Author: Anna-Sophia Carmen Zietan

(1)

Author: Anna-Sophia Carmen Zietan Class Code: Master Thesis (EBM868A20) Student number: S3512479

Email: a.c.zietan@student.rug.nl University: University of Groningen

Faculty: Faculty of Economics and Business Specialization: International Economics and Business

Date: June 18, 2018

Supervisor: Prof. Dr. C.J. Jepma Co-assessor: Prof. Dr. J.H. Garretsen

Master Thesis

Willingness to Pay for the Ecosystem.

(2)

List of Figures

Figure 1 Average Monthly Net Income of Households ... 20

List of Tables

Table 1 Flood Protection Measures ... 4

Table 2 DCE 4-Step Process ... 13

Table 3 DCE Attributes and Levels ... 15

Table 4 Abbreviated Results Table ... 22

Table 5 WTP in CLOGIT Model ... 23

(4)

Abstract

Floods cost money where they collide with civilization and so it has become common sense to spend money before a flood occurs, in order to minimize the money that needs to be spend to clean up damages caused by floods. The following thesis will discuss a number of flood prevention methods but ultimately the focus will be on river renaturalization and natural flood plains. In addition to helping manage floods and prevent damages, flood plains offer a number of benefits for environment and people. Unfortunately, these benefits have the nature of public goods and are thus hard to quantify. Opportunity costs of flood plains, on the other hand, are more easily determined, as natural flood plains cannot be used for intensive agriculture or industry. This becomes a problem, where decisions regarding flood protection measures are made based on cost-benefit analysis. Inevitably, the benefits of natural flood plains will be underestimated, unless one found a way to assign a monetary value to them. Precisely that is the goal of this thesis. Using a discrete choice experiment, the thesis aims to determine the market value, the willingness-to-pay (WTP), for the environmental and social benefits of flood plains. The study finds that people are indeed interested in returning rivers to a more natural state and are willing to pay for such measures despite its economic drawback. However, the willingness-to-pay for the creation of natural flood plains further away was slightly higher than for flood plains nearby where the local economy would be hit.

(5)

1. Introduction

In Germany, 80% of the population live within a 5 km radius from the closest water source (Kummu, et al., 2011). Floods as natural phenomenon frequently take place around flowing waters and are a common problem. In 2013, for example, a single summer flood caused damages amounting to 6.68 billion € in Germany (Bundesministerium des Inneren, 2013). Therefore, it is prudent and necessary to take steps to limit the damage. This can be achieved via different flood protection measures. Dams and dikes are some of the most popular amongst the different options of flood protection. Natural water retention measures, such as natural flood plains, serve as flood protection and improve the ecological state of rivers. Both measures have their advantages and disadvantages, yet, the focus of this thesis will be on natural water retention measures and their implications for the ecosystem. The core conflict here is that natural flood plains cannot be used for intensive agriculture, potentially leading to economic losses for local farmers. It is rather easy to quantify these monetary losses whereas it is very difficult to assess which monetary value the positive effects on the ecosystem have. In order to solve this problem, a discrete choice experiment can be used. With its help, a market can be simulated and participants’ willingness to pay for ecosystem improvements and afforestation can be derived.

A number of such DCEs have been carried out regarding climate change mitigation, endangered species protection, flood protection and water quality some of which will be discussed in more detail in chapter 2. All these studies were carried out on a regional level, however, no DCE for ecosystems of rivers in Germany exists to the best of my knowledge. Thus, the purpose of this thesis is to quantify the benefits from ecosystem improvements associated with the creation of natural flood plains. For this purpose a thought experiment was constructed on the bases of a planned flood protection intiative along the river “Weiße Elster” in the district of Greiz, Germany.

(6)

(7)

2. Motivation and State of Knowledge

2.1. Environment

Floods

In the last few years there was an increase in damages caused by floods (Platzbecker, 2017). In order to understand this occurrence, it is necessary to look at the factors that influence the origin and severity of floods.

Floods are natural events and primarily caused by natural factors, such as extreme or persistent rainfalls or thaw (Bund Naturschutz, 2018). Yet, natural factors are not completely independent from human actions. Climate change, for example, is partly human-induced and led to a more frequent occurrence of extreme weather events in Europe in recent years (Platzbecker, 2017). Moreover, human activities can affect the severity and damage potential of floods. First, wetland areas, which function as natural water and CO2 reservoirs, are reduced by drainages (Nagl, 2018). Second, straightened rivers enclosed by dikes have a higher flow velocity and thus the potential for damage increases (Schönauer & Gößwald, 2002). Third, land sealing hinders the percolation of water thus more rainwater is dispensed into the river systems via urban sewage water systems and flow velocity increases (Schönauer & Gößwald, 2002). Finally, it should also be mentioned that construction on and development of potential or former flooding areas increases the damage potential of floods (Schönauer & Gößwald, 2002).

To cope with the occurrence of floods, different flood protection measures can be applied. Flood Protection Measures

A good flood protection strategy should be based on 3 pillars, i.e. 1. flood prevention,

2. technical flood protection,

(8)

Table 1 Flood Protection Measures

Flood protection strategy Corresponding measures

Flood prevention ✓ land utilization:

- refraining from development/construction on land that is at risk of flooding

- relocation of capital assets or smaller settlements out of the flood area ✓ building resilience

- constructing flood-safe buildings ✓ behavioral precaution

- early-warning systems - emergency plans - insurance

Technical flood protection ✓ dikes and barrages ✓ dams

✓ diversion channels ✓ flood retention basins

✓ controlled retention areas (e.g. polders) ✓ reservoirs and artificial lakes

Natural flood protection ✓ dike relocation

✓ renaturation of river beds ✓ unsealing of land

✓ afforestation ✓ extensive land use

✓ uncontrolled retention areas (e.g. flood plains)

Flood prevention measures aim to reduce the damage potential in the event of a flood, however, they do not influence the flood, meaning the magnitude or duration of the flood, itself. Such measures include preventive measures regarding land utilization (i.e. refraining from development/construction on land that is at risk of flooding, relocation of capital assets or smaller settlements out of the flood area), building resilience (i.e. constructing flood-safe buildings), and behavioral precaution (i.e. early-warning systems, emergency plans, insurance, etc.) (Nagl, 2018).

(9)

Technical flood protection includes dikes and barrages, dams, diversion channels, flood retention basins, creation of controlled retention areas (e.g. polders), and usage of reservoirs and artificial lakes for flood protection by regulating the barrage in accordance with flood requirements (Nagl, 2018). They are essential components of the flood protection concepts in many locations. Nonetheless, experts agree that higher and more massive walls do not necessarily lead to a more effective protection (Schönauer, 2002). Accordingly, even technical flood protection has its limits. Furthermore, technical flood protection, especially in the form of dikes, walls, and measures increasing the outflow capacity, are associated with a higher risk of floods at the upper and the lower reaches of the river (Denhardt, et al., 2008).

