Testing Choice Theory Using Discrete Choice Experiments in Swiss Energy Policy
Matteo Mattmann ii
Promoter: Prof. Dr. Roy Brouwer Co-Promoter: Dr. Ivana Logar
Thesis committee: Prof. Dr. Wouter Botzen Prof. Dr. Michael Getzner Dr. Jürgen Meyerhoff Prof. Dr. Ståle Navrud Prof. Dr. Rolf Wüstenhagen
Cover Design: Åsa Frölander
ISBN: 978-3-906327-95-2 iii
VRIJE UNIVERSITEIT
Testing Choice Theory Using Discrete Choice Experiments in Swiss Energy Policy
ACADEMISCH PROEFSCHRIFT
ter verkrijging van de graad Doctor aan de Vrije Universiteit Amsterdam, op gezag van de rector magnificus prof.dr. V. Subramaniam, in het openbaar te verdedigen ten overstaan van de promotiecommissie van de Faculteit der Bètawetenschappen op dinsdag 17 oktober 2017 om 11.45 uur in de aula van de universiteit, De Boelelaan 1105
door Matteo Mattmann geboren te Zürich, Zwitserland iv promotor: prof.dr. Roy Brouwer copromotor: dr. Ivana Logar
This PhD thesis was funded by the Swiss Federal Institute of Aquatic Science and Technology (Eawag), and is part of the Competence Center for Research in Energy, Society, and Transition (SCCER CREST). v
“Durchaus studiert, mit heißem Bemühn. Da steh ich nun, ich armer Tor! Und bin so klug als wie zuvor;”
J.W. von Goethe in Faust
vii
Summary
Testing Choice Theory Using Discrete Choice Experiments in Swiss Energy Policy
The "Swiss Energy Strategy 2050" proposes to phase-out nuclear power gen- eration and expand renewable sources of energy. Hydropower, an established source of energy in Switzerland, is expected to be one of the key renewables that will be further expanded. In this policy context, this PhD thesis aims to test axioms and assumptions underlying microeconomic choice theory by applying discrete choice experiments (DCE). A DCE is conducted among a representative sample of Swiss respondents and elicits their preferences for an expansion of hydropower. This dissertation contributes to the existing literature by examin- ing how public preferences for expanding hydropower production are linked to public perception of (avoiding) nuclear risks. To this end, hydropower as well as nuclear risks are included in the DCE. This thesis begins with a quantitative meta-analysis of the existing stated preference literature that estimates the non-market values of hydropower exter- nalities. The results of the meta-analysis are used as inputs in designing the DCE. The results of the meta-analysis suggest that deteriorations in vegetation, land- scape, and wildlife are valued negatively, while there is only limited evidence for a significant positive willingness-to-pay (WTP) for mitigating these negative externalities. The avoidance of greenhouse gas emissions proves to exert a sig- nificant positive influence on welfare estimates, but no significant impacts on aesthetic and recreational amenities can be detected. The meta-analysis also re- veals that no stated preference studies so far have considered the link between preferences for renewable sources of energy and nuclear risks. The data obtained from the DCE are used to answer this dissertation’s main research questions. These focus on the standard choice-theory assumptions of viii certain and known preferences and the axioms of continuity and monotonicity. Furthermore, the role of multiple reference points in the framework of prospect theory is investigated. More specifically, the common and idiosyncratic determinants of choice cer- tainty, consistency, and monotonicity are investigated. In contrast to the existing literature, these three concepts are analyzed simultaneously based on the same sample of respondents. The results show that there are significant differences between the choice behavior of certain and uncertain respondents as well as be- tween consistent and inconsistent respondents. Moreover, gender and choice- task complexity prove to be common predictors of choice certainty, consistency, and monotonicity. This thesis also investigates the standard economic axiom of continuous pref- erences in the context of attribute-non-attendance (ANA). A novel methodology to assess ANA is presented based on the monitoring of the respondents’ visual information acquisition process using mouse-tracking. No significant model im- provement is found when including such a visual measure of ANA compared with the standard approach based on stated ANA information. Nevertheless, choice models based on visual ANA result in a slight improvement over choice models that do not take ANA into account and over choice models that use in- ferred ANA information. Finally, the dependence of preferences on (multiple) reference points, a key assumption in prospect theory, is tested. Non-status quo related reference points, associated with comparative risks shown on risk ladders, are expected to affect parameter estimates and welfare measures for a change in hydropower and nu- clear risk. The study confirms the importance of multiple reference points, and shows that, besides the status quo, these other reference points also influence respondents’ choices and welfare measures in DCEs. The results of this thesis support the need for a holistic view on energy policy accounting for the direct and indirect externalities of alternative energy sources in both research and policy. ix
Acknowledgements
First and foremost I would like to thank my supervisors Ivana Logar and Roy Brouwer. It has been a very supportive, friendly, and uncomplicated collabora- tion with both of you. Your comments and ideas often perfectly complemented each other. Ivana, a special word of thanks to you for the infinite amount of time you have invested in giving me very valuable, extensive, and precise feedback on work in progress at various stages. Roy, special thanks to you for your enthu- siasm and for keeping the big picture in sight at times when no light at the end of the tunnel seemed visible. Thank you also for putting me up on various visits to Amsterdam and Waterloo in Canada, and for connecting me with the people in your respective teams. Many thanks also belong to the members of the reading committee for the review of this dissertation: Wouter Botzen, Michael Getzner, Jürgen Meyerhoff, Ståle Navrud, and Rolf Wüstenhagen. Special thanks are due to Mehmet Kut- luay, my connection to the Vrije Universiteit in Amsterdam and the team at IVM, for the motivating and valuable exchange. Thank you Noémie Neverre for the excellent French translation of my survey. I would also like to thank conference attendees in Nancy, Zurich, Cork, and Athens, as well as the organizers and members of the Competence Center for Research in Energy, Society and Transi- tion (SCCER CREST), of which this thesis is part of. This research project was funded by the Swiss Federal Institute of Aquatic Science and Technology (Eawag), and I would like to thank Eawag. A number of great people made the institute an interesting and enjoyable place to work. First of all, there are my team- and office-mates Paola and Markus. Thank you Paola for the humor you brought to our office and for your wise advice regarding finishing a PhD. Thank you Markus for the interesting and useful discussions. Thank you dear other members of ESS: Alex, Alice, Bernhard, Caroline, Fridolin, Jasmine, Mara, Mario, Mika, Mirella, Pauline, Philipp, Simon J., Simon M., and Ulrike. Last but not least, thank you Maja for all the food supply. Thank you.
