1 THE DETERMINANT OF US CONSUMER S ATTITUDES TOWARD SOLAR ENERGY By CHAO LIN LU A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2016
2 2016 ChaoLin Lu
3 To my parents and my wife
4 ACKNOWLEDGEMENTS I would like to express my deepest gratitude to my committee members, Dr. Zhifeng Gao, Dr. Lisa A. House, Dr. Glcan nel and Dr. Matt Hersom for their generous help and support in completion of this dissertation. Most gratefully, I thank Dr. Z hifeng Gao for his encouragement and support during the last four years. He w as always available for my questions, and he was always positive and help me overcome a lot of difficult times. His support and continuous guidance enabled me to complete my dissertation. Moreover I want to thank Dr. Lisa A. House Dr. Glcan nel and Dr. Matt Hersom for their valuable comments and suggestions. These valuable suggestions will greatly benefit me in my future studies. I would also appreciate the support from my fellow graduate students. Because of them, my graduate life was full of happiness memories. Finally, special thanks to my family. Words cannot express how grateful I am to my parents, Chung Wang Lu and A Lei Lee, for all of the sacrifices that they have made on my behalf. I would also want to thank my beloved wife WenMin Su who has supported me all the time to strive towards my goal and encouraged me to be a tough person.
5 TABLE OF CONTENTS P age ACKNOWLEDGMENTS..4 LIST OF TABLES ............................................................................................................7 LIST OF FIGURES ...................................................................... ................. ...................9 LI ST OF ABBREVIATIONS ..................................................................... ......................10 ABSTRACT .............................................................................................. ......................12 CHAPTER 1 I NTRODUCTION ................................... ............... ........... ........ ...... ..................... ... 14 1.1 B ackground.................... .............................. .......... 14 1.2 Objective .......................................................... ........ ................... ...........18 1.3 Method ......... .................................. ......................... ........... .............18 1.4 Outline . 19 2 SOLAR ENERGY: ITS HISTORY AND CURRENT SITUATION ............ .... ....... ...20 2.1 Solar E nergy H istory and D evelopment .................................... ... ........... .20 2.2 Solar E nergy Market in U.S. ....................................... .............. ........... ......21 2 .3 Solar Resource Potential in U.S. ...... ........................................... ..............23 2.4 Promoting Policies f or Solar Energy in U.S. ..... ...................... ...................23 3 LITERATURE REVIEW ............................................. .............. ..............................37 3.1 Studies on Consumers Preference WTP for Renewable Energy .. 37 3.2 Studies on the RPL and LCM Model ................................... ................. ......42 4 METHODOLOGY ........................ .........................................................................47 4.1 Choice Experiment Model . . .47 4.2 Multinomial Logit M odel ... .. ..52 4.3 Latent Class M odel . ...53 4.4 Random Parameters Logit M odel . ..54 4.5 Estimation in W illingness to P ay ..6 5 SURVEY DESIGN .60 5.1 Choice E xperiment D esign 60
6 5.2 Que stionnaire S tructure .62 5.3 Data Collection and Descriptive S tatistics .. 4 6 RESUL TS.. .70 6.1 Sample Char acteristics and Opinions. ...70 6.2 Views on Env ironmental Problems in U.S 70 6.3 The Attitude t owards Solar E nergy ..71 6.4 Description of Attributes in Choice Experiment and the Demographic Variables in M odels .. 2 6.5 MNL and MNLIN Estimation R esults 3 6.5.1 MNL Estimation Results 3 6.5.2 MNL IN Estimation R esults ......74 6.6 RPL Model Estimation R esults .. ...77 6.7 LCM Models Estimation R esults ... 1 6.8 Measures of Fit... ... ..83 6.9 Mean WTP V alue 3 6.10 Estimate a Bid C urve 6 7 CONCLUSION AND DISCUSSION 17 REFERENCE LIST ....................................................................................................121 BIOGRAPHICAL SKETCH....12 7
7 LIST OF TABLES Table P age 2 1 Comparison of solar resource and cumulative solar el ectric capacity in 20168 2 2 Financial incentives and Renewable Portfolio Standard pol icies in U.S29 3 1 Summary of studies on preference for renewable energy ........ ............................45 5 1 Attribute description of group 15 5 2 Attribute description of group 2.. .......................................... .................................66 5 3 Policy profiles for solar energy policies in group 1 ... .. .. .................67 5 4 Policy profiles for solar energy policies in group 1 ..... ....................... .....................67 6 1 Summary statistics for survey respondents....88 6 2 Environmental concerns what people worry about in U.S.. .. .90 6 3 The reasons that prevent people from using solar energy ... 1 6 4 The motivation to use the solar energy systems 2 6 5 Willingness to use the solar energy system .. .. ..93 6 6 Variable definition of the individual characteristics .. ...94 6 7 Estimation results of the MNL model for group 1. ............ ............ ...............95 6 8 Estimation results of the MNL model for group 2. ....... .................. ..............95 6 9 MNL IN estimation results for group 1.......... .........................................................96 6 10 MNL IN estimation results for group 2......................... .................. ........................98 6 11 Basic RPL estimation results for group 1... ...................... ...................... .............100 6 12 RPL C estimation results for group 1................ .............................. ............ ........101 6 13 Basic RPL estimation results for group 2.. ..........................................................102 6 14 RPL C estim ation results for group 2.... .................................................... ........103
8 6 15 Parameter estimat es of LCM model for group 1.. ......... ............................ ...........104 6 16 Parameter estim ates of LCM model for group 2 ........... ............................... ........105 6 17 Comparison of model based on statistical goodnes sof fi t for group 1 ........... .....106 6 18 Comparison of model based on statistical goodnes sof fit for group 2 ........... .....106 6 19 Summary statistics of ind ividual WTP from group 1.... ................. ..... .................107 6 20 Summary statistic s of individual WTP for LCM model from group 1 ..... ...............108 6 21 Summary statistics of indivi dual WTP from group 2........ ................... ................109 6 22 Summary statistics of individual WTP for LCM model from group 2..... ..............110 6 23 Estimated bid curve for Group 1 ..... ................................... ................ .............111 6 24 Estimated bid curve for Group 2... ........ ............................................................112
9 LIST OF FIGURES Figure P age 2 1 The Accumulated installation of PV market in EU and in the rest countries A round the world ... ... 2 2 2 Bottom up modeled system price of PV syst em from 2009 2014 ...... ............... ...33 2 3 The installation of solar electric capaci ty (MW) in the U.S. ..................... ......... .....34 2 4 The cumulative solar capacity by states in U.S. ....................... ............. ...............35 2 5 U.S. Solar water heating installation from 2006 to 2010...... .................... .......... ....36 4 1 F our stages of design of choice experiment was designed. ........ ...............59 5 1 The sample choice q uestion for group 1.. .................................. ................68 5 2 The sample c hoice question for group 2.. .................................. .................69 6 1 Kernel density for percentag e of solar energy from group 1. .. 113 6 2 Kernel density for production incentives from group 1.. ... 113 6 3 Kernel density for rebate programs from group 1. .113 6 4 Kernel density for Job from group 1.. . 114 6 5 Kernel density for ITC from group 1 .. .. 114 6 6 Kernel density fr om group 1. ... 114 6 7 Kernel de nsity for percentage of solar energy from group 2 . .115 6 8 Kernel density for net metering from group 2 .. .. ...115 6 9 Kernel density for property tax incentives from group 2 .115 6 10 Kernel density for Job from group 2. ... ..116 6 11 Kernel density for sales tax incentives from group 2 .116 6 12 Kernel density from group 2.. . ....116
10 LIST OF ABBREVIATIONS AIC Ak aike Information Criterion BIC Baye sian Information Criterion CEM Choice E xperiment M ethod CO2 Carbon Dioxide CVM Contingent Valuation M ethod EIA Energy Information Agency EU European Union GHG Greenhouse g as GW th Gigawatts thermal GW Gigawatts HA Hedonic analysis IEA International Energy Agency IIA Independence of irrelevant alternatives ITC Investment Tax Credit LCM Latent Class Model LCOE Levelized Cost of Electricity MNL Multinomial Logit MWh M e gawatt hour NREL National Renewable Energy Laboratory NOK Norwegian K rone PV Photovoltaic RPL Random Parameter Logit
11 RPS Renewable Portfolio Standard STE Solar Thermal Energy SWH Solar Water Heating SEIA Sola r Energy Industrial Association WTP Willingness to Pay
12 Abstract of Dissertation Presented to the Graduate School Of the University of Florida in Partial Fulfillment of the Requirement for the Degree of Doctor of Philosophy THE DETERMINANT OF US CONSUMERS ATTITUDES TOWARD SOLAR ENERGY By ChaoLin Lu De cember 2016 Chair: Zhifeng Gao Maj or: Food and Resource Economics Solar energy provides several significant advantages, such as reduction of the CO2 emissions, increase of energy supply diversification, security of energy, and regional/national energy independenc e Due to the reduced installation cost and the rapid advances in solar energy technology, the installed capacity of solar power has been increasing over time in the United States. Nevertheless, the solar energy capacity and policies differ significantly across the states. Thus, determining the consumers preferences for solar energy and policies is critical for future solar energy policy development. In this study, we used the stated preference methods to elicit consumer preference for solar energy and determine the impact of solar energy policy on consumers support of solar energy. Our results show that consumers mean WTP was $24.110 per month for ITC, 2.5 times higher than rebate programs (i.e. $9.631) and 4 times higher than pr oduction incentives ( i.e. $6.889 ). Consumers mean WTP for property tax incentives ( $24.691 per month) was 2.8 times higher than net metering (i.e.
13 $8.804) and 1.16 times higher than sales tax incentives (i.e. $21.197 ) The fact that ITC and property tax incentives have high WTP values indicates that the respondents prefer to accept direct subsidy when they install the solar energy system. The results of the study could facilitate the development of the solar policy instruments that can more effectively encourage the solar energy adoption and extend the public understanding of the environmental and economic aspects of solar energy.