Natural flood protection comprises dike relocation, renaturation of river beds, unsealing of land, afforestation, extensive land use, and the creation of uncontrolled retention areas (e.g. flood plains) (Nagl, 2018). The latter will be looked at in a more detail below. Measures of natural water retention offer flood protection on site and possibly have a positive effect on the flood situation in more distant areas around it. The natural flooding areas of the Elbe, for example, led to a reduction of the water levels up to 30 km in anterior areas (Schäfer & Kowatsch, 2015). Since natural water retention needs space, it potentially conflicts with economic interests like agriculture, trade, and industry close to water.

Conflicts of Interest

The Water Framework Directive (WFD) 2000/60/EC of the European Union prescribes that all surface and ground waters should achieve the good (ecological) status, which is defined by a number of factors relating to water quality, biodiversity and structural attributes of a body of water (Bayrisches Landesamt für Umwelt, 2018). The ecological status of a body of water can be expressed using a 5-point scale (Umwelt Bundesamt, 2016).

Grade 1 means that the body of water is in very good ecological condition and there are no man-made changes. Grade 2 is the good status as aspired in the WFD with limited human intervention. Grade 3 stands for moderate status. Grade 4 stands for dissatisfactory status. Grade 5 means that the state of the body of water is poor and that there have been substantial man- made changes.

(10)

likely associated with economic land use. The goal of the WFD, as mentioned, is grade 2, but with consideration of economic interests. No country can afford not altering any area around flowing waters because agricultural space or buildings on-site must be protected from floods. The use of land thus does not only affect flood incurrence, it is an economically relevant question. In order to answer this question, it is necessary to be able to quantify the benefits of a more ecological state of flowing waters. Flood plains are both a good example for an ecological good status and a measure of ecological flood protection. Therefore, flood plains will be looked at more closely in the following section.

Flood Plains and their Benefits

Natural flood plains, which can be used as natural flood protection, provide many benefits; e.g. they:

✓ improve the quality of groundwater and surface waters such as rivers and oceans. The use of fertilizers in agriculture leads to higher nutrient levels in rivers. This can lead to a deterioration of the water quality and harm the river´s ecosystem, possibly causing excessive algae growth and fish kill (Schäfer & Kowatch, 2015). At the bottom of natural flood plains, nutrients and sediments settle down and bacteria convert nitrate into elemental nitrogen via a process called “denitrification”. The natural alteration between high and low tide leads to frequent water exchange and improves the quality of ground water (Schönauer & Gößwald, 2002). Thus, natural flood plains serve as filters. From an economic viewpoint, thus, costly measure for water treatment can be reduced (Schäfer & Kowatch, 2015).

✓ provide climate protection via moor-rich flood plains.

(11)

✓ protect biodiversity.

Natural flood plains are among the most species-rich habitats in Europe, as about two- thirds of all central European biocenosis can be found in flood plains (WWF, 2018). For example, fish use flood plains for protection, as food source, spawning and breeding grounds, thus flood plains directly affect the fish population and their biodiversity (Wolter & Schomaker, 2007). The alteration between high and low tide in natural flood plains has a dual effect on the flora, i.e. floods clear clogged river beds, prevent culmination, and facilitate an exchange of water plants along rivers (Schönauer & Gößwald, 2002).

✓ provide leisure and recreational opportunities.

Natural flood plains do not only appeal to flora and fauna but also offer an attractive getaway for city dwellers by providing room for leisure, recreation and a number of outdoor activities (Schäfer & Kowatch, 2015).

The mentioned benefits, however, are very difficult to quantify. As a consequence, they are mostly left out when cost-benefit analyses are conducted to decide whether a flood protection project should be implemented. The amount of money that it would cost if a flood occurred is assessed instead. In consequence, a more adequate benefit calculation requires the quantification of qualitative benefits. Discrete choice experiments (DCE) can be applied for this purpose.

2.2. Discrete Choice Experiment

(12)

approach offers higher flexibility than revealed preference approach (Louviere, et al., 2000). The willingness to pay estimate, which is needed for cost-benefit calculation, can also be obtained via discrete choice analysis (Hoyos, 2010).

A number of DCEs regarding different environment policies have been conducted which will be briefly discussed below. Layton & Brown (2000) estimated the willingness-to-pay to mitigate the future impact of climate change and found that WTP did not differ significantly between a medium (60 years) and large (150 years) time horizon. Similarly, Longo, et al. (2012) also found that people do care about the long-term horizon even if it is evident that they will not face the consequences themselves. In addition, they found that factors such as cultural identity, political affiliation and participation in environmental organization did have an effect on their WTP estimates (Longo, et al., 2012). Thus, individual-specific variables such as prior knowledge and experiences should be obtained in addition to the usual individual-specific demographic variables.

(13)

3. Theory and Model

3.1. Theory

In the context of Discrete Choice Analysis, two theories are of major importance (Lancsar & Louviere, 2008; Hoyos, 2010): Lancastarian Demand Theory and McFadden´s Random Utility Theory.

Lancaster´s (1966) demand theory breaks away from the traditional assumption that utility comes directly from the good itself, instead, utility is derived from the properties or characteristics of a good. Hence, preference is a ranking of (collections of) characteristics rather than goods, while goods are only indirectly referred to via the characteristics they embody (Lancaster, 1966). All goods consist of several characteristics, and the same characteristic may be shared across several goods (Lancaster, 1966). Consumers choose which (collections of) characteristics they prefer, however, they do not choose which characteristics make up a good (Lancaster, 1966). The theory assumes that the relationship between characteristics of a good and the consumption of that good is objective and linear, and thus, holds true for all individuals (Lancaster, 1966).

Random utility theory assumes that consumers seek to maximize their utility (Komarek, et al., 2011). However, utility is latent and cannot be directly observed only approximated by an individual´s choice, for example in the context of a DCE (Hoyos, 2010). The choice of the individual will depend on a number of factors such as “(1) the objects of choice and sets of alternatives available to decision-makers, (2) the observed attributes of decision-makers, and (3) the model of individual choice and behavior and distribution of behavior patterns in the population” (McFadden, 1973, pp. 106-107). Some of these factors such as the characteristics of the choices and some attributes of the individual can be observed, however, the third factor remains as the unobservable component of an individual´s choice (Komarek, et al., 2011). This unobserved component may stem from not modeled characteristics of the choice, unobserved attributes of the individual, or simply differences in preferences due to heterogeneity within the population (Sammer, 2007).

(14)

where U is the utility that individual i gains from alternative j, V is the observable (or systematic) component and ε is the unobservable (or random) component. The systematic component in turn is modeled as:

𝑉𝑖,𝑗 = 𝛼𝑋𝑖+ 𝛽𝑍𝑖,𝑗

where X is the vector of observed attributes of the individual i, Z is the vector of the attributes of the alternative j as interpreted by individual i, and α resp. β are parameters of X respectively Z. Note that X are alternative-invariant independent variables while Z are alternative-variant independent variables.