xi
Contents
Summary vii
Acknowledgements ix
1 Introduction 1 1.1 Background ...... 1 1.2 Main objective, hypotheses, and research questions ...... 3 1.3 Data collection and econometric analysis ...... 6 1.4 PhD thesis outline ...... 8
2 Hydropower Externalities: A Meta-Analysis 11 2.1 Introduction ...... 11 2.2 Study selection and characteristics ...... 15 2.3 Meta-model ...... 19 2.3.1 Heterogeneity, heteroskedasticity, and non-independence 19 2.3.2 The meta-regression models ...... 20 2.4 Selection and definition of variables ...... 22 2.5 Results ...... 27 2.5.1 Descriptive statistics ...... 27 2.5.2 Meta-regression results ...... 29 2.5.3 Cross-validation ...... 33 2.6 Conclusions and discussions ...... 35 2.A Studies included in the meta-analysis ...... 38
3 Choice Certainty, Consistency, and Monotonicity 41 3.1 Introduction ...... 41 3.2 Choice certainty, consistency, and monotonicity in DCEs ..... 44 3.3 Econometric Models ...... 48 xii
3.4 Case-study description ...... 50 3.4.1 Discrete choice experiment ...... 50 3.4.2 Elicitation of choice certainty, consistency, and monotonicity 52 3.4.3 Sampling procedure and choice experiment design .... 53 3.5 Results ...... 55 3.5.1 Descriptive results ...... 55 3.5.2 Swait-Louviere test results ...... 57 3.5.3 Logit model results ...... 62 3.6 Discussion and conclusions ...... 66
4 Attribute non-Attendance in Discrete Choice Experiments 69 4.1 Introduction ...... 69 4.2 Attribute non-attendance ...... 71 4.3 Case-study description and experimental design ...... 77 4.4 Elicitation of stated and visual ANA ...... 79 4.5 Econometric models ...... 81 4.6 Results ...... 84 4.6.1 Descriptive statistics ...... 84 4.6.2 Stated ANA models ...... 86 4.6.3 Visual ANA models ...... 89 4.6.4 Inferred ANA models ...... 91 4.7 Discussion and conclusions ...... 94 4.A Example choice task ...... 97 4.B Lookup frequency and duration ...... 98 4.C AIC and BIC ...... 99 4.D Baseline ECLC model ...... 100
5 Reference Points for the Valuation of Risk Changes 101 5.1 Introduction ...... 101 5.2 Theoretical framework and hypotheses ...... 104 5.3 Case-study description ...... 108 5.3.1 Choice experiment ...... 108 5.3.2 Risk ladders ...... 109 5.3.3 Covariates ...... 112 5.3.4 Design generation and data collection ...... 113 xiii
5.4 Econometric models and testing procedures ...... 115 5.5 Results ...... 116 5.5.1 Descriptive statistics ...... 116 5.5.2 Choice model results ...... 119 5.5.3 Hypotheses test results ...... 123 5.6 Discussion and conclusions ...... 124 5.A Poe-test ...... 127
6 Conclusions 129 6.1 Summary of the main findings ...... 129 6.2 Directions for future research ...... 132 6.3 Policy recommendations ...... 135
A Choice Experiment Survey 137 A.1 Welcome ...... 137 A.2 Your electricity consumption ...... 137 A.3 Hydropower and nuclear power ...... 139 A.4 Your opinion about dam breaches ...... 139 A.5 Your opinion about nuclear accidents ...... 140 A.6 Risk graph ...... 141 A.7 Expansion of hydropower ...... 143 A.8 Your opinion in a public vote ...... 144 A.9 How hydropower can be expanded (I/II) ...... 144 A.10 How hydropower can be expanded (II/II) ...... 145 A.11 Example situation ...... 147 A.12 Eight hypothetical decision situations ...... 147 A.13 Decision 1 ...... 148 A.14 Decision 2 ...... 149 A.15 Decision 3 ...... 150 A.16 Decision 4 ...... 151 A.17 Decision 5 ...... 152 A.18 Decision 6 ...... 153 A.19 Decision 7 ...... 154 A.20 Decision 8 ...... 155 A.21 Background of your decisions ...... 156 xiv
A.22 Your leisure time ...... 157 A.23 Trust and attitude ...... 158 A.24 About your person (I/II) ...... 159 A.25 About your person (II/II) ...... 160 A.26 On this survey ...... 162
Bibliography 163 xv
List of Figures
2.1 Cross-validation histograms ...... 34
3.1 Choice task example ...... 52 3.2 Stated choice certainty ...... 56 3.3 Choice consistency ...... 57
4.1 Example choice task of the mouse-tracking setup ...... 80 4.2 Descriptive statistics for mean stated ANA ...... 85 4.3 Example choice task ...... 97 4.4 Average lookup frequency over choice tasks ...... 98 4.5 Average lookup duration over choice tasks ...... 98 4.6 AIC and BIC ...... 99
5.1 Schematic illustration of two different risk ladders ...... 105 5.2 Expected changes in utility associated with reference points ... 107 5.3 Choice task example ...... 110 5.4 Risk ladders shown to two samples ...... 111 5.5 Descriptive statistics of risk attitudes and perceptions ...... 117
xvii
List of Tables
2.1 Studies collected in the selection and search procedure ...... 17 2.2 Explanatory variables included in the meta-analysis ...... 23 2.3 Cross-tabulation of mean values of hydropower externalities ... 28 2.4 Meta-analysis regression models ...... 32
3.1 Studies that regress stated choice certainty on its determinants .. 45 3.2 Studies that regress choice consistency on its determinants .... 46 3.3 Attribute and attribute levels in the DCE ...... 51 3.4 Comparison of samples 1 and 2 ...... 55 3.5 Swait-Louviere test results for certain vs. uncertain and consistent vs. inconsistent respondents ...... 60 3.6 Swait-Louviere test results for equality of choice behavior between samples 1 and 2 ...... 60 3.7 Swait-Louviere test results for equality of choice behavior between samples that differ in the position of the repeated choice task ... 61 3.8 Logit regression results ...... 63
4.1 Attributes and attribute levels in the choice experiment ...... 78 4.2 Sociodemographic characteristics of the study sample and target population ...... 84 4.3 Visual ANA statistics ...... 86 4.4 Full attribute-attendance and stated ANA MXL models ...... 88 4.5 Visual ANA MXL models based on different thresholds of lookup frequency and duration ...... 90 4.6 Inferred ECLC ANA model ...... 92 4.7 Shares of respondents displaying stated, visual, and inferred ANA 93 4.8 Baseline ECLC model ...... 100 xviii
