17.11 72.72 -1.76 -27.92 -50.31 -76.62 -120.80 (146.70) (65.72) (105.10) (22.95) (-25.16) (-42.46) (-81.55) (-128.90) 150 50.58 -32.89 22.72 -51.76 -77.92 -100.30 -126.60 -170.80 (- (96.66) (15.72) (55.05) (-27.05) (-75.16) (-92.46) (-178.90) 131.60) ‘Policy Acceptance 245 165 205 120 75 57 18 - Cost’

Overall, the CV of urban dwellers is in all scenarios higher than the surplus of the rural dwellers. People living in Berlin would accordingly benefit more from the water quality improvements defined in the four scenarios. This is also indicated by the policy acceptance cost reported in the last row of Table 6. The acceptance cost is for all scenarios significantly lower for rural than for urban dwellers. Finally, Table 7 reports aggregated figures for the second scenario, the EU-WFD. This scenario is chosen because it is the most policy relevant regarding the EU-WFD objective. The CV is aggregated for both the urban and the rural area separately. Each time we present two figures, one based on the number of inhabitants in the study region (4.4 million) and the other based on the number of households (2.9 million). The number of inhabitants is calculated based on GIS data provided by [33]. The aggregation based on the number of households provides a conservative measure assuming that each household is only willing to pay once and not per member older than 18 years. The number of households was calculated based on the average number of persons per household in Berlin and in Brandenburg.

Working Paper on Management in Environmental Planning 32/2013 - 15 -

Table 7: Aggregated yearly compensating variation

Scenario’s cost Scenario “EU-WFD” Urban Rural Total Individuals 2 890 391 1 526 980 4 417 371 (age 18 an older) € 50 354 650 975 73 661 515 428 312 490 € 100 210 189 233 -2,687 484 207 501 748 € 150 65 669 683 -79 036 484 -13 366 801

Households 1 958 128 907 970 2 866 098 € 50 240 262 345 43 800 475 284 062 820 € 100 94 460 110 -1 598 027 92 862 083 € 150 44 488 675 -46 996 529 -2 507 854 Note: Total I: negative CV has to be compensated; Total II: negative CV is not compensated as it is assumed that people would not have drawbacks from increased water quality

The results from Table 7 concerning the good quality objective of the EU-WFD show that the urban dwellers would benefit much more from any improving water quality. This has two rea- sons: Firstly, urban dwellers are willing to pay higher amounts of money for improving water quality. This is shown in Table 4. Secondly, the population of people 18 years and older is much larger in Berlin (2.9 million people) than in Brandenburg (1.5 million people). Assuming scenario cost’s of €50 per person would result in aggregated benefits of €428 million per year for the whole study region. This figure decreases to €-13 million if the costs would increase to €150 per year per person. In this case the positive benefits experienced by urban dwellers would not compensate the negative effects experienced by the rural dwellers. Additionally, as a conservative measure also the aggregated values based on the number of households are reported. The aggregated CV decreases to €284 million when we assume that each household would only pay as a unit instead of paying per person. If we assume costs of €150 the overall CV would be again negative.

5. Discussion and conclusions

The present study reports the results from the first survey conducted in Germany aiming at eliciting the benefits people would derive from achieving the EU-WFP objective of a good water quality. The results indicate that in the closer metropolitan area of Berlin and Branden- burg people do value changes in water quality in accordance with the objectives of the EU- WFD positively. This finding is in line with other studies carried recently on the WFD objectives in other countries [3, 4]. Moreover, we also found that heterogeneity among respondents is highly significant. It is found at the spatial level as well as with respect to unobserved taste heterogeneity. Following recent studies accounting for spatial differences [4, 9, 10] we find that the spatial distribution of the benefits associated with changed water quality levels varies. In the pre-