The probability P that individual i chooses alternative k over all other alternatives j can be written as:

𝑃_𝑖,𝑘 = 𝑃𝑟𝑜𝑏(𝑈_𝑖,𝑘 > 𝑈_𝑖,𝑗); ∀j ≠ k

This does not give us a complete ranking of preferences from most to least preferred option, however, the discrete choice gives us the most preferred option from a set of competing alternatives (Louviere, et al., 2000). As utility depends on the characteristics of an alternative (Lancaster, 1966), the probability P may be rewritten as:

𝑃_𝑖,𝑘 = 𝑃𝑟𝑜𝑏(𝑉_𝑖,𝑘 + 𝜀_𝑖,𝑘 > 𝑉_𝑖,𝑗+ 𝜀_𝑖,𝑗); ∀j ≠ k

or in other words:

𝑃_𝑖,𝑘 = 𝑃𝑟𝑜𝑏(𝑉_𝑖,𝑘 − 𝑉_𝑖,𝑗 > 𝜀_𝑖,𝑗− 𝜀_𝑖,𝑘); ∀j ≠ k

where the probability that individual i chooses alternative k is calculated as:

𝑃𝑟𝑜𝑏(𝑦_𝑖 = 𝑘) = exp (𝛼𝑋𝑖 + 𝛽𝑍𝑖,𝑘) ∑𝐽_𝑗=1𝑒𝑥𝑝(𝛼𝑋_𝑖+ 𝛽𝑍_𝑖,𝑗)

Note that in estimating the probabilities, the parameters of one alternative are usually set to zero to solve the identification problem (Hill, et al., 2012).

3.2. Model

(15)

essence, is the concept of the contingent valuation (CV) method as employed by Drake et al. (2013) for example. Deviating from that, a DCE involves at least two alternatives of which either one or the other must be chosen (Hoyos, 2010). Such models can be estimated using probit or logit models of binary choice (Hill, et al., 2012). The difference between these probit and logit models lies in the assumptions made regarding the distribution of the error component. The probit model assumes a standard normal distribution, while the logit model assumes a logistic distribution (Hanemann, 1984). Logit models such as the multinomial logit (MNL) model have long been used to analyze discrete choice data as their computation is relatively simple (Hausman & McFadden, 1984). By contrast, the multinomial probit (MNP) model is more complicated as it assumes normal distributions of all unobserved components and thus leads to a multivariate density function (Hoyos, 2010; Layton & Brown, 2000). Based on its greater simplicity and its application in similar studies (i.e. Che, et al. (2014), Layton & Brown (2000), etc.) this thesis will employ a logit model.

Basic MNL models are used to model systematic taste variations based on individual specific (or alternative-invariant) variables (Hill, et al., 2012). With regards to Lancaster´s demand theory, however, a model that also accounts for changes in alternative specific attributes will be more useful in the context of discrete choices. A conditional logit (CL) model allows for individual- and alternative-specific variables to be modeled (Hill, et al., 2012). Unfortunately, a general problem with logit models is the independence of irrelevant alternatives (IIA) restriction (Layton & Brown, 2000). The IIA restriction states that the relative probabilities among a subset of alternatives remain stable even if an alternative is introduced or removed, for example, “if two choices exist, say car and bus in a transportation choice application, that addition of a third choice, subway, will not change the ratio of probabilities of the initial two choices” (Hausman & McFadden, 1984, p. 3). Another issue is dealing with heterogenous consumer preferences and random taste variations (Sarrias, 2016). According to Boxall & Adamowicz (2002) there are three ways to consider heterogeneity: (1) via demographic variables, for example as in Zhai & Suzuki (2008); (2) via segmentation of respondents according to sociodemographic variables, for example as in Komarek, et al. (2011); or (3) via employing a random variable probit or logit models (Boxall & Adamowicz, 2002).

(16)

(17)

4. Methods and Data

According to Hoyos (2010) designing a DCE is a four-step process. Table 2 shows a brief summary of the four steps and the most important specifications that need to be determined in each step.

Table 2 DCE 4-Step Process

Steps Specifications

Definition of Attributes and Levels

✓ Specification of purpose ✓ Specify attributes and levels Experimental Design ✓ Labeled / unlabeled alternatives

✓ Full / fractional factorial design ✓ Main / interaction effects

✓ Construction of choice combinations Questionnaire

Development

✓ Providing background and contextual information ✓ Coding of questions

✓ Piloting and improvements Data Collection ✓ Sampling strategy

✓ Survey administration

4.1. Definition of Attributes and Levels

(18)

the loss of agricultural land whereas residents of the surrounding cities appreciate the benefits of a recreation area won by this. In order to investigate this, a thought experiment was designed. At first, the backstory of the thought experiment shall be described. Along a fictious river called “Schöne Elster” there are two small towns, Oberstadt and Unterstadt. The study participants were asked to imagine that they live in Unterstadt. Oberstadt and Unterstadt are 20 km apart. Both are characterized by intense farming and are affected by floods. The river has been straightened and is surrounded by old ramshackle dikes. The ecological status of the river is poor (grade 5 on the scale for the ecological state of rivers). The starting point is that the deficient flood protection of both towns ought to be improved. For this purpose, the participant were asked to choose their preferred flood protection strategy.(“initiative”) from a set of three alternatives (Initiative 1, Initiative 2, Initiative 3) where the initiatives may differ in four attributes plus one interaction of attributes. A summary of the attributes and levels can be found in Table 3.

Flood protection is indicated by means of the HQ value. HQ stands for the rate with which the water drains off. The number that comes after it (often as an index number) tells the statistical probability of recurrence. HQ5 thus means that a flood happens statistically every five years. In the experiment, the two towns and the land around it each have a HQ20, which means that they are flooded every 20 years. The HQ of the towns themselves and the potential flood plains must be distinguished, for reference please view the images in Appendix H which was also used in the survey to visualize setup of the task.

(19)

Table 3 DCE Attributes and Levels

Attributes Levels

Flood protection strategy in Oberstadt (OS)

o No flood plain (grade 5)

o Flood plain with intensive farming (grade 3) o Natural flood plain (grade 1)

Flood protection strategy in Unterstadt (US)

o No flood plain (grade 5)

o Flood plain with intensive farming (grade 3) o Natural flood plain (grade 1)

Ecological status of the river Schöne Elster (WASSER)

o Grade 1 = the body of water is in very good ecological condition and there are no man- made changes.

o Grade 2 = the good status as aspired in the WFD with limited human intervention.

o Grade 3 = moderate status o Grade 4 = dissatisfactory status

o Grade 5 = the state is poor and there have been substantial man-made changes

Afforestation (BAUM) _{o No trees}

o Planting trees somewhere else

o Planting trees in Oberstadt / Unterstadt Price (PREIS) o 5 € per person, per month

o 7 € per person, per month o 9 € per person, per month

(20)

The ecological status of the river Schöne Elster results from the average ecological status in Oberstadt and Unterstadt. Hence, this value is dependent on the use of the areas around Oberstand and Unterstadt. If the area around Oberstadt, for example, is not used as a flood plain but for intensive farming, which entails a poor ecological condition (5), and the area around Unterstadt is used as a flood plain, which leads to a very good ecological status (1), the average ecological status is a moderate one (3).

Planting trees could also serve as flood protection but is rather important for climate protection. Trees could either be planted in Oberstadt or Unterstadt in the form of parks or next to streets, etc., or somewhere else between Oberstadt and Unterstadt or not at all.