5.1 Attributes and attribute levels in the DCE ...... 109 5.2 Explanatory variables included in the choice models ...... 114 5.3 Sociodemographic characteristics of the study samples and target population ...... 117 5.4 Perceived controllability of the comparative endpoint and middle point risks ...... 118 5.5 Estimated mixed logit models ...... 119 5.6 MWTP estimates for the risk attributes ...... 124 xix
List of Abbreviations
AIC Akaike Information Criterion ANA Attribute Non-Attendance ASC Alternative-Specific Constant BIC Bayesian Information Criterion CV Contingent Valuation DCE Discrete Choice Experiment ECLC Equality-Constrained Latent Class FAA Full Attribute Attendance fMRI functional Magnetic Resonance Imaging GWyr Gigawatt-year HTCM Hypothetical Travel Cost Method IEA International Energy Agency LC Latent Class LR Likelihood Ratio MNL Multinomial Logit MWTP Marginal Willingness-To-Pay MXL Mixed Logit OECD Organization for Economic Co-operation and Development PPP Purchasing Power Parity RP Reference Point SFOE Swiss Federal Office of Energy SP Stated Preference SQ Status Quo TCM Travel Cost Method WTP Willingness-To-Pay
1
Chapter 1
Introduction
1.1 Background
Two trends are currently reshaping the world’s energy landscape: a shift from carbon-based to non-carbon-based sources of energy, and a diminishing role of nuclear power in global energy production. The first trend is observable in the fact that renewable energy sources (excluding hydropower) have experienced double-digit growth rates since 2005 and doubled their share in global power generation within the last 5 years to 6.7% in 2015 (PBL, 2016). Their share in global primary energy consumption has also doubled since 2010, reaching 2.8% in 2015. This is reflected in a decreasing global growth rate of CO2 emissions over the last 15 years: While the 5-year average values for global annual emissions increased by 3.1% between 2001 and 2005, they grew by 2.5% and 1.4% in the years 2006 to 2010 and 2011 to 2015, respectively. In 2015 and 2016, global CO2 emissions growth rates came to a standstill and emissions remained largely un- changed (IEA, 2017; PBL, 2016). This may signal a decoupling of CO2 emissions and economic activity, as the world economy grew steadily in both years, with GDP growth rates of 3.2% and 3.1% for 2015 and 2016, respectively (IEA, 2017;
IMF, 2017). Nevertheless, the regional differences in the CO2 emission trends are considerable. In 2015, the United States and China reduced their CO2 emissions by 2.6% and 0.7%, respectively, while the European Union and India increased their emissions by 1.3% and 5.1%, respectively. The second global trend in energy supply is a shift away from nuclear power. The share of nuclear power in global electricity production peaked in the mid- 1990s at around 20% and has been declining since, reaching 11% in 2015 (IAEA, 2 Chapter 1. Introduction
2015; PBL, 2016). Nevertheless, energy policies with regard to nuclear power diverge in different parts of the world: Germany, Switzerland, and Belgium have decided, with various degrees of legally binding decisions already taken, to phase-out nuclear power. At the same time, a number of countries are expand- ing nuclear electricity production substantially, most notably China, Russia, and South Korea. In response to these trends, and coupled with a weakened public acceptance of nuclear power after the nuclear accident in 2011 in Fukushima, Japan, Switzer- land revised its energy policy in a strategic long-term report – Energy Strategy 2050 (Prognos, 2012; SFOE, 2013). The key components of the Swiss Energy Strategy 2050 include: the expansion of renewable sources of energy including hydropower; an increase in energy efficiency; and the phasing-out of nuclear power. This last aspect has important implications for the country’s electricity supply, which has primarily been based on hydropower and nuclear power over the last decades. Hydropower and nuclear power produced, respectively, 59% and 33% of total electricity in 2016 (SFOE, 2017). Nuclear power is intended to be replaced with, among others, renewable energy sources. With respect to hy- dropower, the Energy Strategy 2050 foresees an increase in electricity production of ca. 9% compared with the current hydropower production (SFOE, 2012). However, the hydropower industry in Switzerland is in a difficult situation for a number of reasons. Most of the hydropower plants currently operate with losses, as a result of considerably lower prices and a reduced variation between peak and off-peak prices on electricity markets in the recent years. There are three main reasons for this: The increasing share of renewable electricity gener- ation; low global coal prices; and the low CO2 emission permit prices of the EU trading system (Barry et al., 2015). Although the environmental benefits of hy- dropower, such as low greenhouse gas emissions, are evident, the hydropower industry in Switzerland has a history of public dispute due to its environmental externalities. The negative effects of hydropower are mainly caused by a loss of connectivity between aquatic systems and altered flow regimes (hydropeak- ing). A loss of connectivity between water bodies affects the migration of animal species. Changes in flow regimes have an impact on wildlife, may endanger floodplains, and cause erosion. Another factor that hampers the industry is the 1.2. Main objective, hypotheses, and research questions 3 low public acceptance of the hydropower technology in Switzerland and else- where (Sternberg, 2008, 2010; Tabi and Wüstenhagen, 2017). This is likely to be associated with the above-mentioned negative environmental externalities of hydropower. Switzerland has the highest proportion (88%) of already developed hydro- power potential worldwide (IEA, 2010). In addition, considering the economic, environmental, and social obstacles that the hydropower industry in Switzer- land currently faces, the envisaged expansion of hydropower production by 9% from its current level is challenging. The Swiss Federal Office of Energy hence concludes that such an expansion is only possible under a considerable improve- ment in economic conditions and public acceptance (SFOE, 2012).