Working Paper on Management in Environmental Planning 32/2013 - 16 - sent analysis, spatial heterogeneity was defined based on respondents place of residence, i.e., being an urban dweller (living in the city of Berlin) or being a rural dweller (living in a county around Berlin in the federal state of Brandenburg). The analysis indicates that substantial differences between urban and rural dwellers are given. The calculated CV measures show that people living in Berlin would benefit more from changes in water quality than people around Berlin. The scenarios concerning quality changes in the west or east of Berlin suggest that changes in the river stretches in the west of Berlin would be more beneficial from societal point of view. However, even after controlling for spatial heterogeneity a significant amount of unobserved taste heterogeneity is present among both urban and rural dwellers suggesting that location is not the only source of difference among respondents. The results are meaningful not only with respect to cost-benefits comparisons but also for systems of fiscal federalism that want to integrate compensations for ecosystem services. From the view of public finance, fiscal transfers are a suitable instrument for internalizing spatial externalities (Ring [34,35]). The river stretches valued are mainly located in the federal state of Brandenburg. As people in the urban area would benefit more from changed water qualities the questions arises who has to bear the costs associated with the management actions required to achieve better water quality levels. As this is a highly political issue, the present paper shows that accounting for spatial units can provide important information for negotiations on fiscal transfers compensating for the provision of ecosystem services. This paper reported on the benefits of improving the water quality. A next step of the analysis is the calculation of the management costs. Within the NITROLIMIT project various scenarios for different quality targets, i.e. target concentrations of surface water quality, are calculated using the MONERIS nutrient emission model (http://moneris.igb-berlin.de). It allows calculating the efficiency of various management measures such as changes agricultural practices or construction of additional wastewater treatment plants for reaching prescribed water quality standards. Knowing to what extent measures are required for reaching a quality target as defined by the EU-WFD allows to determine the costs associated with those measures, and, subsequently to compare benefits and costs. This will provide policy makers with information regarding the economic consequences of pursuing the objectives of the Wa- ter Framework Directive. A rough figure provided by the Federal Ministry of the Environment [11] estimates that the measures required allover Germany would cost 9.4 billion or approximately 20 Euros per captia and year for 2010 to 2015. The policy acceptance costs deter- mined here for the region Berlin-Brandenburg for most scenarios calculated clearly above that figure indicating huge welfare gains from improving the water quality. Taking into account that in other regions in Germany people have on average rather higher incomes, our findings might also suggest that achieving of a good status for all rivers and lakes in Germa- ny will from an economic point of view be highly advantageous.

Acknowledgement

The authors would like to thank Klaus Glenk for valuable comments on an earlier version of this paper. The CE survey was conducted as a part of the research project Nitrolimit (www.nitrolimit.de); funding (Fkz. 033L041D) by the German Federal Ministry of Education and Research is gratefully acknowledged.

Working Paper on Management in Environmental Planning 32/2013 - 17 -

References

[1] R. Brouwer, The potential role of stated preference methods in the Water Framework Di- rective to assess disproportionate costs. J Environ Plann Man 51 (2008) 597-614. [2] N. Hanley, R.E. Wright, B. Alvarez-Farizo, Estimating the Economic Value of Improve- ments in River Ecology Using Choice Experiments: An Application to the Water Framework Directive, Journal of Environmental Management 78 (2006) 183-193. [3] M. Bliem, M. Getzner, Willingness-to-pay for river restoration: differences across time and scenarios, Environmental Economics and Policy Studies 14 (2012) 241-260 [4] R. Brouwer, J. Martin-Ortega, J. Berbel, Spatial Preference Heterogeneity: A Choice Ex- periment, Land Economics 86 (2010) 552-568. [5] K. Glenk, M. Lago, D. Moran, Public preferences for water quality improvements: implica- tions for the implementation of the EC Water Framework Directive in Scotland. Water Poli- cy 13 (2011) 645-662. [6] P. Metcalfe, P. W. Baker, K. Andrews, G. Atkinson, I. J. Bateman, S. Butler, R. T. Carson, J. East, Y. Guéron, R. Sheldon, K. Train, An assessment of the nonmarket benefits of the Water Framework Directive for households in England and Wales, Water Resources Res- search, (2012) in press [7] I. J. Bateman, Bringing the real world into economic analyses of land use value: Incorpo- rating spatial complexity, Land Use Policy 26 (2009) 30-42. [8] P. Tait, R. Baskaran, R. Cullen, K. Bicknell, Nonmarket valuation of water quality: Ad- dressing spatially heterogeneous preferences using GIS and a random parameter logit model, Ecological Economics 75 (2012) 15-21. [9] M. Schaafsma, R. Brouwer, J. Rose, Directional heterogeneity in WTP models for environmental valuation, Ecological Economics 79 (2012) 21-31. [10] S.L. Jørgensen, S. B. Olsen, J. Ladenburg, L. Martinsen, S. R. Svenningsen, B. Hasler, Spatially induced disparities in users' and non-users' WTP for water quality improvements—Testing the effect of multiple substitutes and distance decay, Ecological Econom- ics, in press [11] Federal Ministry for the Environment, Nature Conservation and Nuclear Safety, Water Framework Directive. The path towards healthy waters, Berlin, 2010. [12] S. Hime, I. J. Bateman, P. Posen, M. Hutchins, A transferable water quality ladder for conveying use and ecological information within public surveys, CSERGE Working Paper (2009) EDM 01. [13] S. Ferrini, R. Scarpa, Designs with a priori information for nonmarket valuation with choice experiments: A Monte Carlo study, Journal of Environmental Economics and Man- agement 53 (2007) 342-363. [14] L. Thurstone, A law of comparative judgment, Psychological Review 34 (1927) 273–286. [15] C. F. Manski, Structure of Random Utility Models, Theory and Decision 8 (1977) 229- 254. [16] D. L. McFadden, Conditional logit analysis of qualitative choice behavior, Frontiers in econometrics, New York (1974) Academic Press. [17] D. McFadden, K. Train, Mixed MNL models for discrete response, Journal of Applied Econometrics 15 (2000) 447-470. [18] K. E. Train, Discrete Choice Methods with Simulation. Cambridge 2003. [19] D. A. Hensher, W. H. Greene, The Mixed Logit model: The State of Practice, Transporta- tion 30 (2003) 133-176.