The prices were based on information gathered in conversations with experts from the ThLG. The estimated costs for the formerly mentioned project in the north of the district Greiz were used as references for the prices in the thought experiment. Said project comprises a planned renewal of dikes and the creation of flood plains, whereby a flood protection of HQ100 on- site for the next 100 years would be achieved. The direct beneficiaries of these measures would be the 3,100 residents in the area, who were also taken into account in the benefit-cost analysis of the ThLG. An average ecological status of grade 2 to 3 could be achieved. Thus, this project is corresponds with option 2 in the thought experiment. The total costs of the project are estimated to be 25 million euros. The costs that can be avoided by this project amount to 20 million euros, which can be seen as the benefit of the project. Thus, the costs outweigh the benefits by 5 million euros. This, however, could be counterbalanced by the potential increase of qualitative benefits by the improved ecological status. Therefore, the present master thesis is aiming at assessing the monetary value of these qualitative benefits.

As said before, 25 million euros would be spent on a better ecosystem in combination with moderate farming with HQ10. Hence, the calculation of the costs per person per month looks as follows:

((25 Mio € / 3100 people) /100 years)/ 12 months per year = 6.72 € per person and month. This figure was rounded to 7€ and was used for option 2 of the thought experiment.

(21)

((20 Mio € / 3100 people) /100 years)/ 12 months per year = 5.38 € per person and month. This figure was rounded to 5€ for the thought experiment.

For the last choice option of creating a natural flood plain and gaining a very good ecological status, there was no estimate since the ThLG had ruled out this option as impossible to implement in the concerned area beforehand. For obtaining a regularity of price levels, a price of 9€ was determined for this last option.

The costs of the planting of 130 trees was also estimated but it amounted to such a small amount (namely 0.00€ for planting no trees, 0.11€ for planting trees somewhere else, and 0.26€ for planting trees in Oberstadt / Unterstadt) that it was completely dropped from the thought experiment. All calculations can be found in Appendix A.

In the experimental context, the price for the chosen measures ought to be paid by the citizens of Oberstadt and Unterstadt themselves. The participant is to imagine that he or she will pay for the chosen measure a monthly price. The participants were asked to answer truthfully and under consideration of their actual monthly income.

4.2. Experimental Design

The DCE contains 4 attributes with 3 levels each, which means that there are 34 = 81 possible treatment combinations. In a full factorial design each combination would have to appear at least once, however, would lead to large and inefficient choice set sizes or require excessive numbers of choice sets. Thus, a fractional factorial design was applied. The design was generated via two independent ternary equations, leading to the smallest known regular three level design according to Street & Brugess (2007) with 34-2 = 9 different treatment combinations. The independent equations that were used are:

𝑏₄ = 𝑏₁+ 𝑏₂

𝑏₃ = 𝑏₁+ 2 ∗ 𝑏₂

where b1, b2, b3 and b4 represent the four main attributes OS, US; BAUM and PREIS respectively. The calculation was performed modulo 3 and the attribute WASSER was determined on the basis of the level of OS and US for each combination.

(22)

difficult. The choice sets were determined by applying the two difference vectors (0,1,1,1) and (1,0,1,1), thus, each choice set will be of the underlying structure (0000,0111,1011). This set up ensures that all main effects as well as the two-factor interaction between OS and US can be estimated.

The alternatives (“initiatives”) in each choice set were labeled “Initiative 1”, “Initiative 2” and “Initiative 3” respectively, thus, the alternatives are considered unlabeled. Whether alternatives are labeled or unlabeled has implications for the results, with labeled alternatives the label itself holds information which may reduce the attention afforded to the attributes (de Bekker-Grob, et al., 2010). Since evaluating the attributes of the different flood protection measures is of essence here, an unlabeled DCE will be better to ensure focus on the actual task.

This DCE is a forced choice experiment as no alternative specific constants (ASC) were included. Usually, the inclusion of such status quo alternative is favored, as excluding ASCs can lead to biases, however, ASCs may not work well with unlabeled choice experiments (Hoyos, 2010). On a more practical note the status quo alternative was left out in an effort to keep choice sets as small as possible.

4.3. Questionnaire Development

The survey was conducted online via a web-based questionnaire, which had the advantages that web-based questionnaires can employ the graphical user interface and data is automatically entered into the databases avoiding copying errors (Komarek, et al., 2011). Moreover, the choice sets could be made visible in an appealing way in this questionnaire. Graphics could be used to make it easier for the participants to understand and imagine the situation of Oberstadt and Unterstadt.

(23)

“Statistik”, covered demographic questions. Finally, the fifth part, “Kommentare und Anregungen”, provided space for the participants’ feedback.

Initially, the questions, the explanatory texts and the choice sets of the decision experiment were revised and improved in an iterative process according to the feedback of participants who tested the questionnaire beforehand. The number of initiatives per choice set, for instance, was altered from four to three, and the design of the choice sets was made clearer to simplify and abbreviate the processing. The final version of the questionnaire contained 36 questions on 41 pages. The average processing time was 23 minutes.

4.4. Data Collection

(24)

5. Results

5.1. Descriptive Statistics

The online survey generated 115 responses from all over Germany. Even though it would be preposterous to assume that 115 responses can adequately estimate the preferences of the whole country, the following short section will ascertain how well the survey data fits the German population. To determine the fit, three demographic variables were compared: sex, age and income.

The sex ratio of survey participants is 0.85 men per woman, while the sex ratio in Germany in 2017 was 0.97 men per woman (CIA, 2018). In both cases there are slightly more women than men. The median age of the survey participants is 30 years, while the median age in Germany is 46 years (Worldometers, 2018). The survey, thus, attracted a much younger audience which might be due to the realization of the survey in the form of an online questionnaire and the reliance on websites, online portals and social media distribution. Consequently, 34% of survey participants were students. Finally, the distribution of monthly net income in percent for both, the survey households in 2017 and the German households in 2016 (Statista, 2018) are compared in Figure 1. Even though the correlation between the two distributions is quite high (0.9842) the survey data is clearly skewed towards the lower income regions.

Figure 1 Average Monthly Net Income of Households

(25)

In an initial opinion poll, the participants were confronted with a number of statements regarding flood protection measures. Only after answering the question participants were provided with additional information about the benefits and disadvantages regarding the flood protection measures that were relevant in the DCE, namely dykes and natural flood plains. Responses can be found in Appendix B. Participants signaled high levels of approval for all proposed technical and natural flood protection measure. However, the strongest agreement, with 60.87% of the participants completely agreeing, relates to a flood prevention measure, namely to refraining from construction of buildings in potential flood areas. Approval for the remaining two flood prevention methods, however, is visibly lower. This could be the case because refraining from the construction of buildings in flood areas is something that lies in the future. No one would be forced to change their way of living because of this decision.

5.2. Econometric Analysis

Data from the DCE was analyzed using NLOGIT 6 using the conditional logit model (CLOGIT) and the random parameter logit model (RPLOGIT). The estimations were initially based on a linear indirect utility function of the form:

𝑉_𝑖,𝑗 = 𝛽₁𝑂𝑆 + 𝛽₂𝑈𝑆 + 𝛽₃𝑊𝐴𝑆𝑆𝐸𝑅 + 𝛽₄𝐵𝐴𝑈𝑀 + 𝛽₅𝑃𝑅𝐸𝐼𝑆

where V is the indirect utility of individual i from initiative j.