14 CHAPTER 1 INTRODUCTION 1.1 Background Climate change is one of the many concerns for human activities in the 21 st century (Tingem and Rivington 2009) It may cause heat and cold wave s, increased floods and droughts, and effects the risk of disasters and malnutrition (Panwar et al. 2011) Climate change has become a critical part for many countries st rategies to reach reduction in Greenhouse G as (GHG) emission (Bergmann et al. 2006) In context of raising public awareness and concerns about climate change, the renewable energy capacity has been increasing in the United States (U.S.) in recent years ( O'Keeffe 2014) Renewable energy such as hydroelectric power, wood biomass, geothermal, e thanol, b iodiesel, waste biomass, wind and solar energ ies are considered as important resources in many countries around the world (e.g., Alnatheer 2005; Duke et al. 2005; Gnansounou et al. 2005) A study conducted by t he U.S. Energy Information Agency (EIA) showed that about 66% of electricity consumption was generated by fossil fuel in 2015, while only less than 10% of electricity was produced by renewable energy1. The heavy usage of fossil fuel has increased the level of atmospheric Carbon D ioxide (CO2) which resulted in the global warming and ultimately caused climate change (Hansen et al. 1981) In the U.S., limited supplies of fossil fuel energy and the concerns about climate change have prompted the development of renewable energy (Adua 2008) R enewable energy provides multiple public benefits includ ing environmental improvement ( e.g. reduction of greenhouse gas emission (GHG) ), reduction of the 1 Accessed April 22, 2016, available at https://www.eia.gov/tools/faqs/faq.cfm?id=427&t=3
15 impact of gas and coal price volatility on the economy, as well as increased fuel diversity, national economic security (fossil energy is vulnerable to political instabilities, trade disputes, embargoes and other disruptions) and economic productivity (Menegaki 2008). S olar energy a kind of renewable energy, provides significant envir onmental advantages compared to fossil energy. The advantages of solar energy include the reduction of CO2 emissions, improvement of water resources quality and increase of energy supply diversification and energy security and regional/national energy indep endenc e as well as work opportunities (Tsoutsos et al. 2005) According to the International Energy Agency (IEA) solar power could be the worlds largest s ource of electricity by 20502. There are two main types of solar energy system: s olar photovoltaic (PV) system and solar thermal electricity (STE) system PV system generates electricity directly from sunlight via an electronic process that occurs naturally in semiconductor materials STE system concentrates sun light to produce heat, which is then used by heat engine to run the generator (Xiarchos and Vick 2011) In the first quarter of 2016, 1,665 MW of solar energy were installed in the U.S., more than 700 MW from coal, natural gas and nuclear combined3. In the past, the h igh initial capital cost of solar energy has impeded i nstallation of solar energy (Beck and Martinot 2004) T he p rice of PV system decreased 60% from 2 Accessed April 21, 2016, available at http://www.climatechangenews.com/2014/09/29/ieasolar couldbelargest energy sourceby 2050/ 3 Accessed June 14, 2016, available at http://www.seia.org/news/us solar market track recordbreaking%20year
16 2005 to 2014 which makes solar energy become more accessible and more versatile4. D espite the reduction in the price of PV system solar energy is still more expensive than energ y generated from other sources, such as nuclear, coal, and natural gas. The L eveliz ed Cost of E lectricity (LCOE) of PV system and STE system were $114 M e gawatt hour ( MWh ) and $220 MWh respectively, in 2013 I n the same year the LOCE of conventional coal and conventional cycle nature gas are only $95.1 MWh and $75.2 MWh respectively5. The LOCE of PV system is expected to be close to conventional coal as it s cost decreases continuously ( Hernndez Moro and Martnez Duart 2013) This will make PV system more competitive to other energy (Goodrich et al. 2012) For the solar electric capacity in the U.S. about 7.3 GW of PV system were installed in 2015, which increased over 3 times compared with installed capacity in 20056. This remarkable growth in solar electr ic capacity indicates that solar energy plays an important role in renewable energy Public support for renewable energy has increased rapidly over the past 20 years and many policies were adopted explicitly to promote renewable energy (Beck and Martinot 2004) These policies tried to compensate for the cost related barriers by providing subsidies to renewable energy (Beck and Martinot 2004) For example, the U.S. fe deral government provided investment tax credit (ITC) that provides a 30% federal tax credit claimed against the tax liability In addition, s everal state s offered 4 Accessed April 22, 2016, available at http://www.seia.org/researchresources/solar industry data 5 Energy information administration. Annual Energy Outlook 2015. Accessed April 22, 2016, available at https://www.eia.gov/forecasts/aeo/electricity_generation.cfm 6 Accessed April 22, 2016, available at http://www.seia.org/researchresources/solar industry data
17 rebate program s, produ ction incentives, sales tax incentives and property tax incentives (Xiarchos and Vick 2011) A study investigating the solar energy use in the U.S. showed that both solar energy regulation and solar resource availability in a state have an influential impact on the adoption rate (Xiarchos and Vick 2011) For example, California has relative abundant solar resource, which has 5.26 kWh/m2/day (annual average, 3rd in the U.S.) and fav orable solar energy policies, which make s this state has the highest cumulative solar electricity capacity (13,241MW)7 in U.S. A lthough solar energy resource in states such as Oklahoma is relatively abundant, which has 4.71 kWh/m2/day (annual average, 10th in the U.S.) t he cumulative solar capacity (5.4 MW) is quite low when compared to other states in the U.S because of the lack of favorable solar energy policies. Other than solar energy resource availability and policies, consumer attitude and preference for solar energy are al so important factors that determine the adoption of solar energy ( Farhar 1999; Ek 2005; Wiser 2007) Previous studies estimated positive consumers Willingness to Pay ( WTP ) for renewable energy ( Batley et al. 2001; Zarnikau 2003; Nomura and Akai 2004) and other studies further re ported that individuals have a preference for solar energy over other renewable energy (Roe et al. 2001; Nomura and Akai 2004; Yoo and Kwak 2009) However, previous studies for renewable energy does not explore consumer preference of different types of policies aimed to promote solar energy. Understanding consumer preference of these policies is important as policies with the same cost may have different impacts on consumer adoption of solar energy if these polic i es have various installation cost reduction 7 The amount of cumulative solar electric capacity installed until March 2016
18 focuses. This project fills the gaps in the current literature by eliciting consumer s preference and their WTP for different solar energy promotion policies, which provides significant implications for the design of more effective government policy that aim to increase solar energy adoption. 1.2 Objective Due to the reduced production cost of solar energy the installed capacity of solar power has been increasing over time in the U.S. Interestingly, the installation rate of solar energy system differs very much across states. Solar ene rgy resource availability and policies have shown to be the incentives to promote solar energy. The objective of this study is to explore the consumer s preferences and their WTP for different solar energy promotion policies in the U.S Particularly this study determined the consumer pre ference for different solar energy policies ; and consumer WTP for different attributes of solar energy policy. T he results could help understand the impact of economic aspects of policies on consumer adoption of solar energy, therefore to promote solar energy to the public by designing the most effective solar energy policy instruments. 1.3 Method This study uses ch oice experiment s to elicit consumers preference for solar energy promotion policy. Choice experiment is a stated preference valuation method that to assign the WTP for environmental goods and services. Choice experiment is incentive compatible and is bas ed on random utility theory and Lancasters theory of utility maximization. The data for this study was collected using an online survey that was distributed by a survey company to its national representative consumer panels.
19 The WTP for solar energy policies was derived from several different variants of multinomial logit model with stated choice data. These include a simple multinomial logit (MNL) model, a simple MNL with interaction terms between respondent characteristics an d policy attribute (MNL IN), a Latent C lass model (LCM), a basic Random Parameter Logit (RPL) model, and RPL model with the correlation between attributes (RPLC). We estimate the mean consumer WTP for each of several different policies and the degree of heterogeneity among respondents for each policy. Finally, we estimate the bid curves that are derived from the multiple linear regressions. The dependent variable represents the mean WTP for solar energy policies and demographic variables are ind ependent variables, showing consumers responses to demographic variables such as gender, income, education, and having children, as well whether the respondent is a homeowner or renter. 1.4 Outline The rest of this study is organized as follow. Chapter 2 present s current situation and history of the solar energy. Chapter 3 presents a literature review, including previous studies on the consumers preference WTP for renewable energy and previous applications of and comparison between the RPL and LCM model. Chapter 4 discusses the methodology, including choice experiment design, Multinomial Logit model, Latent Class model, and Random Parameters Logit model. Chapter 5 explains the details of the survey conducted. Chapter 6 presents the results of MNL, MNLIN, RPL, and LCM model. The conclusion remarks and further implications are provided in Chapter 7.
20 CHAPTER 2 SOLAR ENERGY: ITS HISTORY AND CURRENT SITUATION 2 .1 Solar Energy History and D evelopment According to EIA, solar energy technology has developed over 200 years. However, the development of PV system was not started until 1876 when William Grylls Adam s and Richard Evans Day discov ered that selenium can produce e lectricity after it is irradiated by light1. In 1954, Daryl Chapin, Calvin Fuller, and Gerald Pearson invented PV system in the U.S. (Chapin et al. 1957) This is the first time that solar cell s could convert enough energy to electricity; nevertheless the cost of solar cells was n ot affordable for everyone. Afterward, in the 1970s Elliot Berman manufactured a lower cost solar cell that de creased the price from $100/w to $20/ w (Berman 1985) In the 2000s, the solar power facilities and the rooftop solar power system s started to be used in the U.S.2. Moreover d uring 2003 to 2013, the cost of PV installation rapidly decreased from $8/W to $3/W in the U.S U ntil 20 13 the cumulative capacity of PV system reached about 4.93 GW in the U.S.3 In the same year, there were 38.35 GW of PV system installed globally, including European countries (10.97 GW) and countries in Asia, North and South America, a nd Australia (27.38 GW) (Figure 1 Accessed April 22, 2016, available at https:// www1.eere.energy.gov/solar/pdfs/solar_timeline.pdf 2 Accessed April 27, 2016, available at http://pureenergies.com/us/how solar works/timelineof s olar power in the united states/ 3 European Photovoltaic Industry Association, 2014, EPIA Global Market Outlook for Photovoltaics 20142018. Available at: http://www.epia.org/fileadmin/user_upload/Publications/EPIA_Global_Market_Outlook_for_Photovoltaic s_20142018__Med ium_Res.pdf
21 2 1 ). Among these countries, Germany was the top market with 3.3 GW in Europe. In Asia, China was the top mar ket with 11.73 GW (Figure 2 1 ). Solar thermal energy (STE) can be used to heat water and heat or cool buildings. In the STE markets, the world total installed amount in 2014 was about 374.7 Gigawatts thermal (GWth) with 44.1 GWth in the European Union (EU) and 262.3 GWth in China, which was the top STE market in the world. The remaining installed capacity of 68.3 GWth was shared by the U.S. and Canada (17.7 GWth), Asia excluding China (10 GWth), and Latin America (8.7 GWth)4. Historically, numerous solar thermal systems were installed in the U.S. However, in th e mid 1980s, the numbers of installations were decreasing due to the expiration of federal solar tax (Sherwood 2007) 2 .2 Solar Energy M arket in the U.S. The cost of PV system is a critical factor in the development of solar energy market. U.S. Department of Energy statistics data indicates that the c ost of producing solar energy has been decreasing in recent years For residential PV system the modeled system price decreased from $6.97/w in 2009 to $3.12/w in 2014, a reduc tion of 55 %. For commercial PV system, the modeled system price decreased from $5.1 6/w in 2 009 to $2.17/w in 2014. For utility ground mount, the modeled system price of PV decreased from $4.4/w in 2009 to $1.78/w in 2014 ( Figure 22 ). This price reduction indicates that the solar energy is becoming more affor dable than that of 5 years ago. The p rice reduction is primarily related to a precipitous drop in PV system module prices in recent years 4 Accessed April 27, 2016, available at http://www.iea shc.org/solar heat worldwide
22 The high volatility of gas price in recent years has further boosted the demand for solar energy and provided the incentive to develop this industry (Xiarchos and Vick 2011; Barbose et al. 2015) This situation brought positive effects on the job market in the U.S. since more solar PV installers are needed in the j ob market (Kammen et al. 2004) Statistics from Solar Energy Industries Association (SEIA)5 shows that in 2015 the solar PV system of 7,260 MW was installed, which is an increase of 700% over the past ten years (Figure 23), and the total installed capaci ty reached 27.4 GW. Each year 30,000 solar water heating and cooling systems are installed in the U.S. which earns $435 million in annual revenue. The cumulative solar capacity between states varies dramatically in U.S. in 20166 (see Figure 24 ). The cumulative solar electricity capacity in California is 13,241 MW the highest among states in the U.S. The second ranked state is Arizona with 2,303 MW, followed by New Jerse y with 1,632 MW cumulative solar electricity capacities The STE provides highly reliable energy for heating and cooling. Approximately 30,000 Solar Water H eating (SWH) installations annually were installed from 2008 to 2010 (see Figure 25). The market penetration of STE in U.S. is relatively low compared with other countries, with only 9 GWth of installed capacity ranking 36th in the world. As a result, STE did not grow as fast as PV system in U.S. and this is the reason that this study focuses on PV system. 2 .3 Solar Resource P otential in U.S. 5 Accessed April 28, 2016, available at http://www.seia.org/researchresources/us solar market insight 6 Accessed April 28, 2016, available at http://www.seia.org/researchresources/top 10solar states