Working Paper on Management in Environmental Planning 32/2013 - 18 -

[20] D. Rigby, K. Balcombe, M. Burton, Mixed logit model performance and distributional as- sumptions: preferences and GM foods, Environmental and Resource Economics 42 (2009) 279–295. [21] M. Sillano, J. D. Ortúzar, Willingness-to-pay estimation with mixed logit models: some new evidence, Environment and Planning A 37 (2005) 525–550. [22] R. Scarpa, M. Thiene, K. Train, Utility in WTP space: a tool to address confounding random scale effects in destination choice to the Alps, American Journal of Agricultural Eco- nomics 90 (2008) 994-1010 [23] A. Daly, S. Hess, K. Train, Assuring finite moments for willingness to pay in random coefficient models, Transportation 39 (2012) 19-31. [24] B. Cooper, J. Rose, L. Crase, Does anybody like water restrictions? Some observations in Australian urban communities, Australian Journal of Agricultural and Resource Econom- ics 56 (2012) 61-81. [25] M. Bierlaire, BIOGEME: a free package for the estimation of discrete choice models, Proceedings of the 3rd Swiss Transport Research Conference, Monte Verita, Ascona, Switzerland, 2003. [26] M. Bierlaire, An introductory tutorial to BIOGEME Version 1.8, Monte. Verita, Ascona, Switzerland, 2009. [27] C. Lawrence, J. Zhou, A. Tits, User’s Guide for CFSQP Version 2.5: A C Code for Solv- ing (Large Scale) Constrained Nonlinear (Minimax) Optimization Problems, Generating It- erates Satisfying All Inequality Constraints, Technical Report TR-94-16r1, Institute for Sys- tems Research, University of Maryland, College Park, MD 20742, 1997. [28] S. Hess, K. E. Train, J. W. Polak, On the use of a Modified Latin Hypercube Sampling (MLHS) method in the estimation of a Mixed Logit Model for vehicle choice, Transportation Research Part B-Methodological 40 (2006) 147-163. [29] W. H. Greene, D. A. Hensher, J. M. Rose, Using Classical Simulation-Based Estimators to Estimate Individual WTP Values, in: R. Scarpa, A. Alberini (Eds.), Applications of Simu- lation Methods in Envionmental and Resource Economics. Dordrecht, 2005, pp. 17-33. [30] W. M. Hanemann, Welfare Evaluations in Contingent Valuation Experiments with Dis- crete Responses, American Journal of Agricultural Economics 66 (1984) 332-341. [31] J. Huber, K. Train, On the similarity of classical and Bayesian estimates of individual mean partworths, Marketing Letters 12 (2001) 259-269. [32] R. Scarpa, M. Thiene, Destination choice models for rock climbing in the Northeastern Alps: A latent-class approach based on intensity of a latent-class approach preferences, Land Economics 81 (2005) 426-444. [33] Statistical Office for Berlin-Brandenburg, Data on population statistics from the regional reference system, Potsdam, 2012. [34] I. Ring, Ecological public functions and fiscal equalization at the local level in Germany. Ecological Economics 42 (2002) 415-427. [35] I. Ring, Integrating local ecological services into intergovernmental fiscal transfers: The case of the ecological ICMS in Brazil. Land Use Policy 25 (2008) 485-497.

Working Paper on Management in Environmental Planning 32/2013