The RPLOGIT model was applied using the panel command with nine choice sets for each individual. The following specifications regarding the distributions of the random parameter were made, i.e. random effects for the variable BAUM were drawn form a normal distribution, while random effects for variables OS and US were drawn from a triangular distribution. The triangular distribution is based on the standard continuous uniform distribution and may be used as an alternative to the normal distribution that restricts the range of variation in a parameter (Department for Transport, 2014; Greene, 2016). In addition, the Halton command was added, which draws random numbers in a more intelligent way, thereby reducing the number of iterations and the simulation error (Greene, 2016).

(26)

model were each estimated for only alternative-specific variables and for alternative- and individual-specific variables. Table 4 shows the abbreviated results of these estimations, the full table (including the individual-specific variables) can be found in Appendix E and the corresponding NLOGIT commands and outputs in Appendix D. A list of the abbreviations of the individual-specific variables is provided in Appendix F.

Table 4 Abbreviated Results Table

CLOGIT RPLOGIT Rhs Rhs+Rh2 Rhs Rhs+Rh2 OS 0.6569*** 0.6627*** 1.0349*** 1.0652*** US 0.6569*** 0.6438*** 1.0038*** 1.0707*** BAUM 0.3030*** 0.3061*** 0.3715*** 0.3808*** PREIS -0.1282*** -0.1301*** -0.1293*** -0.1310*** A_INIT1 -0.1112 -0.0199 0.1527 0.5851 A_INIT2 0.3695*** 0.0402 0.4396*** 0.3942 TsOS_X 2.0703*** 2.1434*** TsUS_X 2.2914*** 2.4620*** NsBAUM_X 0.1827 0.1743 Individual- specific Var see Appendices E & F see Appendices E & F Log Likelihood -1115.3286 -1115.3286 -1115.3286 -1115.3286 R2 0.1475 0.1563 0.1988 0.2108 R2 Adj 0.1450 0.1439 0.1953 0.1981 Pseudo R2 - - 0.2140854 0.2259305

(27)

A special feature regarding the goodness-of-fit of the RPLOGIT models is the likelihood ration index (or pseudo R-squared) which is used to measure the goodness-of-fit for discrete choice data (Hoyos, 2010). According to Hoyos (2010) a value greater than 0.2 already indicates a well fitted model, thus the models estimated appear satisfactory with regards to this indicator. For all four models the alternative-specific variables are highly significant and show the same signs. OS, US and BAUM have positive coefficients indicating that providing a higher level of an attribute, i.e. better ecological state in the case of OS and US, and more trees close by in the case of BAUM, increases the likelihood of choosing that initiative. PREIS has a negative coefficient meaning a higher price decreases the likelihood of that alternative being chosen. The observed characteristics of the individuals, on the other hand, seem to have little effect on the choice behavior and are mostly insignificant. Exceptions to this are the variables coding personal experience of discomfort due to flooding (F19_1), specialized knowledge regarding floods or flood protection due to occupation (F22_2), age (F24_2), and income (F34_2). Lastly, the individual WTP-estimates and the mean WTP with a 95% confidence interval were calculated. Tables 5 and 6 (for full tables see Appendix G) shows the mean WTP and standard error for both models. In both cases WTP for OS is slightly higher than for US, which may indicate that indeed, participants placed greater value on better ecosystems somewhere else, if a better ecological state at home comes at the price of local agriculture.

Table 5 WTP in CLOGIT Model

WTP for… Mean Std. Err.

OS 5.12385 0

US 4.96687 0

BAUM 2.36356 4.16e-17

Table 6 WTP in RPLOGIT Model

WTP for… Mean Std. Err.

OS 7.82396 0.46546

US 7.70921 0.54493

(28)

6. Conclusion

The purpose of this thesis was to determine the willingness to pay for better ecosystems along rivers in Germany. In addition, the DCE was designed to also examine whether WTP differs if ecosystem improvements are carried out on-site or someplace further away. The study found that WTP for a better ecosystem exists, however, improvements further away scored a slightly higher WTP than ecosystem improvements on-site. A relevant consideration here is the fact that if land is used for the renaturation of a river its usability for economic purposes, such as agriculture, declines. In addition, a positive WTP was determined for planting trees locally, respectively on-site.

A number of limitations may apply to the results generated in this thesis. First and foremost, the data collected may be biased towards a younger audience due to the execution of the survey in the form of a web-based survey. Furthermore, highly complex choice tasks may overburden participants. In this case, participants may be inclined employ very simple decision-making heuristics which may lead to non-compensatory heuristics that ignore a number of attributes (Hoyos, 2010). This may be the case here, as the topic of floods and flood protection required a significant amount of background information and the thought experiment was rather complex as well. This criticism was confirmed by some of the feedback received from participants on the feedback portion of the questionnaire. In addition to the choice task being complicated itself, a large number of consecutive choice tasks may lead to loss of attention and negatively impact an individual´s preference stability (Day & Prades, 2010). According to Bech, et. al. (2011) nine choice sets are already considered a lot, which was also confirmed by participants. Finally, there is the question of validity of whether the respondents’ stated choice, especially in a hypothetical market, is conform with their actual behavior or not (Hoyos, 2010). In the case of a public good, the option to perform an external validity test by comparing stated to revealed preference does not exist.

(29)

(30)

7. References

Bateman, I., Day, B., Georgiou, S. & Lake, I., 2006. The Aggregation of Environmental Benefit Values: Welfare Measures, Distance, Decay and Total WTP. Ecological Economics. Bayrisches Landesamt für Umwelt, 2018. Grundlagen und Ziele der Wasserrahmenrichtlinie. [Online] Available at:

https://www.lfu.bayern.de/wasser/wrrl/grundlagen_und_ziele/index.htm

Bech, M., Kjaer, T. & Lauridsen, J., 2011. Does the Number of Choice Sets Matter? Results from a Web Survey Applying a Discrete Choice Experiment. Health Economics.

Boxall, P. & Adamowicz, W., 2002. Understanding Heterogeneous Preferences in Random Utility Models: A Latent Class Approach. Environmental & Resource Economics.

Bund Naturschutz, 2018. Hochwasserkatastrophen sind Menschenwerk. [Online] Available at: https://www.bund-naturschutz.de/wasser/hochwasser.html [Accessed 01.05.2018].

Bundesministerium des Inneren, 2013. Bericht zur Flutkatastrophe 2013: Katastrophenhilfe, Entschädigung, Wiederaufbau, s.l.: s.n.

Campbell, D., Hutchinson, W. & Scarpa, R., 2008. Incorporating Discontinuous Preferences Into the Analysis of Discrete Choice Experiments. Environmental and Resource Economics. Chen, H. & Cosslett, S., 1998. Environmental Quality Preference and Benefit Estimation in Multinomial Probit Models: A Simulation Approach. American Jounral of Agricultural Economics.