23 One of the key factors that deter mine the development of the solar energy is solar resource7. Solar resource is defined as the amount of solar energy at various locations on Earth (units usually ki lowatt per square meter per day ( kWh/m2/day) (Xiarchos and Vick 2011) The solar resource data is given by National Renewable Energy Laboratory (NREL). The sout hwestern part of the U.S. has the greatest potential ac cess to the solar energy ( Table 21 ). A comparison between cumulative solar electric capacity and solar resource of states demonst rate that California, Arizona, Texas, Colorado, North Carolina, Hawaii, and Nevada make use of the abundant solar resources in their states very well but other states such as Oklahoma and Utah did not do very well to access solar energy. A lthough the solar resource is not high in New Jersey, Massach usetts, and New York, these three states are among the top ten states for cumulative solar electric capacity. Theoretically, high solar radiation means an area has high potential to develop the solar energy the comparison between solar resource and cumulative solar electricity capacity indicates that there are other factors to affect the solar energy adoption. 2 .4 Promoting Policies for Solar E nergy in U.S. Solar energy promoting policies may explain the disparity between solar resource and solar energy installation capacity across states in the U.S. Scarpa and Wills (2010) pointed out that if the government wanted to reach sustainability and security of renewable energy supply, it must adopt some policies to correct the market failure (Scarpa and Willis 2010) Solar energy policies were built to stimulate solar energy 7 National Renewable Energy Laboratory(NREL), 2014. Solar Maps. Accessed April 22, 2016, available at http://www.nrel.gov/gis/solar.html
24 market growth and maturity (Barbose et al. 2015) The main goal of these policies is to encourage adoption of solar energy and further change societies behavior of fossil fuel consumption and make solar energy competitive without subsidies. T here are three kinds of schemes for government to promote solar energy. One scheme is federal financial incentive including ITC The s econd scheme is state incentives including state financial incentives and support incentives. The state financial incentives are rebate programs, production incentives, sale tax incentives, property tax incentives and t he state supporting incentive is a Renewable Portf olio Standard (RPS) (Xiarchos and Vick 2011) The RPS indicates a minimum share of electricity from renewable energy sources and it is an incentive for renewable producers to reduce costs (Berry and Jaccard 2001) T he t hird scheme is financial considerations for promoting solar energy including commercial bank loans and net metering. ITC is a policy at the federal level and all other programs such as rebate programs, production incentives, sale tax incentives, property tax incentives and RPS are at the state level. The ITC8 gives a tax credit for 30% of the cost of a residential and commercial solar system This tax credit is used for both existing homes and for new construction when the homeowners buy a solar energy system and plan to install it. For example, if the solar energy system cost the homeowner $10,000, the homeowner will get a $3,000 in federal tax credit back. I t has been a crucial tool to boost the amount of solar project nationwide. This policy started in 2006 and is set to end in 2016. However, it will be extended to 2019 and then will be reduced to 22% by 2 021 After 2023, the credit will go 8 Ac cessed April 23, 2016, available at http://www.seia.org/policy/financetax/solar investment tax credit
25 to zero. The benefit of ITC is to increases the U.S. solar manufacturing capacity and reduces the cost of solar for consumers Therefore, this policy can bring strong financial incentive for residential and commercial use of solar energy in U.S. Solar energy policies at states level vary significantly. For the financial considerations of the solar system, net metering can let the installer get the additional income for the excess electricity they generate from the solar system. Forty six states adopted a net metering policy to promote solar energy. Forty three states adopt this policy at the state level and three sta tes, Idaho, South Carolina and Texas implemented voluntary utility programs by utility companies (Table 22 ). Other state level policies include rebate programs, production incentives, sales tax incentives and property tax incentives. Rebate programs provide a discount for solar energy installation. Production incentives are policies that pay residents for every kWh generated from the solar system Sales tax incentives provide the exemption from state sales tax for a n installing solar system. Pro perty tax incentives also provide an exemption from state property tax for installing solar system (Xiarchos and Vick 2011) Some states have multiple options for rebates programs as well as production incentives, sale tax incentives and property tax incentives. For supporting state incentives, the Renewable Portfolio Standard (RPS) is regard ed as the most important policy to promote the solar energy (Wiser and Barbose 2008) RPS is a law or other piece of regulation that mandates t hat a certain percentage of a state s energy production comes from renewable resources by specified targ et dates. RPS usually builds in targets that increase over time. For example, a state can regulate that its renewable energy generation to incre ase by 3%
26 each year for the ten years. In other words, the RPS encourages the utility company to invest more in solar energy. Twenty eight states in U.S already established the RPS (T able 22 ). Many of the RPS mandate goals are as high as 30%, even 40% o f the total energy production in the next 6 to 16 years. The top ten states for cumulating solar electricity capacity all have adopt ed the RPS policy. Conversely, some states, like Florida and Vermont have not adopt ed th e RPS policy Net metering lets the customers who generate their own electricity from solar energy to feed electricity they do not use back into the grid. Net metering is a billing mechanism that cre dits solar energy system customers for electricity they produce to the grid. Therefore, if the residents install a PV system on the houses rooftop, it is possible to generate more electricity than house uses. If a home has the net metering system, the electricity meter will run backward to get the extra financial credit for extra solar power. Statistics in Table 2 1 show that the net metering is the most popular policy promoting solar energy (46 states adopted it) in the U.S. Solar tax exemptions include both property and sale tax exemptions provide by the state government to residents and company that install the solar energy.9 Property tax incentives allow residents to exclude the adde d value of a solar energy system from the valuation of their property for taxation purpose For example, renewable energy property tax exemptions in Nevada permits people to apply for a property tax exemption to 55 percent of these property values for 20 years. Table 21 shows that there are 27 states that off er the property tax exem ptions. Sale tax exemptions offer an exemption 9 Solar Energy Industries Association (SEIA), Solar Tax Exemptions. A ccessed April 25, 2016, available at http://www.seia.org/policy/financetax/solar tax exemption
27 from the state sales tax for installing a solar energy system. For instance, a sales tax exemption was provided in Arizona for the sale of solar energy devices and the installation of solar energy systems. The purpo se of sale tax and property tax exemptions are to help lower the installation cost of a s olar energy system. In Table 22 there are 21 states that provide a sale tax exemption.
28 Table 21 Comparison of solar resource and cumulative solar electric capacity in 201610. State Cumulative solar electric capacity installed (MW) (as of March 2016) Solar resource**(kWh/m2/day) Arizona 2,303* 5.64 California 13,241* 5.26 Hawaii 564* 5.22 Nevada 1.240* 5.16 Texas 534* 5.04 Utah 263 4.99 Colorado 540* 4.84 Oklahoma 5.4 4.71 North Carolina 2,087* 4.43 New Jersey 1,632* 4.01 Massachusetts 1,020* 3.81 New York 638* 3.71 Note: Top ten state with the highest cumulative solar capacity ** The average amount of solar energy is 4.29 kWh/m2/day 10 Resource: National Renewable Energy Laboratory(NREL), http://www.nrel.gov/gis/solar.html
29 Table 22 Financial incentives and Renewable Portfolio Stand ard Policies in the U.S.11. States Rebate programs Production Incentives Sale Tax incentives Property Tax incentives RPS* Net metering Electricity production from renewable energy (kWh) 12 Alabama S tate policy; : V oluntary utility program only 11 Source: Database of State Incentives f or Renewable & Efficiency, 2014 12 Accessed June 15, 2016, available a t https://www.eia.gov/electricity/monthly/pdf/epm.pdf
30 Table 22. Continued States Rebate programs Production Incentives Sale Tax incentives Property Tax incentives RPS* Net metering Electricity production from renewable energy (kWh) Massachusetts S tate policy; : V oluntary utility program only
31 Table 22 C ontinued States Rebate Program Production Incentives Sale Tax incentives Property Tax incentives RPS Net meterin g Electricity production from renewable energy (kWh) Rhode Island S tate policy; : V oluntary utility program only
32 Figure 21. The accumulated installation of PV market in the EU (right) and in the rest countries around the world (left) in 201313. 13 Resource: European Photovoltalc Industry Association http://www.cleanenergybusinesscouncil.com/site/resources/files/reports/EPIA_Global_Market_Outlook_ for_Photovoltaics_20142018__Medium_Res.pdf
33 Figure 22. Bottom up modeled system price of PV system from 2009 to 201414. 14 Resource: U.S. Department of Energy https://emp.lbl.gov/sites/all/files/pv_system_pricing_trends_presentation_0.pdf 0 1 2 3 4 5 6 7 8 2009 2010 2011 2012 2013 2014Price of PV system($/w)Year residential ($/w) commercial ($/w) utility ground mount ($/w)
34 Figure 23. The installation of solar electric capacity (MW) in the U.S. from 2000 to 201515. 15 Resource: Solar Energy Industries Association (SEIA) http://www.seia.org/researchresources/us solar market insight 0 1000 2000 3000 4000 5000 6000 7000 8000 Electricity capacity (MW) Year
35 Figure 24. The cumulative solar capacity by states in U.S. until March 201616. 16 Resource: Solar Energy Industries Association (SEIA) http://www.seia.org/researchresources/top10solar states 0 2000 4000 6000 8000 10000 12000 14000 Cumulative solar capacity(MW) State
36 Figure 25 U.S. SWH installations from 2006201017. 17 Resource: Solar Energy Industries Association http://www.seia.org/researchresources/solar heatingcoolingenergy securefuture 0 5000 10000 15000 20000 25000 30000 35000 40000 2006 2007 2008 2009 2010Number of SWHYear
37 CHAPTER 3 LITERATURE REVIEW In this research, I used the discrete choice model to determine consumers preference toward solar energy promoting policies in the U.S. T his method has been used in many studies to analyze the individuals preference, including transportation choice (Shen 2009; Greene and Hensher 2003; Caussade et al. 2005) recreational choice (Boxall et al. 1996; Horne et al. 2005; Bos et al. 2004) and choice of products or services (Loureiro and Umberger 2007; Swait and Adamowicz 2001) In t his chapter the previous studies on the consumers preference and WTP for ren ewable energy and applications of RPL and LCM model will be discussed. 3.1 Studies on Consumers Preference WTP for Renewable Energy Because of the perceived environmental benefits and decreasing installation cost of solar energy system, peoples interest in solar energy has been increasing and the demand for solar energy has inc reased significantly in the past 10 years. Many studies have been conduct ed to elicit consumers preferences and WTP for renewable energy. Batley et al. (2001) explored the advantages and disadvantages of renewable energy in the UK C ontingent Valuation M ethod (CVM) was used to estimate consumers WTP for green electricity from 692 received polls They showed that WTP varies with social status and income, suggesting income is a significant factor in determining the WTP values, and the higher the i ndividuals income the more they will pay extra for renewable energy Roe et al. (2001) explored the U S consumers WTP using Hedonic analysis (HA) in the U.S They concluded that consumers were willing to pay a small amount for
38 the green electricity. There are two methods in this study : one is a consumer based conjoint survey and the other is HA Data was collected from a total of 835 people in eight different cities in the U.S: Cincinnati, Ohio; Holyoke, Massachusetts; Houston, Texas; Jacksonville, Florida; Riverside, California; Philadelphia, Pennsylvania; Portland, Oregon; and Salt Lake City, Utah. The r esults indicated that consumers are willing to pay more for the renewable electricity that can reduce air emissions The HA suggested that the annual premium for renewable energy was $59.4 and for solar energy was $83.2. Zarnikau (2003) examined the WTP for electricity utility investment in renewable energy and energy effic iency in Texas by a method of openended CVM. A Tobit model was used to explain WTP for renewable energy from a survey with a total of 200 to 250 consumers that were selected from Deliberative polls. The dependent variable is the extra dollars that consumer is willing to pay and t he independent variables are female, owned a home, education, age, salary, and race. They show ed that i nformation about energy resource options can increase consumers WTP for renewable energy and they also proved that WTP increased with age, education, and salary. Ek (2005) co mpared the publ ic and private attitudes toward the green electricity in Sweden by CVM and estimated the WTP using a binomial logit model A total of 1000 households were interviewed while 453 of these refused to participate. This study concluded that if consumers value the environment and are willing to pay more for green electricity, the consumption of renewable e nergy should be increasing. The results also indicated that the Swedish householders had a positive attitude toward wind power.