1.2 Main objective, hypotheses, and research ques- tions
This PhD thesis focuses on the public acceptance of hydropower expansion. It does so by investigating the values that the general public attaches to the vari- ous positive and negative externalities related to hydropower. These values are typically elicited using stated preference research methods, that is, research in which the public at large is asked to answer questions about the externalities involved, using different forms of survey methods. In the past two decades, discrete choice experiments (DCE) have become the most important stated pref- erence (SP) elicitation approach (Johnston et al., 2017). The general objective of this PhD thesis is to quantify and explain variation in the non-market values at- tached to hydropower externalities. More specifically, it aims to test axioms and assumptions of microeconomic choice theory in the context of DCEs. A quan- titative meta-analysis summarizes and synthesizes the results of the existing SP literature on hydropower externalities, and generates new insights that are used as inputs in the design of a DCE in this thesis. The choice survey that is subse- quently designed and implemented consists of four versions which differ with respect to their methodological characteristics. Four independent samples of re- spondents answered the different versions of the DCE. This setup allows this dissertation to investigate the choice behavior of survey participants in a DCE and, in particular, it enables it to gain new insights into choice certainty, choice 4 Chapter 1. Introduction consistency, and choice monotonicity, as well as the continuity of preferences and their dependence on reference points. All of these concepts are linked to the axioms and assumptions of choice theory. This dissertation is embedded in the context of Swiss energy policy, as the DCE elicits public preferences and willingness-to-pay (WTP) values for an expansion of electricity production by hydropower in Switzerland. In the first part of this dissertation, a meta-analysis is conducted by esti- mating meta-regression models using the estimated non-market values for hy- dropower externalities. These values are elicited from existing SP studies im- plemented in different parts of the world. This is the first meta-analysis in the non-market valuation literature that explicitly focuses on hydropower and its external effects. Next, a DCE is conducted as part of an online survey. The policy context de- scribed in the previous section serves as a case study for the implementation of the DCE. Research on the external effects of renewable energy typically investi- gates either direct externalities, e.g. the effects on wildlife, vegetation, and land- scape, or indirect externalities, e.g. greenhouse gas emissions. Greenhouse gas emissions constitute an indirect externality, because they are not directly caused by renewable energy sources but represent avoided external effects of conven- tional sources of electricity. In the Swiss energy context, one of the major indirect externalities of renewable energy sources is the avoidance of nuclear risk. Until now, the effect of avoiding nuclear risk on public risk perception and preferences for renewable energy has not been studied in the valuation literature. Hence, this PhD thesis contributes to the existing literature by its central hypothesis, which postulates that public preferences for expanding hydropower production are linked to public preferences for avoiding nuclear risk. To this end, a choice attribute related to nuclear risk is included in the DCE. Hartmann et al. (2013) is the only study that explicitly investigates the relationship between public at- titudes and perceptions of nuclear power and preferences for the adoption of green electricity. However, in contrast to Hartmann et al. (2013), this thesis fo- cuses on this relationship based on stated preference research. The main objective of the DCE is to investigate a number of axioms and com- mon assumptions made in consumer choice theory (e.g. Jehle and Reny, 2001; Mas-Colell and Whinston, 1995). The underlying theoretical framework of DCEs 1.2. Main objective, hypotheses, and research questions 5 is provided by Lancaster’s theory of consumer demand (Lancaster, 1966), and by random utility theory (McFadden, 1974; Thurstone, 1927). Lancaster’s theory of demand states that utility is not derived from goods directly, but from their indi- vidual characteristics. Random utility theory has its roots in the rapidly increas- ing availability of survey data in the 1960s, which created the need for linking ob- served behavior to existing microeconomic consumer theory (McFadden, 2001). Random utility theory assumes that utility functions exist, and that respondents choose in accordance with their utility functions. Therefore, the standard eco- nomic axioms and assumptions of consumer theory as outlined, for example, in Jehle and Reny (2001), are maintained in the random utility framework. This PhD thesis tests some of these axioms and assumptions. Additionally, it tests an important assumption about choice behavior in Kahneman and Tversky’s (1979) prospect theory. First, the standard economic axiom of monotonicity and the conventional assumptions of stable and known preferences are investigated. This is accom- plished by testing the null hypothesis that preferences are known, consistent, and monotonic, and by examining the determinants of choice certainty, choice consistency, and choice monotonicity and their impact on choice behavior. The contribution of this PhD thesis to the existing literature is that this thesis inves- tigates choice certainty, consistency, and monotonicity simultaneously, using the same choice responses. This allows for the identification of both common and idiosyncratic drivers of these constructs. Second, the axiom of continuous preference relations is tested by analyzing attribute non-attendance (ANA). Continuous preference relations imply contin- uous indifference curves, and hence assume that choice participants adopt com- pensatory decision-making rules when making choices in a DCE (e.g. Lagarde, 2010). The existence of ANA violates the continuity axiom. This thesis assesses how a novel methodology to capture visual ANA can contribute to a better un- derstanding of ANA behavior. Specifically, the common approach for analyzing ANA in the existing literature through stated or inferred ANA information is extended with a novel, visual approach for capturing ANA behavior based on mouse-tracking. Finally, this dissertation studies the dependence of preferences on reference points, which is an important assumption of choice behavior in prospect theory. 6 Chapter 1. Introduction
In contrast to large parts of the valuation literature, the original text of Kahne- man and Tversky (1979) states a variety of possible reference points that do not have to coincide with an individual’s status quo. Drawing on prospect theory, this PhD thesis examines the possibility that comparative risks displayed on a risk ladder may serve as reference points and have an impact on an individual’s choice in a DCE. The majority of the DCE literature on reference points focuses on reference points that are linked to the characteristics of the choice tasks (e.g. their baseline levels). In contrast, the last chapter of this thesis contributes to the DCE literature by exploring the role of reference points which are induced independently and prior to the actual choice tasks. The main research questions addressed in the thesis can be summarized as follows:
1. What are the main determinants of the non-market values for hydropower externalities?
2. What are the common and idiosyncratic determinants of choice certainty, choice consistency, and choice monotonicity in DCEs, and what is the role of choice complexity?
3. How does visual ANA data obtained from mouse-tracking perform in ex- plaining ANA behavior compared with stated and inferred ANA?