Che, Y. et al., 2014. Residential Preferences for River Network Impovement: An Exploration of Choice Experiments in Zhujiajiao, Shanghai, China. Environmental Management.

CIA, 2018. The World Factbook. [Online] Available at:

https://www.cia.gov/library/publications/the-world-factbook/geos/gm.html [Accessed 01.05.2018].

(31)

de Bekker-Grob, E. et al., 2010. Labeled versus Unlabeled Discrete Choice Experiments in Health Economics: An Application to Collateral Cancer Screening. Value in Health.

Denhardt, A., Hirschfeld, J., Drünkler, D. & Peschow, U., 2008. Kosten-Nutzen-Analyse von Hochwasserschutzmaßnahmen, s.l.: Umwelt Bundesamt.

Department for Transport, 2014. Supplementary Guidance: Mixed Logit Models, London: s.n. Drake, B., Smart, J. & Termansen, M., 2013. Public Preferences for Production of Local and Global Ecosystem Services. Regional Environment Change.

Ewing, G. & Sarigöllü, E., 2000. Assessing Consumer Preferences for Clean-Fuel Vehicles: A Discrete Choice Experiment. Journal of Public Policy & Marketing.

Greene, W., 2016. NLOGIT Version 6: Reference Guide. New York: Econometrics Software Inc.

Hanemann, W., 1984. Welfare Evaluations in Contingent Valuation Experiments with Discrete Responses. American Journal of Agriculutural Economics.

Hausman, J. & McFadden, D., 1984. Specification Tests for the Multinomial Logit Model. Econometrica.

Hill, R., Griffiths, W. & Lim, G., 2012. Principles of Econometrics. 4th ed. Asia: John Wiley & Sons.

Hoyos, D., 2010. The State of the Art of Environmental Valuation with Discrete Choice Experiments. Ecological Economics.

Johnson, F. & Mathews, K., 2001. Sources and Effects of Utility-theoretic Inconsistency in Stated Preference Surveys. American Journal of Agricultural Economics.

Komarek, T., Lupi, F. & Kaplowitz, M., 2011. Valuing Energy Policy Attributes for Environmental Management: Choice Experiment Evidence from a Research Institution. Energy Policy.

Kummu, M., de Moel, H., Ward, P. & Varis, O., 2011. How Close Do We Live to Water? A Global Analysis of Population Distance to Freshwater Bodies. PLoS One.

(32)

Lancsar, E. & Louviere, J., 2008. Conducting Discrete Choice Experiments to Inform Healthcare Decision Making: A User´s Guide. Pharmacoeconomics.

Layton, D. & Brown, G., 2000. Heterogeneous Preferences Regarding Global Climate Change. The Review of Economics and Statistics.

Longo, A., Hoyos, D. & Markandya, A., 2012. Willingness to Pay for Acillary Benefits of Climate Change Mitigation. Environment Resource Economics.

Louviere, J., Hensher, D. & Swait, J., 2000. Stated Choice Methods: Analysis and Application. New York: Cambridge University Press.

McFadden, D., 1973. Conditional Logit Analysis of Quantitative Choice Behavior. In: P. Zarembka, ed. Frontiers in Econometrics. New York: Academic Press.

Michel, B., Plättner, O. & Gründel, P., 2011. Klima-Hotspot Moorböden, Braunschweig: 02/2011 ForschungsReport.

Nagl, G., 2018. Ökologischer Hochwasserschutz - Raum für naturnahe Auen, Gewässer und Feuchtgebiete und Schutz für die Menschen. [Online] Available at: https://www.bund-naturschutz.de/wasser/hochwasser.html [Accessed 01.05.2018].

Platzbecker, E., 2017. The Impact of Climate Change on the Occurrence of Floods and the Resulting Economic Costs in Europe, Groningen: Master Thesis University of Groningen. Puckett, S. & Hensher, D., 2009. Revealing the Extent of Process Heterogeneity in Choice Analysis: An Empirical Assessment. Transportation Research Part A: Policy and Practice. Sammer, K., 2007. Einflus von Ökolabling auf die Kaufentscheidung: Evaluation der

Schweizer Energieetikette Mittels Discrete-Choice-Experimenten, s.l.: Dissertation University of St. Gallen.

Sarrias, M., 2016. Discrete Choice Models with Random Parameters in R: The Rchoice Package. Journal of Statistical Software.

(33)

Schönauer, S., 2002. Als Flüsse zu Straßen wurden - Der Gewässerausbau europäischer Ströme und seine Folgen für die Umwelt, s.l.: Bund für Umwelt und Naturschutz Deutschland.

Schönauer, S. & Gößwald, M., 2002. Breitwasser statt Hochwasser - Unsere Flüsse brauchen mehr Raum, s.l.: Bund für Umwelt und Naturschutz Deutschland.

Statista, 2018. Verteilung der Privathaushalte in Deutschland nach monatlichem Haushalts-nettoeinkommen im Jahr 2016. [Online] Available at:

https://de.statista.com/statistik/daten/studie/3048/umfrage/privathaushalte-nach- monatlichem-haushaltsnettoeinkommen/ [Accessed 01.05.2018].

Street, D. & Burgess, L., 2007. The Construction of Ooptimal Stated Choice Experiments: Theory and Methods. Hoboken: John Wiley & Sons.

Umwelt Bundesamt, 2016. Ökologischer Zustand der Fließgewässer. [Online] Available at: https://www.umweltbundesamt.de/daten/wasser/fliessgewaesser/oekologischer-%20zustand-der-fliessgewaesser#textpart-2 [Accessed 01.05.2018].

Wolter, C. & Schomaker, C., 2007. Die fischökologische Bedeutung von Flussauen und ihre Berücksichtigung bei der fisch-basierten Gewässer-Zustandsbewertung. In: N. U. Odertal, ed. Nationalpart-Jahrbuch Unteres Odertal (4). s.l.:s.n., pp. 28-37.

Worldometers, 2018. Germany Population. [Online] Available at:

http://www.worldometers.info/world-population/germany-population/ [Accessed 01.05.2018]. WWF, 2018. Die Biologische Waschanlage der Flüsse. [Online] Available at:

https://www.wwf.de/themen-projekte/fluesse-seen/lebensraeume/auen/ [Accessed 01.05.2018].

(34)

Appendix A: Price Calculations

The levels for the attribute price were determined based on the prices of the cost associated with the renewal of dikes, creation of (natural) flood plains and afforestation.

Renewal of dikes + creation of (natural) flood plains1

[0] No Flood Plain

((20 Mio € / 3100 people) /100 years)/ 12 months per year = 5.38 € ~ 5 € per person and month

[1] Flood Plain with Intensive Farming

((25 Mio € / 3100 people) /100 years)/ 12 months per year = 6.72 € ~ 7 € per person and month

[2] Natural Flood Plain [1] + ([1] – [0]) = 9 € → 7€ + (7€ - 5€) = 9€ per person and month Afforestation2

Planting of one tree for 24 people: 3100 / 24 = 130 [0] No Trees 0€

[1] Planting of 130 Trees Somewhere Else

(((3200 €/ tree) x 130 trees) / 100 years) /12 months per year = 11 cent* [2] Planting 130 Trees in OS & US

((7300 €/ tree x 130 trees) / 3100 people / 100 years) / 12 months per year = 26 cents* *Planting in OS or US assumes that trees will be planted inside the town where the ground has already been sealed (i.e. with streets, pedestrian walks, market places, etc.). Consequently, the ground needs to be unsealed and prepared. This cost does not arise “Somewhere Else” as these trees are assumed to be planted outside of developed areas where costs are lower.