39 Bergmann et al. (2006) discussed the impact of renewable energy attributes on the household prefere nce for renewable energy technologies in Scotland by Choice Experiment M ethod ( CEM ) T hree attributes, including landscape, wildlife, and air quality in the choice experiment were used in this study The results demonstrated that the households are willing to pay more to decrease the impact on landscape changes and to reduce harm to wildlife, as well as to improve the air quality Borchers et al. (2007) explored consumers preference and WTP for green electricity in the U.S. usin g CEM. The attributes in the study contain ed monthly electricity usage and green energy sources, including w ind, solar, biomass and farm methane. A total of 128 c onsumers were interviewed in Delaware in 2006. The results suggest that there is a positive WTP for green electricity and individuals had a preference for solar energy compared to other green sources Longo et al. (2008) did not focus on the preference for different types of energy source. Instead their study elected the WTP of UK energy users for different characteristic energy source using CEM The attributes in this work are annual reduction in GHG emission due to renewable energy increase, annual length of electricity shortages in minutes, change in the number of employees in the electricity sector, and electricity bill. The data was analyzed by a MNL model and a condition logit model. T he utility parameters were all positive except price and annual length of electricity shortages in minutes indicating that the WTP is positive for reducing GHG emission, avoiding energy blackouts, and increasing employment. Navrud and Brten (2007) use d the CEM to elicit peo ple s preference in Norway and WTP fo r different energy sources, and also discussed whether the preference
40 differ ed if the people used a renewable energy. The attributes for this study include multiple type s of energy source, such as wind power, hydropower, and natural gas, annual fee on the electricity, and size of power. The data was analyzed by a MNL model and a condition logit model for a rural subsample (91 respondents ) urban subsample (98 respondents) and the total sample (189 respondents ) They found that people in rural areas have a higher WTP than those in urban areas and that if the respondent is female, the demand for compensation is reduced by 745 Norwegian K rone ( NOK ) While the e ffect of education is ambiguous, the age of respondents seems to play a role in the demand for compensation. Yoo and Kwak (2009) made use of CVM to estimate consumers WTP for green electricity in Korea. A total of 800 interviews were conducted by a face to face interview in 2006. The results showed that 63.4% of respondents knew about the PV system which was the most well known renewable energy. Furthermore, t ide power generation and wind power (8.1% of respondents) are the seco nd best well known renewable energ ies. The annual households WTP for green electrici ty is $1.8 and increased with income. Scarpa and Willis (2010) utilized the CEM to obtain the household WTP for renewable energy in the UK There are two kinds of CEM: primary heating choice experiment and discretionary technology choice experiment. The questionnaire with 1279 respondents was composed of a stratified sample of households across England, Wales and Scotland in 2007 S ix attributes including capital cost, energy bill, m aintenance cost, the source of recommendation (friend or plumber) contract length, and inconvenience of the system were used in the primary heating choice experiment.
41 For the discretionary technology choice experiment, there were five attributes including types of system, capital cost, energy bill per month, maintenance cost, and the source of recommendation. The results suggest that consumers WTP was between 2.61 British Pound to 3.21 British Pound in capital costs to reduce annual fuel and further adopt renewable energy. This value also can be applied t o solar electricity, solar thermal, and wind power. Cicia et al. (2012) explored the peoples preference for wind, solar, biomass, nuclear, and fossil energy in Italy from a survey of 504 questionnaires collected in 2009. The data was analyzed by a latent class model that tak e s into account socio economic covariates to explain the sample heterogeneity. They found that 3 equally distributed classes and the utility parameters for nuclear power are all negative, indicating that the nuclear power could bring the negative welfare impact for respondents. The mean WTP value was about 37. 5 for solar and wind energy Yoo and Ready (2014) used C EM to explore the consumers attitudes toward renewable energy from a survey with 783 respondents in Pennsylvania in the U.S The attributes were percentage of renewable electricity in Pennsylvania by 2020, impact on jobs, and impact on electricity price and taxes. The results showed that the average households are willing to pay $55 per year to increase wind power and $42 per year to increase the solar power. Overall, CVM and CEM are the two most widely used methods to estimate consumers preference for renewab le energy or green electricity. While both CVM and CEM are survey based methods CEM allows respondents to make a tradeoff between different attributes. CEM allows estimates of relative WTP values, which are useful for
42 the policy maker. Nevertheless, t he result of WTP should not regard as actual WTP but rather as consumers relative preference (Menegaki 2008) Despite the heterogeneity of previous studies, we can summarize the main findings from the previous literature. Firstly, solar and wind usually are the most well known renewable energ ies I nsomuch that if the studies tried to compare differ renewable energy sources, respondents are willing to pay more for the solar and wind. Secondly, the WTP is higher among younger and higher income. In addition, WTP increased with the respondents concern with the e nvironmental issues. At last most previous studies just discuss the preference for green energy sources but do not further explore what policy factors may motivate people to use the renewable energy. 3.2 Studies on the RPL and LCM Model In this study, I explore the WTP for solar energy using several different variants of MNL model with stated choice data, including a MNL model, a MNL model with interaction terms between respondent characteristics and attributes, a Latent Class M odel (LCM) assuming discrete heterogeneous preferences, and a random parameter logit model (RPL) assuming continuous heterogeneous preference. The RPL model is defined by accommodating heterogeneity as a continuous function of the parameters. T he utility parameters are random following some ex ante specified distribution. Conversely, LCM model can be regarded as a semi parametric version of RPL (Sagebiel 2011) and let preferences differ by different classes Both RPL and LCM models relax some of the restrictions of the MLN model that is nested with both models (Yoo and Ready 2014) The likelihood ratio tests can be used to compare between MNL and RPL and between MNL and LCM. Usually, LCM
43 and RPL model s have a better statistic fit relative to MNL (Beharry Borg and Scarpa 2010; Greene and Hensher 2003; Shen 2009) However, a comparison between RPL and LCM cannot be made with likelihood ratio tests because one model is not nested with the other (Yoo and Ready 2014) The following reviewed some key studies on RPL and LCM models. Green and Hensher (2003) compared LCM and RPL models in term of choice elasticities for a change in travel times and mean WTP estimates. Both model s provided alternative way s to capture the unobserved heterogeneity and variability in unobserved sources of utility. They made use of choice probabilities under LCM and RPL for each alternative and examined the relationship between both models. Three latent classes were determine d based on statistical value. Compared with the value of t ravel time savings (willingness to pay) derived from RPL and LCM model s, the LCM model could provide more information than RPL model The study showed that there was a weak relationship between tw o models. They found that respondents behavioral sensitivity to an attribute (changes in travel time) is decreasing in the LCM relative to the RPL model Shen (2009) applied the stated choice survey datasets of Japan to investig ate the difference between LCM and mixed logit model for transport mode choice. F ive attributes in this stated choice experiment ; included (1) in vehicle time including delay time caused by a traffic jam, (2) access time, (3) frequency, (4) travel cost a nd (5) negative impact on the environmental caused by transport. There were 467 individual s answer ed the questionnaire with 8 choice sets. Furthermore, Shen (2009) also used a test on non nested model s based on Akaike Information Criterion ( AIC ) to help
44 determine which model is better than another one. The results of this study suggest ed that the LCM performs better than the RPL in both datasets. Kosenius (2010) examined consumers preference heterogeneity for water quality attributes and they also determined welfare estimates for three nutrient reductions in the Gulf of Finland. The dataset was composed of 1900 Finns of ages 1880 drawn from Census Register, analyzed with MNL, RPL, and LCM models. The results from MNL and RPL show ed that clear water is the most important water quality characteristics followed by the desire for fewer occurrences of blue green algae. The LCM model revealed that the order of relative importance of attributes depends on age, household income, coastal residenc e, and vacation home ownership. They also concluded that the RPL model was better than the LCM model when the sample is weighted to correct for sampling bias. Beharry Borg and Scarpa (2010) used LCM and RPL models to analyze unobserved taste heterogeneity and estimated WTP for an improvement in coastal water quality This study used a dataset of 284 respondents and reported the results of two choice experiment for two group: snorkellers and nonsnorkellers They found that based on statistical criteria, a LCM outperforms a RPL for snorkelers, but a LCM did not perform well in nonsnorkellers
45 Table 31. Summary of studies on preference for renewable energy Sources Goods to be valued C ountry Methodology Results Batley et al. (2001) Green electricity UK CVM WTP varies with social status and income. 34% of respondents want to pay more to adopt green electricity. Roe et al. (2001) Green electricity USA HA The WTP of the annual premium for 100% renewable energy was $59.4 and for solar energy was $83.2. Zarcnikau (2003) Renewable energy USA (TX) CVM Consumers willing to pay a small premium for renewable energy. WTP increased with age, education and salary. Nomura and Akai (2004) PV system and wind energy Japan CVM WTP is about 2000 yen ($17). Bergmann et al. (2006) Renewable energy Scotland CEM Households were willing to pay more to d ecrease high impact landscape changes, reduce harm to wildlife, and i mprove the air pollution. Ek(2005) Wind power Sweden CVM 1. A positive toward wind power. 2. The probability of finding an individual is decreasing with age and income. Borchers et al. (2007) Green electricity USA CEM Positive WTP and the individuals have a preference for solar energy over the other renewa ble energy. Navrud and Braten (2007) Renewable energy Norwegian CEM People in urban had a higher WTP than in rural. Longo et al. (2008) UK CEM WTP is positive for reducing GHG emission avoiding energy blackouts, and increasing employment
46 Table 31. Continued Sources Goods to be valued C ountry Methodology Results Yoo and Kawk(2009) Green electricity Korea CVM PV system was the most well known renewable energy.Annual household s WTP for green electricity was $1.8 and increased with income. Cicia et al (2012) Renewable energy Italy CVM The WTP for was about 37.5 for solar and wind. The WTP in agriculture is 21 for biomass Yoo and Ready (2014) Renewable energy USA CEM The average households were willing to pay $55 for wind power an d $42 per year for solar power. Note: CEM, CVM and HA denotes choice experiment method, contingent valuation method and Hedonic analysis.