4. Do comparative risks shown on risk ladders serve as reference points and influence preferences for a change in risk?
1.3 Data collection and econometric analysis
Two data collection processes took place for the purpose of this PhD thesis. First, a database was created based on secondary data derived from existing SP stud- ies and publications for the meta-analysis, and second, a survey including a DCE was conducted to collect primary research data. A database of existing research that values the external effects of hydropower was constructed in order to iden- tify the main determinants of the economic values for hydropower externalities (research question 1). The created database consists of 29 international stud- ies, which together generate 81 observations. Three different meta-regression 1.3. Data collection and econometric analysis 7 models are applied. A baseline model is estimated using weighted least squares regression analysis. The observations are weighted by the sample size of the sur- veys in the original studies. This procedure controls for differences in the vari- ances of the values in the database by assuming that variances are smaller for observations that are obtained from surveys with larger sample sizes. Two other models control for systematic differences in mean welfare estimates between the studies, and for differences in the influence of the regressors on the dependent variable. The DCE that follows the global meta-analysis elicits public preferences for an expansion of hydropower in Switzerland specifically, and aims to generate data on public preferences for hydropower expansion. This serves to answer the methodological research questions 2 to 4. The DCE was implemented among 1,000 households that constituted a representative sample of the German- and French-speaking Swiss population (roughly 95% of the total population of Switzer- land, the remaining 5% live in an Italian-speaking region). Survey pretesting included 20 face-to-face interviews and two rounds of online pretests with 220 and 350 respondents. For the final DCE, the respondents were split into four dif- ferent, independently recruited representative samples, each comprising ca. 250 households. Each sample received a slightly different questionnaire version in order to be able to answer the different research questions. Compared with the baseline version, the three other versions differed with respect to: the presence of questions on choice certainty; the monitoring of the respondents’ information ac- quisition process by mouse-tracking; and the reference points included in a risk ladder representing the changes in hydropower and nuclear power risk under valuation. Different econometric techniques are applied. For the purpose of answer- ing question 2, binary logit models and random-effects ordered logit models are estimated. Binary logit models are used to regress choice consistency and choice monotonicity on possible explanatory variables, and random-effects or- dered logit models are employed to identify drivers underlying stated choice certainty. To answer research questions 3 and 4, mixed logit (MXL) models are estimated. In contrast to the fixed effects multinomial logit model, MXL models allow for random taste heterogeneity across individuals and correlation between unobserved factors over alternatives and choice tasks. Question 3 is addressed 8 Chapter 1. Introduction by running MXL models with attribute parameters for respondents who state non-attendance to an attribute restricted to zero. The same procedure is applied for analyzing visual ANA information. Inferred ANA is assessed using equality- constrained latent class (ECLC) models. Each class in this model describes a spe- cific pre-defined pattern of ANA behavior. MXL models are also used in order to identify the effects of different risk ladders on choice behavior (research question 4). The Swait and Louviere (1993) test procedure is applied in the course of an- swering the research questions 2 and 4. For research question 2, the tests assess whether there are statistically significant differences between the choice behav- ior of respondents who are (un)certain about and (in)consistent in their choices. Furthermore, the effect of including follow-up questions on choice certainty and including the same choice task in a different position in the choice-task sequence is examined using the same procedure. In answering research question 4, the Swait-Louviere test is applied to compare two split-samples of respondents who were shown risk ladders that differ with respect to the ranges of probabilities of comparative risks.
1.4 PhD thesis outline
Chapter 2 aims to answer research question 1. This study was first presented at the 22nd Annual Conference of the European Association of Environmental and Resource Economists in Zurich in June 2016, and has been published as Mattmann, Logar, and Brouwer (2016a) in Energy Economics. It presents a meta- analysis of existing SP research on the economic value of the positive and neg- ative external effects of hydropower. For this purpose, a database with the eco- nomic values of the non-market impacts of hydropower electricity generation is constructed. The main aim of the meta-analysis is to quantify and explain the economic values for positive and negative hydropower externalities. Different meta-regression model specifications are estimated to test the robustness of the determinants of these non-market values. The impact of key methodological features of the valuation studies on the results is also investigated. Chapter 3 attempts to answer research question 2. It focuses on the consumer theory axiom of monotonicity and the assumptions that consumer preferences 1.4. PhD thesis outline 9 are known and stable. More specifically, Chapter 3 tests whether choices are based on known, stable, and monotonic preferences, and investigates the com- mon and idiosyncratic determinants of choice certainty, consistency, and mono- tonicity based on the results of the DCE. For this purpose, choice certainty, con- sistency, and monotonicity are regressed on possible drivers. In doing so, two different measures of choice task complexity are compared: The entropy of a choice task, and the utility difference between the alternative that is chosen and the second-best alternative. Moreover, this chapter tests the equality of choice behavior of respondents who differ with respect to choice certainty and consis- tency. It also investigates the effect of including choice certainty follow-up ques- tions after each choice task, and compares the choice behavior of respondents who were shown a repeated choice task in a different position in the choice task sequence. Chapter 4 answers research question 3 and investigates the standard economic axiom of continuity. This chapter was first presented at the 23rd Annual Confer- ence of the European Association of Environmental and Resource Economists in Athens in June 2017. It presents the first application of mouse-tracking to analyze ANA in DCEs. Mouse-tracking is applied to record the frequency and duration of uncovering attribute information in the choice process. Mouse-tracking func- tions similarly to eye-tracking, but can be applied online and allows for a larger sample size. The information obtained from mouse-tracking is used to generate a visual definition of ANA, while stated ANA information is collected by means of a follow-up question after the DCE. The performance of choice models based on stated, inferred, and visual ANA information is compared. Chapter 5 focuses on research question 4 and assesses a key assumption of prospect theory: the dependence of preferences on reference points. This chap- ter argues for the existence of multiple reference points. The DCE values the changes in the risk of dying caused by a hydropower and a nuclear power acci- dent. Risk ladders are used to communicate the risk information to respondents. Two different risk ladders are presented to two independent samples of respon- dents. The risk ladders differ with respect to the range of risk probabilities that serve as a benchmark for the risks being valued. One sample is shown a risk ladder with a high reference point, that is, a risk ladder with a wide range of comparative risk probabilities that include high risk events, whereas the other 10 Chapter 1. Introduction sample is shown a risk ladder with a low reference point, i.e. a risk ladder with a narrow range of comparative risk probabilities that encompass lower risks. On both risk ladders, the change in risk that is being valued is identical for both sam- ples. Chapter 5 hypothesizes that, in addition to the status quo probability of the valued risks, comparative risks presented on the risk ladders represent reference points that influence the valuation of risk changes. Chapter 6 discusses the results that are presented in the Chapters 2, 3, 4, and 5 and concludes. This chapter also identifies the need for further future research, and summarizes policy-relevant insights. 11
Chapter 2
Hydropower Externalities: A Meta-Analysis1
2.1 Introduction
As a result of both, increasing efforts to decarbonize economies and substan- tially diminished social and political acceptance of nuclear energy production following the 2011 accident in Fukushima, Japan, renewable energy sources are set to play a more prominent role worldwide. This is reflected in various na- tional energy policies. Germany and Switzerland, for example, decided to phase out nuclear energy production and replace its share in national electricity pro- duction primarily with renewable energy sources (SFOE, 2013). Renewable en- ergy sources avoid many negative externalities of conventional energy produc- tion based on fossil or nuclear fuels, which typically involve long-term conse- quences such as the impacts of greenhouse gas emission on climate change or the accumulation of radioactive waste. However, renewable sources of energy often operate with lower energy densities than non-renewable energy carriers, which results in spatially larger production facilities (Wüstenhagen, Wolsink, and Bürer, 2007). As a consequence, other types of externalities such as threats to biodiversity or aesthetic impacts occur.