1_{Costs were taken from a (similar) project in the district of Greiz. Cost estimates for this project were provided}

by the ThLG.

2_{Source online:}

(35)

Appendix B: Opinion Poll

Flood protection (FP) type I… completely agree rather agree rather disagree completely disagree don’t know Technical FP Dikes and dams should be made higher and more massive. 31.88% 47.83% 11.59% 3.62% 5.07%

Technical FP Reservoires should be included in flood protection concepts via regulation of barrages oriented on flood levels.

46.38% 40.58% 5.07% 0.00% 7.97%

Natural FP Natural retention areas should be created by relocating dikes and renaturation of water bodies.

55.07% 33.33% 5.07% 1.45% 5.07%

Natural FP Land that is not or minimal used (e.g. closed down commercial space) should be unsealed so that water can drain away.

46.38% 36.23% 10.14% 4.35% 2.90%

Natural FP The ability of the soil to absorb water should be increased by afforestation.

50.00% 32.61% 9.42% 1.45% 6.52%

Flood prevention There should not be any construction on areas that are very likely to be flooded.

60.87% 31.16% 5.80% 0.72% 1.45%

Flood prevention Assets in areas that are very likely to be flooded (residential buildings or industrial /commercial facilities) should be relocated if possible.

15.22% 33.33% 31.88% 12.32% 7.25%

Flood prevention Residents should more attend to private provision (building precaution, insurance, etc.)

(36)

Appendix C:

NLOGIT Output for Models with WASSER

CLOGIT;Lhs=CHOIC

;Choices=INIT1,INIT2,INIT3

;Rhs=one,OS_X,US_X,BAUM_X,WASSER_X,PREIS_X$

Discrete choice (multinomial logit) model

Dependent variable Choice

Log likelihood function -950.87305 Estimation based on N = 1035, K = 7 Inf.Cr.AIC = 1915.7 AIC/N = 1.851

---Log likelihood R-sqrd R2Adj Constants only -1115.3286 .1475 .1446 Note: R-sqrd = 1 - logL/Logl(constants) ---Chi-squared[ 5] = 328.91104 Prob [ chi squared > value ] = .00000 Response data are given as ind. choices Number of obs.= 1035, skipped 0 obs

---+---| Standard Prob. 95% Confidence

CHOIC| Coefficient Error z |z|>Z* Interval

---+---OS_X| -.00790 .57921 -.01 .9891 -1.14314 1.12734

---+---***, **, * ==> Significance at 1%, 5%, 10% level.

Fixed parameter ... is constrained to equal the value or had a nonpositive st.error because of an earlier problem.

(37)

---RPLOGIT;Lhs=CHOIC

;Rhs=one,OS_X,US_X,BAUM_X,WASSER_X,PREIS_X ;Fcn=OS_X(t),US_X(t),BAUM_X(n)

;Pds=9;Halton$

Random Parameters Multinom. Dependent variable

Log likelihood function

Logit Model CHOIC -893.63495 Restricted log likelihood -1137.06372 Chi squared [ 10](P= .000) 486.85754

Significance level .00000

McFadden Pseudo R-squared .2140854 Estimation based on N = 1035, K = 10 Inf.Cr.AIC = 1807.3 AIC/N = 1.746

---Log likelihood R-sqrd R2Adj No coefficients -1137.0637 .2141 .2103 Constants only -1115.3286 .1988 .1949 At start values -950.8731 .0602 .0556 Note: R-sqrd = 1 - logL/Logl(constants) ---Response data are given as ind. choices Replications for simulated probs. = 100 Used Halton sequences in simulations. RPL model with panel has 115 groups Fixed number of obsrvs./group= 9 Number of obs.= 1035, skipped 0 obs

---+---|Random parameters in utility functions...

OS_X| .12175 .2242D+07 .00 1.0000 *********** *********** US_X| *********** .09066 .2242D+07 .00 1.0000 *********** BAUM_X| .37150*** .06001 6.19 .0000 .25388 .48912

|Nonrandom parameters in utility functions...

WASSER_X| -.91313 .2242D+07 .00 1.0000 ***********

***********

PREIS_X| -.12925*** .03143 -4.11 .0000 -.19085 -.06765

A_INIT1| .15265 .12122 1.26 .2079 -.08493 .39023

A_INIT2| .43956*** .12604 3.49 .0005 .19253 .68658

|Distns. of RPs. Std.Devs or limits of triangular...

(38)

---Appendix D: NLOGIT Commands and Outputs

Model does not contain a full set of ASCs. R-sqrd is problematic. Use model setup with ;RHS=one to get LogL0.

CLOGIT;Lhs=CHOIC

;Rhs=one,OS_X,US_X,BAUM_X,WASSER_X,PREIS_X ;Pds=9;Halton$

---+---OS_X| -.00790 .57921 -.01 .9891 -1.14314 1.12734

---+---***, **, * ==> Significance at 1%, 5%, 10% level.

Fixed parameter ... is constrained to equal the value or had a nonpositive st.error because of an earlier problem.

---Discrete choice (multinomial logit) model

(39)

RPLOGIT;Lhs=CHOIC ;Choices=INIT1,INIT2,INIT3 ;Rhs=one,OS_X,US_X,BAUM_X,WASSER_X,PREIS_X ;Fcn=OS_X(t),US_X(t),BAUM_X(n) ;Pds=9;Halton$

OS_X| .12175 .2242D+07 .00 1.0000 *********** *********** US_X| *********** .09066 .2242D+07 .00 1.0000 *********** BAUM_X| .37150*** .06001 6.19 .0000 .25388 .48912

WASSER_X| -.91313 .2242D+07 .00 1.0000 ***********

***********

PREIS_X| -.12925*** .03143 -4.11 .0000 -.19085 -.06765

A_INIT1| .15265 .12122 1.26 .2079 -.08493 .39023

A_INIT2| .43956*** .12604 3.49 .0005 .19253 .68658

TsOS_X| 2.07029*** .26323 7.86 .0000 1.55437 2.58621 TsUS_X| 2.29135*** .26841 8.54 .0000 1.76528 2.81742 NsBAUM_X| .18270 .11595 1.58 .1151 -.04456 .40995 ---+---nnnnn.D-xx or D+xx => multiply by 10 to -xx or +xx. ***, **, * ==> Significance at 1%, 5%, 10% level. ---Random Parameters Multinom.