47 CHAPTER 4 M ETHODOLOGY 4.1 Choice E x periment M odel State d preference method has been one of the most productive areas of applied research over the last 30 years to understand and explore the individual behavior when choosing among several options (e.g. goods, sites, and services) It can provide insightful information to better u nderstand consumer behavior, the effects of changes in product or service design, price strategy, distribu tion channel and communication st rategy (Louviere et al. 2000) The advantage of the stated preference methods is that they can estimate demand for a new product or service w hen it is not traded in the marketplace (Louviere et al. 2000) The CEM is one of the important state d preference valuation tool s to estimate the WTP. CEM has been widely applied in transportation research (e.g. transportation mode choice), environmental valuation (e.g. wetland management scenario) marketing ( e.g. food label choice) and public health (e.g. health care choice) (Hensher 1994; Louviere 1988; Louviere 1992) This m ethod i s based on two theories: Lancasters theory of utility maximization based on product characteristics and random utility theory (OKeeffe 2014; Manski 1977; Cascetta 2009; McFadden 1980) Lancasters theory of consumer demand described that the consumers derived utility is not from goods but the characteristics or attributes that goods possess. The value of a good is the sum of the value of its characteristic s (Lancaster 1966) Random utility theory stated that not all determinants of individuals value are observed by researcher s, so the utility function can be divided into observed and stochastic components (McFadden 1973) M odel ing
48 individual choice behavior requir e s three components including choice s to decisionmakers t he observed attributes of the choices by decision makers, and the random components that cannot be observed by researchers. The distribution of random components determines behavior patterns in the population and the specific choice of the mode ls For the random utility theory, l et Uim be the utility of the i th alternative for the mth individual. Assume each utility value can be partitioned into two factors: a systematic component, Vim and a random component, The random component defines unobserved individual idiosyncrasies of tastes. = + (4 1 ) = ( 4 2 ) s are utility parameters and only utility parameters are independent of m. C is a vector of choice attributes. Utility parameters can be expressed to vary across the sampled observations or a function of contextual influences such as the socioeconomic characteristics. On the other hand, the systematic component is the part of utility which can be observed. The random component is the utility contributed by attributes unobserved. W e can assume that indivi duals attempt to choose an alternative that maximize s the individuals utility. The assumption is that the individual m will choose the alternative i if and only if (iff) > ( 4 3 )
49 From the Equation 41 and 43 The alternative i is chosen iff + > + (4 4 ) We can rearrange the Equation 44 ( ) > ( ) ( 4 5 ) We cannot observe the ( ) Thus we have to calculate the probability that ( ) will be more than ( ) or the probability that ( ) will be smaller than ( ) There are four stages in the application of choice modeling (F igure 41) : (1) selection of attributes and assignment of levels (2) ch oice of experiment design, (3) construction of experimental co ntext and, (4) sampling and estimation strategy At the first stage, f ocus group, pi l ot study, and previous studies are a good starting point to determine the selection of attributes and assignment of levels (Martinsson et al. 2001; Hanley et al. 2001) The selection of attributes must be expected to affect the respondents choice. Besides the determining the attributes, it is important to explore the interaction effect between attributes. In additional, if we try to evaluat e the welfare m easure, a monetary attribute should be included such as price or cost (Martinsson et al. 2001) Feasible, realistic, non linearly space and span of the range of respondents preference maps must be considered to decide the attribute level (Hanley et al. 2001) Therefore, it is important to make the levels more realistic and attributes level can describe the current situation. A baseline alternative can be included
50 in each choice set This opinion can show the responde nts current feasible choice (Hanley et al. 2001) At the second stage, the experiment design is used to combine the levels of attributes into alternatives or profiles. The purpose of experiment design is to efficiently create a set of choice scenarios ( set s) where attribut es are uncorrelated among choice alternatives Optima l design is very important in the choice of experiment design (Martinsson et al. 2001) Martinsson et al (2001) stated that there are two steps to develop the experiment design: one is to obtain the optimal combination of attributes and levels and t he other is combining the alternatives or profiles into a choice set. Choice experiment design focuses on the way to find the efficient choice set s and commonly use approach is D optimal design. The purp ose of D optimal design is to extract the maximum amount of information from a selection of attribute combinations The efficiency measure is D efficiency: D efficiency = [| | ] (4 6 ) Where Q is the number of attributes level combination and is the covariance matrix of a vector of attributes level combination. The reason to use the D optimal design in the linear model is that it is less computationally burdensome and it could be directly obtained from some statistical software, such as Statistical Analysis System ( SAS) (Shen and Saijo 2009) For a linear model in which the response variable is a linear function of the explanatory variables, the efficiency of a ( ND ) d esign matrix X is based on the information matrix XX The definition of p is the number of parameters in the linear
51 model and is the number of design points. An efficient design will have a small variance matrix and the eigenvalues of ( ) provide measure of its size. The most useful index is D efficiency. Defficiency is a function of the geometric mean of the eigenvalues, which is given by | ( ) | / (Kuhfeld 2005) D efficiency=100* | ( ) | / (4 7 ) The value of D efficiency scaled from 0 to 100. If D efficiency is 0, one or more parameters cannot be estimated. If D efficiency is 100, the design is balanced and orthogonal (Kuhfeld 2005) In other words, if the values of D efficiency are between 0 and 100, all the parameters in the model can be estimated, but cannot reach D optimal design. There are four rules for efficient design based on a nonlinear model: (1) orthogonality; attribute level s are independent of each other (2) level balance; each level appears an equal number of times (3) minimal overlap; minimizing the number o f times that each level appears in a choice set, and (4) utility balance ; no choice set is in a dominant alternative (Herriges and Kling 1999) At the third stage, optimal attribute combinations identified by the experimental design are grouped into choice set s to be presented to respondents. We assume that the respondents have a rational, stable, transitive, and monotonic preference. Also, we assume that they do not have any problem to complete the survey and do not get too tired to finish the ques tionnaire (Kuhfeld et al. 1994) The final stage is sampling and estimation strategy. Given the survey population, a sampling strategy could be a simple random sample, a stratified random sample, or a
52 choice based sample. For parameter estimation, we can estimate the utility parameters and compare and select the best model by conducting overall goodness fit tests. 4 .2 Multinomial Logit Model Multinomial logit (MNL) model is the basic choice model based on random utility theory. For the MNL model, there is a single vector of characteristics, which describes the individual and a set of parameter vectors. The g alternatives are characterized by a set of h attributes. Respondent i choose among G alternatives. The model underlying the observed data is following random utility specification. ( ) = = + = 1 (4 8 ) The is a single parameter vector. The random individual terms ( , ) a re assumed to be identically independently distributed with type 2 extreme value distributions Under these assumptions, the probability that individual i chooses alternative g is > (4 9 ) Specifically, the probability is ( = ) = ( ) ( ) (4 10) Where i s the index of the choice made, is a vector of choice attributes, and is a vector of preference parameters (also called marginal utilities) to be estimated. Individual characteristic s cannot be put into the utilit y function alone because individ ual characteristic are invariant across alternatives (Yoo and Ready 2014) If the
53 researchers want to include the individual characteristics in MNL, they need to consider the interaction between individual characteristics and choice attributes (Champ et al. 2012) 4 .3 Latent C lass Model Latent Class M odel ( LCM) assumes that there are S segments i n the population. Preference of individual differ s among segments. However, individual preference within each segment is homogeneous. The LCM model assigns each respondent to a segment according to each individual characteristics and choice behavior (Yoo and Ready 2014) Suppose individual i who belongs to segment s chooses alternative g in a choice situation h | = + | (4 11) Where is a vector of choice attributes, | is an unobserved component within a class, and is a classspecific vector of parameter s. Supposing Independence of Irrelevant A lternatives ( IIA ) hold within a class the probability of respondent i choosing alternative r in a choice situation h | = ( ) ( ) (4 12 ) In LCM, it is necessary to define the probability of class membership in each class for each respondent. The class membership probability of indivi dual i being classified to class S can be rewrit ten as a multinomial logit model:
54 = ( ) ( ) (4 13) In Equation 413, the is a class specific parameter vector and is individual characteristic s including income, age, etc. This class specific parameter shows the influence of indi vidual characteristic on the probability of being in a class s. The number of classes can be determined by the researcher but the class probabilities are subject to a statistical procedure rather than behavioral assumptions. To identify the optimal number of classes, measures of fit li ke AIC or Bayesian Information C riterion ( BIC ) are commonly used (Segebiel 2011) 4 .4 The Random Parameters Logit Model R andom parameter logit (RPL) model also refers to mixed logit model, mixed multinomial model or hybrid logit model We can assume that an individual ( i =1,,I) has a choice among g alternatives in each of h choice situation. Suppose i ndividual i consider s the full set of offered al ternatives in choice situation h with the highest utility the utility with each alternative g evaluated by each individual i in choice situation h is pre sented in Equation 414. U= X+ (4 14) where is the full vector of explanatory variables including attributes of the alternatives in choice situation h and are not observed directly by analyst and are treated as stochastic influences. is assumed to be identically independently distributed with extreme value distribution. We can rewrite the Equation 414 to indicate the Thus:
55 = + + = + + (4 15) where is a vector of individual characteristics is the associated parameter matrix and is a lower triangular matrix The random effect, is characterized by E  = 0 ) and Var [ ] = diag [ ] The mixed logit model assumed a general distribution for which can be normal, lognormal, uniform, or triangular. For example, the lognorm al model is = ( + + ) (4 16 ) where, is normally distributed. The unconditional probability that individual i will choose alternative g given the specific characteristics of choice set and the model parameters is expected value of the conditional probability. E quation 417 indicates that the probability density is introduced by the random component in the model f or The unconditional probability is ( , ) = ( \, ) ( \, ) (4 17) Where elements of are the parameters of the distribution of The ( \, ) is density function of given and The random variation in is derived by the random vector The random parameters define the degree of preference heterogeneity through the standard deviation of the parameters. When the sign of attribute has a theoretical expectation (e.g. we expected a negative sign on the cost /price parameters), the possibility of different distributional