1This chapter is published as: Mattmann, Matteo, Ivana Logar, and Roy Brouwer (2016). "Hy- dropower Externalities: A Meta-Analysis". Energy Economics 57, pp.66-77. It was also presented at the 22nd Annual Conference of the European Association of Environmental and Resource Economists in Zurich in June 2016. 12 Chapter 2. Hydropower Externalities: A Meta-Analysis
Much of the existing research related to the economic valuation of renew- able energy focuses on the newer technologies of wind, solar, biomass and bio- fuel. Recent examples include studies which value externalities from: wind power generation (Alvarez-Farizo and Hanley, 2002; Ek, 2006; Ek and Pers- son, 2014; Ladenburg and Dubgaard, 2007); biomass (Susaeta et al., 2011); or a mixture of various renewable energy sources (Bergmann, Colombo, and Han- ley, 2008; Bergmann, Hanley, and Wright, 2006; Komarek, Lupi, and Kaplowitz, 2011; Kosenius and Ollikainen, 2013; Ku and Yoo, 2010; Longo, Markandya, and Petrucci, 2008). In contrast, the amount of research that has been conducted on the effects and economic values of more established technologies such as hy- dropower is rather limited. Since the role of hydropower as a source of renew- able energy is expected to expand further worldwide (e.g. Jacobson and Deluc- chi, 2009), an understanding of individuals’ preferences for its effects on the en- vironment, recreational activities, and aesthetic values is of crucial importance to inform an effective and efficient energy transition. Hydropower is a renewable source of energy with a long history (Paish, 2002). The product of hydropower generation is electricity, a standard market good that can be sold directly to electricity consumers, and it is therefore usually not considered in valuation studies. The same holds for the employment effects of hydropower operations. However, hydropower electricity production typi- cally generates a number of positive and negative side effects that affect different groups of stakeholders, for which they are, in most cases, not (directly) compen- sated. These effects of hydropower depend not only on the size of operation and the geographical location, but also on the type of hydropower facility. That is, run-of-the-river facilities, usually operating with constant water flows and gen- erating electric base load, have different effects than storage plants that depend on dams to store water, which is released at times of peak demand. The effects of storage plants with natural water feeding can differ again from the effects of pumped-storage plants that pump water from a lower to a higher reservoir. In general, most of the external effects of hydropower are caused by hydropeaking and disconnected water bodies. Reduced connectivity refers to the disconnec- tion of water bodies caused by hydropower dams and run-of-the-river facilities. Changes in flow (hydropeaking) occur only in the case of storage hydropower plants. Hydropeaking causes non-natural flow patterns, i.e. high variability in 2.1. Introduction 13 discharge, water levels, and flow velocity of water bodies. The various effects caused by different types of hydropower plants will be briefly summarized be- low. Recreation is an important service provided by aquatic ecosystems (Boyd and Banzhaf, 2007), which may be impaired by hydropower. Examples of such ser- vices affected by hydropower operations include various types of recreational activities such as kayaking or rafting (Aravena, Hutchinson, and Longo, 2012; Hynes and Hanley, 2006), fishing (Filippini, Buchli, and Banfi, 2003; Gogniat, 2011; Håkansson, 2009; Loomis, Sorg, and Donnelly, 1986; Navrud, 2004; Rob- bins and Lewis, 2009) or visiting waterfalls (Ehrlich and Reimann, 2010). Most studies observe that these recreational activities are negatively influenced by hy- dropower due to hydropeaking and the disconnectivity of water bodies, both of which impede water sports and endanger fish populations, thereby reducing the value of angling. It is, however, conceivable that hydropower may also generate positive effects on recreational opportunities: for example, by creating artificial lakes suitable for water sports. Getzner (2015) empirically compares the recre- ational value of free-flowing sections of a river with dammed stretches and finds higher recreational benefits on free-flowing sections than on dammed stretches of rivers for a variety of recreational activities. The environmental effects of hydropower are manifold. A positive environ- mental externality of hydropower electricity production is lower greenhouse gas emission compared with most other sources of electricity production (see Weisser (2007) for a literature overview of greenhouse gas emissions by differ- ent electricity production technologies). The reduction in the emission of green- house gases depends, however, on reservoir size and type, the extent of flooded vegetation, soil type, water depth, and climate conditions. Especially methane emission can form a significant source of greenhouse gas release in the case of the hydropower reservoirs of storage plants in tropical regions (e.g. Barros et al., 2011; Delsontro et al., 2010). Pumped-storage plants without natural water feed are used for load balancing only, and do not directly reduce greenhouse gas emissions since they consume more electricity than they generate. Negative environmental externalities of hydropower also stem from either re- duced connectivity of aquatic systems or altered flow regimes. Reduced connec- tivity especially affects the migration of fish and other animal species. Changes 14 Chapter 2. Hydropower Externalities: A Meta-Analysis in flow patterns (hydropeaking) change sedimentation levels, and can lead to rapid changes in water temperature. Both of these effects have an impact on invertebrates, which are usually very sensitive to altered temperature and sed- iments (e.g. Bruno et al., 2009). In addition, non-natural hydropower flow pat- terns may endanger floodplains, threaten fish and bird species, and cause ero- sion. Hydropower projects, especially the construction of dams, artificial lakes and reservoirs, may also affect artifacts of important cultural, historical and geolog- ical value that are flooded during the construction phase of hydropower stor- age plants (Han, Kwak, and Yoo, 2008; Lienhoop and MacMillan, 2007; Navrud, 2004). Direct, potentially negative, aesthetic impacts of hydropower often stem from hydropower-related facilities such as dams, access tracks, pipelines, build- ings, and the lack of vegetation due to these installations (Hanley and Nevin, 1999). Run-of-the-river plants cause aesthetic degradation as well. It has been shown that free-flowing rivers have higher aesthetic value compared with rivers affected by hydropower facilities (Born et al., 1998). Furthermore, pylons con- necting remote hydropower plants might adversely affect views and scenery (Aravena, Hutchinson, and Longo, 2012). The main objective of this paper is to synthesize the empirical evidence on the economic valuation of hydropower externalities in a meta-analysis. In contrast to a recent meta-analysis on the willingness-to-pay for green electricity (Sundt and Rehdanz, 2015), we focus explicitly on hydropower and its externalities. This is, to our knowledge, the first study to conduct such an analysis. A main research question addressed in this paper is whether the positive hydropower external- ities outweigh the negative ones. The purpose of the meta-analysis is not only to review and evaluate the existing literature, but also to explain study-to-study variation by focusing on differences between valuations for various positive and negative types of hydropower externalities, as well as on key methodological characteristics such as sensitivity to scope. In order to do this, the external effects of hydropower production are first identified and classified. Next, the drivers of welfare estimates for the non-market effects of hydroelectric production technol- ogy are examined in a meta-regression model. The remainder of this paper is structured as follows. Section 2.2 describes the search procedure and selection of studies included in the meta-analysis. Section 2.2. Study selection and characteristics 15
2.3 explains the main econometric issues in meta-modeling and the estimated models. Section 2.4 considers the factors that influence the economic values of hydropower externalities. The results of the estimated meta-regression models are presented in Section 2.5 followed by conclusions in Section 2.6.