Dependent variable Log likelihood function

(40)

CLOGIT;Lhs=CHOIC

;Rhs=one,OS_X,US_X,BAUM_X,PREIS_X$

---Log likelihood R-sqrd R2Adj Constants only -1115.3286 .1475 .1450 Note: R-sqrd = 1 - logL/Logl(constants) ---Chi-squared[ 4] = 328.91104 Prob [ chi squared > value ] = .00000 Response data are given as ind. choices Number of obs.= 1035, skipped 0 obs

(41)

---CLOGIT;Lhs=CHOIC

;Rhs=one,OS_X,US_X,BAUM_X,PREIS_X

;Rh2=F18,F19,F20,F22,F24,F25,F29,F32,F33,F34,STUDENT,ERWERBST$

Log likelihood function -940.98950 Estimation based on N = 1035, K = 30 Inf.Cr.AIC = 1942.0 AIC/N =

1.876

---Log likelihood R-sqrd R2Adj Constants only -1115.3286 .1563 .1439 Note: R-sqrd = 1 - logL/Logl(constants) ---Chi-squared[28] = 348.67814 Prob [ chi squared > value ] = .00000 Response data are given as ind. choices Number of obs.= 1035, skipped 0 obs

(42)

---RPLOGIT;Lhs=CHOIC

;Rhs=one,OS_X,US_X,BAUM_X,PREIS_X ;Fcn=OS_X(t),US_X(t),BAUM_X(n) ;Pds=9;Halton$

McFadden Pseudo R-squared .2140854 Estimation based on N = 1035, K = 9 Inf.Cr.AIC = 1805.3 AIC/N =

1.744

--***, **, * ==> Significance at 1%, 5%, 10% level. ---OS_X| 1.03488*** .13181 7.85 .0000 .77654 1.29321 US_X| 1.00379*** .12634 7.95 .0000 .75618 1.25141 BAUM_X| .37150*** .06001 6.19 .0000 .25388 .48912

PREIS_X| -.12925*** .03143 -4.11 .0000 -.19085 -.06765

A_INIT1| .15265 .12122 1.26 .2079 -.08493 .39023

A_INIT2| .43956*** .12604 3.49 .0005 .19253 .68658

TsOS_X| 2.07029*** .26323 7.86 .0000 1.55437 2.58621

TsUS_X| 2.29135*** .26841 8.54 .0000 1.76528 2.81742

NsBAUM_X| .18270 .11595 1.58 .1151 -.04456 .40995

(43)

---+---x RPLOGIT;Lhs=CHOIC ;Choices=INIT1,INIT2,INIT3 ;Rhs=one,OS_X,US_X,BAUM_X,PREIS_X ;Rh2=F18,F19,F20,F22,F24,F25,F29,F32,F33,F34,STUDENT,ERWERBST ;Fcn=OS_X(t),US_X(t),BAUM_X(n) ;Pds=9;Halton$

OS_X| 1.06519*** .13861 7.68 .0000 .79351 1.33687

US_X| 1.07072*** .13656 7.84 .0000 .80307 1.33838

BAUM_X| .38082*** .06051 6.29 .0000 .26221 .49943

(44)

--***, **, * ==> Significance at 1%, 5%, 10% level. ---INI_F292| -.04574 .10623 -.43 .6667 -.25395 .16246 INI_F322| .04247 .12979 .33 .7435 -.21192 .29685 INI_F332| .20482 .20128 1.02 .3089 -.18969 .59932 INI_F342| -.11336* .06576 -1.72 .0847 -.24225 .01553 INI_STU2| .20199 .42717 .47 .6363 -.63525 1.03924 INI_ERW2| -.05858 .38531 -.15 .8792 -.81378 .69662

TsOS_X| 2.14343*** .27822 7.70 .0000 1.59813 2.68873

TsUS_X| 2.46204*** .28834 8.54 .0000 1.89691 3.02718

NsBAUM_X| .17426 .11750 1.48 .1381 -.05603 .40455

(45)

(46)

(8488) (7239) F20_2 .13082 -.05356 (4731) (8290) F22_2 .24604 .57754** (2542) (0399) F24_2 .01215* .01976** (0953) (0389) F25_2 -.20690 -.37973 (6151) (4791) F29_2 -.01893 -.04574 (8139) (6667) F32_2 .06789 .04247 (4910) (7435) F33_2 .11604 .20482 (4756) (3089) F34_2 -.04014 -.11336* (4182) (0847) STU_2 .07998 .20199 (8036) (6363) ERW_2 -.04958 -.05858 (8629) (8792) Log Likelihood -1115.3286 -1115.3286 -1115.3286 -1115.3286 R2 .1475 .1563 .1988 .2108 R2 Adj .1450 .1439 .1953 .1981 Pseudo R2 - - .2140854 .2259305

(47)

Appendix F: Abbreviations of Individual-specific Variables

Question No. Question Text Answer Options

F18_1 F18_2

Have you ever been directly affected by a flood, i.e. were your belongings damaged?

0 = No 1 = Yes

F19_1 F19_2

Have you ever been indirectly affected by a flood, i.e. have you experienced one or multiple of the following situations because of a flood? 0 = No 1 = Yes F20_1 F20_2

Have you ever gathered information on floods or flood protection? (Not in professional context)

0 = No 1 = Yes

F22_1 F22_2

Have you ever gathered information on floods and flood protection in the context of your professional activity?

0 = No 1 = Yes

F24_1 F24_2

How old are you? RATIO

F25_1 F25_2

Do you live in Germany? 0 = No 1 = Yes

F29_1 F29_2

What is your highest educational achievement?

ORDINAL

F32_1 F32_2

How many people live in your household, you included?

ORDINAL

F33_1 F33_2

How many of the people living in your household are children under the age of 18?

ORDINAL

F34_1 F34_2

How much was the average monthly net income of your household in the year 2017?

(48)

Appendix G: STATA WTP Results Tables

WTP in CLOGIT model

Variable Obs Mean Std. Err. [95% Conf. Interval]

C_OS_wtp 115 5.12385 0 5.12385 5.12385

C_US_wtp 115 4.96687 0 4.96687 4.96687

C_BAUM_wtp 115 2.36356 4.16e-17 2.36356 2.36356

WTP in RPLOGIT model

Variable Obs Mean Std. Err. [95% Conf. Interval]

RP_OS_wtp 115 7.823955 .4654561 6.90189 8.74602

RP_US_wtp 115 7.709211 .5449345 6.6297 8.788722

(49)

(50)

Author: Anna-Sophia Carmen Zietan

Master Thesis

Willingness to Pay for the Ecosystem.

Table of Contents

List of Figures

List of Tables

Abstract

1. Introduction

2. Motivation and State of Knowledge

2.1. Environment

2.2. Discrete Choice Experiment

3. Theory and Model

3.1. Theory

3.2. Model

4. Methods and Data

4.1. Definition of Attributes and Levels

4.2. Experimental Design

4.3. Questionnaire Development

4.4. Data Collection

5. Results

5.1. Descriptive Statistics

5.2. Econometric Analysis

6. Conclusion

7. References

Appendix A: Price Calculations

Appendix B: Opinion Poll

Appendix C:

NLOGIT Output for Models with WASSER

---Appendix D: NLOGIT Commands and Outputs

Appendix F: Abbreviations of Individual-specific Variables

Appendix G: STATA WTP Results Tables