56 assumption for attributes should be examined. It is important to explore the type of draws, the distributional assumptions, and multiple choice situation per individual. In other words, it is a first step to select the distribution of the random parameters. One can make use of different distribution on each attribute. The most popular and common distributions are normal, triangular, uniform, and lognormal distributions. For example, t he lognormal distribution is used if the attributes need to be a nonnegative /negative sign. A uniform distribution with a (0, 1) bounds is sensible when the model has dummy variables 4 .5 Estimation in Willingness to P ay Utility parameters can be estimated using the models discussed above and WTP can be calculated as a function of the preference parameters. If a nonprice attribute is continuous, or discrete and coded using dummy coding, then it is common to calculate t he consumers WTP for this attribute as the negative ratio of the attribute coefficient to the price (or cost) coefficient. First, assuming a linear random utility function, consumer utility (Vig) can be defined as = + + (4 18) Where Pig is the price of alternative g for person, is the marginal utility of price for person i is the nth attribute of alternative g for per son i is the marginal utility of the nth attribute, and is a stochastic disturbance of alternative g for person i N is the number of the alternative g
57 WTP for the nth attribute is the amount of money that consumer would be willing to pay to keep utility unchanged when nth attribute changed ( Gao and Schroeder 2009). Therefore, the WTP is the price premium that cons umers could pay for the change. We can derive the following equation: + + = ( + ) + + = ( ) (4 19 ) This part worth (or implicit price) formula represents the marginal rate of substitution between income and the attribute in question, i.e., the marginal welfare measure (willingness to pay or willingness to accept) for a change in an attribute. In most cases, the a is negative because of the relationship between price and quantity demand. Therefore, if the attribute increases the consumers utilities, ( ) is big ger than zero. It explains that the consumers would be willing to pay a premium for the quality improvement. For the mixed logit model, it is a challenge to select the distribution for an individual parameter. However, more people intend to focus on the ratio of random parameters, as in the derivation of WTP estimate For example, if two preference parameters have triangular distribution then the ratio of these parameters will have a discontinuous distribution wit h a singularity unless a r ange of the denominator is forced to exclude zero. The ratio of two normal distribution has the similar problems. To derive the WTP, the researchers can use all the information in the distribution or the mean and
58 st andard deviation. Suppose we have a mo del with a fixed cost parameter and an attribute whose parameter is normally distributed with mean and standard deviation The WTP for the attribute is distributed normally with and standard deviation However, this method ignored the sampling variance in the point estimates. For example, l et be the vector with elements We can produce a covariance matrix for all the estimated parame ters. A 3x3 symm etric matrix is to extract the part for (cal led it W). Therefore, we can draw random observations from the normal distribution which has mean and covariance matrix W. Next, bid curve can be estimated which is to estimate the changes in WTP for a change in explanatory variables, inc luding socioeconomic variables, such as, WTP = X + (4 20 ) X is a vector of explanatory variables such as income, age, gender and so on. is a vector of parameter and is an error term i n the Equation 420 The error term is assumed to have a normal distribution with zero mean and constant variance. To estimate a bid curve, the sign and size of estimated parameters are very important because the sign represents the direction between WTP and explanatory variables and size show the strength between WTP and explanatory variables.
59 Figure 41 F our stages of design of choice experiment was designed. Choice of experimental design Construction of experimental context Sampling and estimation strategy Selection of attributes and assignment of levels
60 CHAPTER 5 SURVEY DESIGN 5 .1 Choice Experiment D esign The goal of the research is to estimate the prefe rence of the people living in the U.S. with respect to the policies to promote solar energy. Six promotion policies such as ITC, rebate programs, production incentives, net metering, sales tax incentives, and property tax incentive are considered in this survey. The CEM also include three additional important factors that may affect people preference such as including the percentage of solar energy target by 2020, impact on electricity prices and impact on the job The percentage of solar energy in the energy sector is about 0.05% in 2015 which is levels of this attributes in the baseline. Besides the percentage of solar energy attribute in the baseline choice of this choice experiment is set to zero. If a choice experiment is designed with too many attributes, respondents could not make tradeoffs bu t instead utilize other strategies such as heuristics or lexicographic decision, which is not consist ent with the random utility theory, a key assumption in the choice experiment (Witt et al. 2009) Therefore, I divided the solar energy policies into two groups. Both groups include attributes such as the percentage of solar energy by 2020, impact on electric ity prices and impact on the j ob T he first group of solar energy policies includes ITC Rebate P rograms and Production I ncentives (see Table 51). The second group of solar energy policies includes Net Metering, Sale Tax Incentives and Property Tax I ncentives (see Table 5 2 ). The above method results in a small number of attributes in each CE M decreasing the task complexity and enable s a compensatory decision (Witt et al. 2009) Witt et al. (2009)
61 also concluded that blocked attribute desig ns may provide flexibility when it is not possible or desirable to reduce the number of attributes. The policies in group 1 help examine the impact of installation cost and income from producing solar electricity on consumer preference of solar energy. While the policies in group 2 help determine the impact of tax exemption and additional income from solar energy. For instance, t he ITC and rebate programs in group 1 offer the direct discount when people install a sol ar energy system. The production incentives enable solar energy installer to receive the additional income for electricity generate d from solar energy system For group 2, the property tax and sales tax exempt sale s tax and state property tax from installing a sola r system. The net metering allows the installer to receive additional income for excess electricity. The choice experiments were based on two groups of solar energy policies were designed using SAS software (Kuhfeld 2005) The information in Table 5 1 and 52 shows that the choice experiment includes six attributes, each having three levels. A full factorial design that can estimate all the main effect and higher order interaction effect will result in 3 = 729 attribute combinations or solar energ y policy profiles. However, it is impossible for an individual to evaluate all the 729 policy profile, and a subset of the profiles need to be selected. A fractional factorial design can be used to generate an experiment with a fewer number of policy profiles depending on the effects (main effect, two way interaction, three interactions etc.,) in which a researcher is interested. When using the fractional factorial design, it is important to construct efficient experiment so that the selected sub profiles can accurate estimate effect of each profile as much as possible.
62 For the group 1 policies I can get 13 optimal policy combinations (Table 53 ). T he D efficiency is 94.8 I randomly allocated the 13 policy profiles without replace ment to generate the Option B in the choice experiment. Then I pared the policy profiles in Option A and B to generate a choice experiment, which ensures orthogonality between attributes of a choice profile. Similarly, a choice experiment with the second g roup of solar energy policies is designed with the same method because the number of policies and the levels for corresponding policies is the same for the solar energy policies in groups one and two (Table 54). 5. 2 Questionnaire S tructure The questionnaire includes an informed consent and four other sections. The first page presents informed consent. Informed c onsent is a voluntary agreement for respondent s to agree that they fully understand and accept the survey purpose, benefit, and risks. The first section asks about basic demographics of respondents such as age, gender, race, level of education, annual income, and marital status as well as the size of the house, and whether the respondent owns or rents his home. The second section asks respondents attitude towards climate change and their major environmental concerns in the US. The third section asked respondents about their attitude toward solar energy which included whether the respondents believed that solar energy was a major source of alternative energy in the U.S. and whether the respondents had solar energy system or planned to install a solar energy system i n t he next two years. T his section also asked respondents the reasons that prev ent them from using a solar system and the benefit s that may motivate them to use solar energy systems. More specifically, this
63 section also asked whether respondents they installed solar electricity system or solar water heating system. The fo u rth section included the choice experiment that asked the respondents to choose their most favorite option from the three options that were described as solar energy policy profiles (Figure 51 and Figure 52). The first two options are the solar energy policy profiles generated from the choice experiment design and the last option is the status quo that only has the policy percentage of solar electricity target by 2 020 to equal to 0.5%, which is the percentage of electricity generated by solar energy in 2015. Before respondents start answering the choice experiment questions, we told them that the state government would like to increase the installation of solar el ectric capacity and want to know what are the most favorable solar energy policies preferred by general public. We also told them that all the choice ma de in the choice experiment may have a real impact on the development of solar energy policies and that will impact their electricity expenditure. Emphasizing the consequentiality of respondents choice in the survey may reduce the hypothetical bias and motivate respondents to provide more incentive comparable answer (Loomis 2014) We then presented the det ailed definitions of the six solar energy policy attributes and asked them to do a practice to calculate the potential cost saving if a policy attribute was set to a certain level (e.g. ITC=0.3). To make the practice realistic, we told them the calculation is based on rebate programs and production incentives, which were $1.0/watt and $0.15/kWh, respectively. The F igure 51 and 52 are the sample choice question for group 1 and group 2. The completion of the questionnaire was about 15 20 min utes
64 5.3 Data Collection and Descriptive S tatistics Toluna, an online survey company, delivered the surveys to its representative consumer panels in the U.S. in December 2015. A total of 614 completed q uestionnaires were collected. All t he 614 respondents (303 from group 1, and 311 from group 2) were over 18 years old and did not work for a market research, advertisi ng, or utility company We randomly assigned the survey respondents to one of two groups with different solar energy policies in the choice experiment. For t he f irst group, the choice experiment policies were ITC, Productive Incentive, and Rebate P rograms; for the second group, the choice experiment policies were Net M etering, Sale s Tax Incentives, and Property Tax I ncentives. All other parts of the two questionnaires were exactly the same. Because e ach respondent was presented with 13 choice set s, the choice experiment in the first and second groups result in 3,9391 and 4,043 valid observations, respectively. 1 13 choice set*311=3939
65 Table 51. Attribute description of group 1 Attributes Description Levels Value Percentage of solar energy Percentage of electricity in U.S. that comes from solar energy by 2020 3 0.5% 1.0% 1.5% ITC A tax credit for solar systems on residential and commercial properties 3 0% 30% 40% Rebate programs Provide the discount for solar energy installation 3 $0/W $0.2/W $0.4/W Production incentives Policies that pay residents for every kWh from solar electricity 3 $0/kWh $0.15/kWh $0.3/kWh Price Additional amount of electricity would have to pay relative to the baseline choice 3 $0 $11 $22 Job The new jobs opportunities created by solar energy industry 3 0 50,000 100,000