2.2 Study selection and characteristics
The non-market valuation of the externalities of hydropower production consti- tuted the main criterion for a study to be included in the meta-analysis. More specifically, all studies that generated primary valuation data of the non-market impacts of electricity production by hydropower were considered for inclusion. We included all studies in which hydropower production was identified as a source of the externalities. This involves studies that valued the externalities of hydropower exclusively (roughly 80% of all observations), as well as studies which value the external effects of renewable energy in general, but explicitly mention hydropower to be one of these (20% of the observations included). For example, a study that values increased water flows due to modified hydropower operation schemes would be included in the analysis, whereas a study that esti- mates the value of increased water flows without explicitly specifying that these changes in water flows are caused by hydropower operation would not be in- cluded. Applying this selection criterion ensured that individuals took their preferences for hydropower into account when valuing the external effects. The search procedure was conducted in 2014. Online databases that were browsed included Google Scholar, Scopus, Econlit and RePEc. ProQuest was used to search specifically for relevant PhD theses. The search included both published and unpublished papers, working papers, conference papers, PhD theses, Master’s theses, government and non-government reports. Keywords that were used in the search process included, among others, the following terms and combinations thereof: hydropower, hydroelectric, stated preferences, re- vealed preferences, contingent valuation, conjoint analysis, choice experiment, travel cost, hedonic pricing, externalities, dams, and recreational benefits. Table 2.1 provides a list of the studies included in the meta-analysis collected by the search and selection procedures described above. Most of the studies ob- tained are articles published in international peer-reviewed journals, but there 16 Chapter 2. Hydropower Externalities: A Meta-Analysis are also two reports, two working papers, one conference paper, a PhD the- sis, and two Master’s theses. Three reports could not be obtained despite an extensive search procedure. Other studies that were excluded to avoid double counting analyzed data that had already been used in one or more other relevant publication. Five papers valued the externalities of renewable energy in general without explicitly mentioning hydropower, and thus the economic values of the effects could not be ascribed to hydropower. Furthermore, two publications re- ported only aggregated economic values for the relevant population that could not be transformed to individual welfare estimates. The earliest study was carried out in 1983, while the other studies were con- ducted over a period of 18 years between 1993 and 2011. The majority of the studies were carried out in Europe (70%), followed by South America (13%), the United States (9%), and Asia (9%). With respect to the valuation methods, most studies applied stated preference methods, such as contingent valuation (CV) or discrete choice experiments (DCE); two studies used revealed preference meth- ods (travel cost method (TCM)); and three combined revealed and stated pref- erence approaches, using the hypothetical TCM (HTCM). Out of a total of 29 studies, 81 observations could be used in the subsequent meta-analysis. 15 stud- ies contributed only one observation. Studies provided more than one observa- tion when using different samples of respondents (for example, distinguishing between users and non-users of a resource) or because they valued various com- binations of hydropower externalities. A few studies also varied the method- ological aspects in split samples. The number of respondents underlying each observation varies considerably (between 45 and 1933), with an average of 361 respondents per observation. Eight observations (9.9%) included respondents who were directly affected by hydropower externalities. These are, for exam- ple, anglers, who were asked to value the number of fish in a river affected by hydropower. Peer-reviewed papers included in the analysis received, on aver- age, 39 citations measured by the Google Scholar citation index, with one study having a maximum of 136 citations (up to December 2014). Finally, the share of hydropower in total national electricity production (in the year of the survey) was included as a measure for the energy mix in a country (IEA, 2014a,b). Na- tional shares of hydropower vary widely, with an average of 38% of electricity produced by hydropower in the countries where the surveys were conducted. 2.2. Study selection and characteristics 17
TABLE 2.1: Studies collected in the selection and search proce- dure (ordered by study year)
# Study Authors Type of publication Country Nat. Valuation Nc year (years of hydro. Methodb publication) sharea
1 1983 Loomis, Sorg, Journal article (Journal USA 13.7% CV 1 and Donnelly of Environmental (1986) Management) 2 1993 Kosz (1996) Journal article AUT 71.5% CV 1 (Ecological Economics) 3 1993 Navrud Report & Journal NOR 99.6% CV 2 (1995, 2001) article (Hydropower and Dams) 4 1994 Biro (1998) Journal article (Ambio) TUR 39.1% CV 1 5 1996 Loomis (1996) Journal article (Water USA 9.6% CV 3 Resources Research) 6 1997 Hansesveen Master’s Thesis NOR 99.3% CV 3 and Helgas (1997) 7 1998 Bergland Report NOR 99.4% CV 3 (1998) 8 1998 Filippini, Journal article CHE 53.7% HTCM 1 Buchli, and (Applied Economics) Banfi (2003) 9 1998 Hanley and Journal article (Energy GBR 1.4% CV 1 Nevin (1999) Policy) 10 1998 Loomis (2002) Journal article (Water USA 7.8% HTCM 1 Resources Research) 11 2002 Han, Kwak, Journal article KOR 1.0% DCE 1 and Yoo (Environmental Impact (2008) Assessment Review) 12 2002 Sundqvist Doctoral Thesis SWE 45.2% DCE 1 (2002) 13 2003 Bothe (2003) Working Paper ISL 83.4% CV 1 14 2003 Hynes and Journal article (Land IRL 2.4% TCM 1 Hanley (2006) Use Policy) 15 2003 Bergmann, Journal article GBR 0.8% DCE 6 Colombo, (Ecological Economics) and Hanley (2008) 16 2004 Håkansson Journal article (Journal SWE 39.6% CV 8 (2009) of Environmental Planning and Management) 17 2004 Navrud Report NOR 98.8% CV 1 (2004) 18 Chapter 2. Hydropower Externalities: A Meta-Analysis