66 Table 52. Attribute description of group 2 Attributes Description Levels Value Percentage of solar energy Percentage of electricity in U.S. that comes from solar energy by 2020 3 0.5% 1.0% 1.5% Price Additional amount of electricity would have to pay relative to the baseline choice 3 $0 $11 $22 Net metering The installer to get the additional income for the excess electricity they generate from the solar system 3 $0/kWh $0.5/kWh $1.0/kWh Sales tax incentives The exemption from state tax for installing solar energy 3 0% 50% 100% Property tax incentives The exemption from state property tax for installing solar system 3 0% 50% 100% Job The new jobs opportunities created by solar energy industry 3 0 50,000 100,000
67 Table 53. Policy profiles for solar energy policies in group 1. Profile % of solar energy ITC Rebate programs Production incentives Electricity prices Job 1 1.5 0.4 0.0 0.15 11 100000 2 1.5 0.3 0.4 0.30 0 100000 3 4 1.5 0.3 0.0 0.00 22 50000 1.5 0.0 0.4 0.15 22 0 5 1.5 0.0 0.2 0.30 11 50000 6 1.0 0.4 0.2 0.00 0 0 7 1.0 0.3 0.4 0.15 11 50000 8 1.0 0.0 0.0 0.30 22 100000 9 0.5 0.4 0.4 0.30 22 50000 10 0.5 0.3 0.2 0.15 22 100000 11 0.5 0.3 0.0 0.30 11 0 12 0.5 0.0 0.4 0.00 11 100000 13 0.5 0.0 0.0 0.15 0 50000 Table 54. Policy profiles for solar energy policies in group 2. Profile % of solar energy Net metering Sale tax incentives Property tax incentives Electricity price Job 1 1.5 1.0 0.0 0.5 11 100000 2 1.5 0.5 1.0 1.0 0 100000 3 4 1.5 0.5 0.0 0.0 22 50000 1.5 0.0 1.0 0.5 22 0 5 1.5 0.0 0.5 1.0 11 50000 6 1.0 1.0 0.5 0.0 0 0 7 1.0 0.5 1.0 0.5 11 50000 8 1.0 0.0 0.0 1.0 22 100000 9 0.5 1.0 1.0 1.0 22 50000 10 0.5 0.5 0.5 0.5 22 100000 11 0.5 0.5 0.0 1.0 11 0 12 0.5 0.0 1.0 0.0 11 100000 13 0.5 0.0 0.0 0.5 0 50000
68 Which of the following policy combination do you prefer the most? Option A Option B Option C % of solar electricity by 2020 0.5% 1% 0.5% ITC 0% 30% 0% Rebate programs $2.0/W $2.0/W 0 Production incentives $0.0/kWh $0.15/kWh 0 Additional amount of electricity prices (per month) $11 $11 0 Impact on Jobs 100,000 50,000 0 Please choose your preferred option: Option A Option B Option C Figure 51 the Sample Choice Question for Group 1
69 Which of the following policy combination do you prefer the most? Option A Option B Option C % of solar electricity in 2020 1% 1% 0.5% Net metering $0/kWh $1.0/kWh 0 Sale tax incentives 0% 50% 0 Property tax incentives 100% 0% 0 Additional amount of electricity prices (per month) $22 $0 0 Impact on Jobs 100,000 0 0 Please choose your preferred option: Option A Option B Option C Figure 52 the Sample Choice Question for Group 2
70 CHAPTER 6 RESULTS 6.1 Sample Characteristics and O pinions The demographics of the respondents including gender, education, employment status, and household income be fore tax are reported in Table 6 1 Compared to the 2014 American Community Survey, the distribution of females and males of our samples (whole survey or those in group 1 or 2), was very close to that of the U.S. population. About 52% of the r espondents were female. The median household income was $50,000 to $74,999, consistent with the national median household income of $53,482. Our samples had a higher education level than the national average, mainly because that the survey was conducted online. Our sample had a larger proportion of respondents aged 18 to 45 than that of t he U.S. population (53% vs 47%). Our sample also ha s a slightly larger proportion of Whites than that of the U.S. population (80% vs 77%). Abo ut 75% of the respondents in our sample lived in town houses, not in apartment or condominium Finally, a bout 70 % of the respondents in our sample owned their home, not rent a home higher than homeownership rate (64%)1 in 2015 in the U.S 6.2 Views on Environmental Problems in U.S. Table 62 presents the statistics of the respondents environmental concerns. About 50% of the respondents indicated that they worried about natural disasters, including tsunamis, flooding, earthquakes, and drought. About 49% respondents were 1 Accessed September 26, 2016, available at http://www.census.gov/housing/hvs/files/currenthvspress.pdf
71 concerned about energy conservation issues, including renewable energy development, energy efficiency, and fossil fuel depletion. About 48% of respondents were also concerned about the contamination of drinking water. Conversely, respondents were least concerned wi th persistent extreme weather (about 35%). These results indicate that for the respondents, issues related to climate change and the development of renewable energy play an important role in public concerns with environmental issues. 6.3 T he Attitude T o ward Solar E nergy Table 63 shows the main reasons that prevent ed respondents from using solar energy. A bout 38% respondents said that the high installation cost was the most important impediment to using solar energy. An a dditional 20% stated that high installation cost is the important factor that prevent s them from using solar energy. Next, about 39% of the respondents considered the main reason that prevent s them from installing solar energy as the reliability of solar energy caused by weather and location of the house. Interestingly, the efficiency of the solar cell and increasing home insurance are not important factor s for the respondents not to install a solar energy system. These results demonstrated that installation costs are the most important reason preventing people from installing solar energy system, followed by the reliability of solar energy system that may be caused by weather and location of the house. Table 64 shows the motivations to use a solar energy system. A bout 42% of the respondents said that a reduced electricity bill is the most important motivator for using solar energy. An a dditional 26% stated that the reduced electricity bill is the important factor that motivate d them to use solar energy. A bout 59% of respondents stated that energy independence and reducing crude oil consumption are the most important or
72 important factor s to drive their use of sola r energy. About 56% also admitted that the environmental benefits are the most important or important factor s to prompt the use of solar energy. These results indicate that energy independence, reducing electric bills, and environmental benefit s are the key factors that motivated the respondents to adopt solar energy s ystem s. On the other hand, maintai ning the solar energy system was not an important factor that prevented respondents from adopting a solar energy system. Table 6 5 shows that about 13.2% of the respondents owned a solar energy system, including 7.5% who had a solar energy system and about 5.7% who had a solar energy and planned to install more in the next two years. About 22.8% of respondents planned to install solar energy in the next two years. However, more than 60% of respondents had no plan or did not know whether they would install solar energy in the next two years, which indicates that there are still strong barriers to prevent homeowners from installing solar energy system. 6.4 Description of Attributes in Choice Experiment and the D emographic Variables in Models Because each respondent made 13 choices in the choice experiment, the 303 and 311 respondents in the first and second groups result in 3939 and 4043 observations for regression analysis, respectively. All variables and their definitions used in the regression are reported in Table 6 6 The attributes variables incl uded Job, Price, Percentage of Solar Energy, Production Incentives, Rebate P rog rams, and ITC in the first regression; and Job, Price, Percentage of Solar Energy, Net Meterin g, Sales Tax I ncentives, and Property Tax I ncentives in the second regression. According to the
73 previous studies discussed consumers preference towards renewable energy, we included six demographic variables as potential factors that may affect respondents preference for solar energy policy including gender (Batley et al. 2001) whether household income was more than $100,000/year (Shen and Saijo 2009; Batley et al. 2001) age (Yoo and Ready 2014; Batley et al. 2001) whether the respondent had children (Ek 2005) whether the respondent received a bachelors degree (Ek 2005) and climate change awareness (Yoo and Ready 2014) Based on our pilot survey, we also consider ed two additional demographic variables, whether household income was less than $35,000 /year and whether a respondent was renting a house. 6 .5 MNL and MNL IN Estimation R esults 6 .5 .1 MNL Estimation R esults The first models esti mated were MNL model s for groups 1 and 2 The results of group 1 are shown in Table 67 The sign of the coefficient represents the influence of the attributes on the probability of choosing an option with certain attributes. For instance, a positive sign of an attribute indicates that the presence/increase of that attribute will increase respondents probability of choosing an option carrying that attributes More specifically, t he coefficient s of an attribute represent the marginal indirect utility of that attribute. For group 1, all attributes have the expected signs. The sign of P ercentage of solar energy and Job attributes are both positive and statistically significant at the 1 % level indicating that new job opportunities created by the solar energy industry and a higher percentage of solar electricity target by 2020 increase respondents preference for solar energy. On the other hand, the coefficient of R ebate programs and ITC attributes were statistically significant at the 1 % level and the
74 coefficient of the Production incentives attr ibute was stati stically significant at the 10% level These results indicate that th e reducing solar energy syst em installation cost or generating additional income for solar system installers can increase respondents adoption of solar energy system The results of group 2 are shown in Table 68 For group 2, all attributes have the expected signs. The sign of P ercentage of solar energy and Job attributes are also positive and statistically significant at the 1% level meaning that new job opportunities created by the solar energy industry and a higher percentage of solar electricity target by 2020 increase respondents preference for solar energy. The coefficient s of Net metering, Sale s tax incentives and P roperty tax incentives attributes were statistically significant at the 1 % level These results show that decreasing sales tax and property tax or earning additional income for solar energy installers also can promote the respondents adoption of the solar system 6 .5 .2 MNL IN estimation result s The MNL IN mo del s are usually estimated as an alternative to MNL model s to reveal the preference heterogeneity (Yoo and Ready 2014) In th e MNL IN model, the respondent s characteristic s and attitudinal variables including gender, age, climate change awareness, higher income household, lower income household, having children, education, renting a house and having a solar energy system interact with choice experiment attribute variables T able 6 9 shows the estimated coefficients fro m the MNL IN model for group 1.
75 W e fit the choice model with a different coefficient on every attribute for the demographic variables in this study Therefore, we can create the interaction variables for group 1 The sign of P ercentage of solar energy is positive and statistically significant at the 1 % level, indicating that a higher percentage of solar electricity target by 2020 increase respondents preference for solar energy. The coefficient of P ercentage of solar energy *a ge is negative and statistically significant at the 10% level, pointing out that a higher percentage of solar electricity target by 2020 increases young respondents preference for solar energy. For the ITC attribute t he coefficient of ITC climate change awareness i s positive and statistically significant at the 5% level, suggesting that increasing ITC promote s respondents who concern climate change to adopt solar energy. For the Rebate programs attribute, the coefficient of Rebate programs*Having a solar energy system is negative and statistically significant at 5% level, indicating that increasing Rebate programs will impede the respondents who already have solar energy systems to further use solar energy For the Job attribute, the coefficient of Job*climate change awareness is positive and statistically significant at the 1% level, meaning that the new job opportunities created by solar sector promote respondents who are concern ed about climate change t o adopt solar energy. T he coefficient of Job*renting a house is negative and statistically significant at the 5% level, indicating that the new job opportunities created by solar sector promote respondents who own a house to adopt solar energy. The coeffi cient of Job*having a solar energy system is positive and statistically significant at the 1% level, indicating that the new job opportunities created by solar sector promote respondents who have solar energy systems to further adopt solar energy.