# Study Authors Type of publication Country Nat. Valuation Nc year (years of hydro. Methodb publication) sharea
18 2005 Longo, Journal article GBR 40.5% DCE 4 Markandya, (Ecological Economics) and Petrucci (2008) 19 2006 Kataria (2009) Journal article (Energy SWE 43.1% DCE 7 Economics) 20 2006 Robbins and Journal article (Journal USA 6.8% TCM 2 Lewis (2009) of the American Water Resources Association) 21 2006 Ku and Yoo Journal article KOR 0.9% DCE 3 (2010) (Renewable and Sustainable Energy Reviews) 22 2008 Aravena, Journal article Energy CHL 40.5% CV 1 Hutchinson, Economics) and Longo (2012) 23 2008 Ponce et al. Journal article (Water CHL 40.5% CV 10 (2011) Resources Management) 24 2008 Kosenius and Journal article (Energy FIN 22.1% DCE 1 Ollikainen Policy) (2013) 25 2009 Ehrlich and Journal article EST 0.4% CV 1 Reimann International Journal (2010) of Geology) 26 2010 Guo et al. Journal article (Energy CHN 17.2% CV 2 (2014) Policy) 27 2011 Gogniat Master’s Thesis CHE 51.5% HTCM 1 (2011) 28 2011 Klinglmair Conference Paper AUT 55.0% DCE 3 and Bliem (2013) 29 2011 Klinglmair, Working Paper AUT 55.0% DCE 10 Bliem, and Brouwer (2012) Notes: a IEA (2014a,b). b CV: Contingent Valuation; CE: Choice Experiment; HTCM: Hypothetical Travel Cost Method; TCM: Travel Cost Method. c Number of observations included in the meta-analysis. 2.3. Meta-model 19
2.3 Meta-model
2.3.1 Treatment of heterogeneity, heteroskedasticity, and non- independence
Meta-regression models can be classified according to the way they address and treat data heterogeneity, heteroskedasticity of effect-size variances, and non- in- dependence of observations from the same studies (Nelson and Kennedy, 2008). This section explains these three issues and how they are tackled in our study. Data used in a meta-analysis come from a variety of papers, authors, and countries. Furthermore, there are often individual-specific differences between survey participants, and the effect-size that forms the dependent variable in a meta-analysis might suffer from inconsistencies between studies (Smith and Pat- tanayak, 2002). In other words, studies may differ with respect to their design el- ements, but they may also differ regarding their examined effect-size (Ringquist, 2013). Apart from enhancing the comparability of effect-sizes by adjusting avail- able data from primary studies and dropping observations that lack comparabil- ity, the standard treatment of data heterogeneity in economic studies is to con- trol for differences in effect-size by including independent variables (Nelson and Kennedy, 2008; Smith and Pattanayak, 2002). In this study, control will be in- cluded for the differences between the types of hydropower externalities valued, the sample characteristics, and the methodological features of different studies. The primary studies used in meta-analysis usually do not have the same (es- timated) variances owing to differences in study-specific characteristics (Nelson and Kennedy, 2008). The standard assumption of the ordinary least squares (OLS) estimator of homogeneity is thus in general violated (Ringquist, 2013). In order to mitigate heteroskedasticity of effect-size variances and to control for differences in the quality of study results, the observations are ideally weighted by the inverse of their variances, resulting in weighted least squares regression (e.g. Lipsey and Wilson, 2001). By applying weights in this manner, more ac- curate studies with lower variances receive higher weights in the meta-analysis. Since, in this study, we only have information available about estimated vari- ances of a fraction of the primary studies, we weight the individual observations by the square root of the study sample sizes, as is commonly done in the meta- regression literature (see Nelson and Kennedy (2008) for an overview on stud- ies which apply this procedure). This ensures that studies with larger sample sizes (and therefore, as expected, also lower variances) receive more weight in 20 Chapter 2. Hydropower Externalities: A Meta-Analysis the analysis. As a consequence, the issue of heteroskedasticity is mitigated, and we ensure that the observations which we consider to be more reliable receive higher weights in the analysis. It is common procedure in meta-analysis to draw several effect-sizes from each study. Since observations drawn from the same study usually share some common characteristics, it must be assumed that there is within-study correla- tion between observations (Nelson and Kennedy, 2008). Various procedures ex- ist to mitigate this issue, such as including only one observation per study, or including only mean values of various observations from the same study. How- ever, since the number of primary studies, and hence observations that are used in a meta-analysis, may be limited, it is in many cases unavoidable to use all the observations obtainable from each study. Furthermore, the use of several observations from the same study provides some estimation leverage because many elements of the research design of these observations remain the same (Ringquist, 2013). If various observations per study are used, it is necessary to control for within-study correlation by explicitly taking the hierarchical data structure into account. This can be done, for example, by using panel data mod- els or calculating cluster-robust standard errors (Nelson and Kennedy, 2008). Both approaches are applied in this study.
2.3.2 The meta-regression models
We apply a variety of different approaches to address the issues described in Sec- tion 2.3.1, resulting in three different models. In Model 1 we use cluster-robust standard errors, where studies are set as clusters. This enables us to take the cor- relation between value estimates from the same studies into account. Cluster- robust standard errors assume independent observations across, but not within, clusters. Model 2 is a random-effects panel model with individual studies de- fined as cross-sectional units. Model 3 is an extension of the random-effects model that allows not only intercept coefficients, but also slope parameters to be random (Cameron and Trivedi, 2005). The baseline model (Model 1) is estimated by weighted least squares, and is specified as follows (e.g. Harbord and Higgins, 2008):