76 Table 6 10 shows the MNLIN estimate results for group 2. The respondents characteristics and nine attitudinal variables such as gender, age, climate change awareness, higher income household, lower income household, having children, education, renting a house and having a solar energy system interact with choice experiment attribute variables For the Percentage of solar energy attribute, t he coefficient of the percentage of solar energy *Rent ing a house is negative and statistically significant at the 5 % level indicating that r espondents who own a house care more about a higher percentage of solar electricity target by 2020 than respondents w ho rent a house. T he coefficient of P ercentage of solar energy *c limate change awareness is positive and statistically significant at the 5% level, implying that the respondents who concern about climate change attach great value to a higher percentage of solar electricity target by 2020 than respondents who do not concern about climate change. For the Property tax incentives attribute t he coefficient of Property tax incentives*high income household is positive and statistically significant at the 1% level, meaning that reducing property tax motivate the respondents who have a higher income (over $100000/year) to adopt solar energy. The coeffici ent of the Property tax incentives*age is positive and statistically significant at the 5% level, meaning reducing property tax increase the respondents preference to adopt solar energy. Finally, th e coeffici ent of Property tax incentives* having children is positiv e and statistically significant at the 1% level including that reducing property tax motivate the respondents who have children to adopt solar energy.
77 For the Net metering att ribute, the coefficient of Net metering*c limate change awareness is positive and statistically significant at the 1% level ; raising the cap on the solar net metering increases the preference for respondents with climate change awareness to use solar energy. The coefficient of Sales tax incentives* Having a solar energy system is negative and statistical ly significant at the 5% level implied that reducing sales tax incentives can increase the preference for respondents with having a solar energy system to further use solar energy. For Job attribute, the coefficient of Job*climate change awareness is positive and statistically significant at the 10 % level suggesting that new job opportunities created by solar energy industry increase the probabili ty of respondents with climate change awareness to use solar energy. T he coefficient of Job*Lower income household is negative and statistically significant at the 1% level, explains that new job opportunities created by solar energy industry cannot incre ase low income respondents preference for solar energy. 6 .6 RPL model Estimation R esult s The first step to estimate a RPL model is to determine how many draws to use for simulation. Hensher and Green (2009) stated that 500 draws were more than necessary; f or this study, 500 Halton draws are used to obtain stable estimates for RPL model The second step is to determine which variables are random. Common practice is to estimate a RPL model in which coefficients of all attributes are random and check whether the standard deviation of random coefficients is statistically significant or not (Yoo and Ready 2014) It is common to use a normal distribution for the random coefficients of the non price attributes (Shen 2009; Kosenius 2010; Greene and
78 Hensher 2003) to estimate the RPL model. In this study, we suppose that the coeffi ci ents of all non price attributes are normal ly distributed and the price coefficient has a lognormal distribution. The reason for using the normal distribution is that consumers could have either a positive or negative preference for the nonprice coefficients. On the other hand, economic theory predicts that price coefficient should be negative. Table 611 show s the estimated results from the basic RPL model for group 1. The sign of percentage of solar energy and Job attributes are both positive and statistically significant at the 1% and 5% level, indicating that new job opportunities created by the solar energy industry and a higher percentage of solar electricity target by 2020 increase respondents preference for solar energy. In addition, the coefficient s of Rebate programs ITC Production incentives attributes w ere statistically significant at the 1% level. These results indicate that the reducing solar energy syst em installation cost or generating extra income for solar system installers can increase respondents adoption of solar energy system. The s tand ard deviation parameters of Percentage of solar energy Rebate program s, ITC, Job and Price attributes are also statistically significant which indicates heterogeneity in respondent s preference for these attributes. Table 612 report s the results of a RPL C model that further considered the correlation between every attribute for group 1. This model is statistically significant (Chi square value equal to 1010.686 with 28 degrees) I n comparison to the Basic RPL model for group 1 we co nclude that the RPL C model is better (the log likelihood ratio test produces a Chi square value equal to 636.62 (i.e. 2*( 4224.54 ( 3906.23))).
79 The sign of P ercentage of solar energy and Job attributes are both positive and statistically significant at the 1% and 5% level, indicating that new job opportunities created by the solar energy industry and a higher percentage of solar electricity target by 2020 increase respondents preference for solar energy. T he coefficient s of Rebate programs ITC, and Production incentives were statistically significant at the 1% level. These results indicate that the reducing solar energy system installation cost or generating extra income for solar system installers can increase respondents adoption of solar energy system. Th ese results are consistent with the basic RPL model. For the covarianc e of a random variable, a positive covariance implies that larger parameter estimates for individuals along the distribution on one attribute are generally associated with larger estimates for that same individual in the parameter space for the second attribute. For example, t he coefficient in t he covariance of random parameters between Job and P ercentage of solar energy attributes is negative and statistically significant at the 5% level, indicating that if respondents pay more attention to Job created by solar industry they could care less about the goal for solar energy installation Similar, if respondents care about the goal for solar energy installation, their support for solar energy is less affected by the reb ate programs and ITC Table 6 13 show s the estimated coeffi cient from the basic RPL model for group 2 The sign of P ercentage of solar energy attribute is positive and statistically significant at the 1% level, indicating that a higher percentage of solar electricit y target by 2020 increases respondents preference to install a solar energy system The sign of Net metering, P roperty tax incentives and S ales tax incentives attributes is positive and
80 statistically significant at the 1% level, meaning reducing property tax or sales tax or increasing the cap on net metering promote respondents to use solar energy. The s tandard deviation parameters of Percentage of solar energy Net metering, Property tax incentives Sale s tax incentives, Job and Price attributes are also statistically significant at the 1% level indicating heterogeneity in respondent s prefer ence for these attributes. In Table 6 14 w e report the results of a RPL C model output for group 2. The model is statistically significant (Chisquare value equal to 2140.94 with 28 degrees). I n comparison to the Basic RPL model for group 2 we co nclude that the RPL C model is better (the log likelihood ratio test produces a C hi square value equal to 1486.24 (i.e. 2*( 4194.89( 3451.77))). The coefficient estimates of every attribute, such as Percentage of solar energy Net metering Property tax incentives and S ales tax incentives are consistent with these in a basic RPL model. I n other words, these attributes are both positive and statistically significant, implying that thes e attributes are associate d with a higher probability of supporting the solar energy Furthermore, Property tax incentives has the highest impact on the utility of every respondent of group 2 for this experiment For the covariance of a random variable in group 2 the coefficient in the covariance of random parameters between Property tax incentives and P ercentage of solar energy attributes is positive a nd statistically significant indicating that if respondents pay more attention to property tax incentives they could care less about the goal f or solar energy installation. I f respondents care about the goal for solar energy installation, they are more willing to support sales tax incentives On the other
81 hand, if the respondents care about the job created from the solar industry, they also support the property tax incentives and sales tax incentives. 6 .7 LCM Model Estimation Re sult s Compared with the MNL model, the LCM model allows us to simultaneously analyze the preference heterogeneity and estimate the probability of a respondents choice After comparing the AIC and average class probabilities in different LCM we decided to use a 3 class LCM not 2or 4 cl ass LCM model s. Table 615 shows the results of 3 class LCM model for group 1. The overall model is significant. Average class probabilities are 23.1% for class 1, 54.1 % for class 2 and 22.8 % for class 3. For class 1 the coefficient of Job is positive and statistically significant at the 1% level, indicating the new job opportunities created by solar energy industries increases respondents preference for solar energy. T he coefficients of P roduction incentives, R ebate programs, and ITC are also positive and statistically significant at the 1% level, implying that reducing solar energy installation cost or generating additional income for solar system installers can increase respondents adoption of solar energy system Therefore, class 1 was named as Prosolar energy. On the other hand, only the coefficient of the R ebate program s attribute is positive and statistically significant in class 2 indicating that only reducing solar energy installation cost can increase respondents adoption of solar energy In class 3 t he coefficient of the Job attribute is negative and statistically significant at the 1% level indicating that new job opportunities created by solar energy industry decrease the respondents preference to use the solar energy.
82 For the socioeconomic characteristic s, the fixed parameters were set f or class 3 therefore, the parameters of the other two classes should be interpreted in relation to the third class. In class 1, the coefficient of Climate change awareness was significant at 5 % level, indicating that the respondents with climate change awareness are more likely to be classified into class 1 In cla ss 2 the coefficient of Age was negative and significant at the 1% level implying that younger respondents are more li kely to be classified into class 2. By comparing with the AIC and average class probabilities in LCM for group 2, we decided to use a 3 class LCM not 2 or 4 class LCM model s. T ab le 6 16 shows the result of class membership for gro up 2 in the 3 class LCM model. T he overall model is significant. Average class probabilities for the 3 class are 55.6% for class 1 29.7% for class 2 and 14.7% for class 3. For class 1, the coefficient of Job and Percentage of solar energy attributes are statistically significant and positive at the 1% significance level, indicating that the new job opportunities created by the solar energy industry and a higher percentage of solar electricity target by 2020 increase respondents preference for solar energy. Net metering, Sales tax incentives and Property tax incentives attributes are also statistically significant and positive at the 1 % level meaning that decreasing sales tax and property tax or earning additional income for solar energy can promote the respondents adoption of the solar system. Therefore, class 1 was named as Pro solar energy. On the other hand, the coefficient of Job attribute is statistically significant and negative at the 10% level in class 2 showing that new job opportunities created by solar energy industry cannot increase respondents preference to adopt the
83 solar energy. In class 3 only the coefficient of Price attribute is statistic ally significant at the 1% level therefore it was named as No solar energy class. For class 1 the coefficient of L ower income household was statistically significant and negative at the 5% level and the coefficient of having children was statistically significant and positive at the 5% level This indicate s that low income households are less like to be Pro solar energy while those who had children are more like in class 1. For the second class, the coefficient of Age was negative and significant at the 10% level, which shows that younger respondents are more likely to be classified into class 2 6.8 Measures of F it Table 6 17 and 618 contrast the measures of fit of five models, including the MNL, MNL IN, RPL(basic), RPL C, and LCM models. In group 1, the MNL model fit the data worst because of it had the highest value o f AIC and highest absolute value of log likelihood at convergence compared with the other four models. The two models that fit the data best are the RPLC and LCM models; the LCM model has a slightly lower AIC value and a lower absolute value of log likelihood value at convergence than the RPLC model. T he RPL model fit the data much worse than the RPLC model which implies that a RPL without the correlation of attributes is not sufficient to obtain full preference heterogeneity. In group 2, the MNL model still fit the data worst because it had the highest AIC value and highest absolute value of log likelihood at convergence compared to the other four models. The two models that fit the data best are the RPLC and LCM models. The RPL C model has a slightly lower AIC value and a lower absolute value of l og likelihood
84 value at convergence than the LCM model, which implies that the RPLC model fit the data much better than the LCM model in group 2. 6 .9 Me an WTP V alue For m odel s of group 1, we estimated the mean and standard deviation of WTP (Table 6 19 and 620) for the MNL, MNL IN RPL C and LCM model s. The WTP is the money that the respondents can give up for an improvement in the quality of a product or service For instance, a new job opportunity cr e ated by the solar energy industry by one unit allows an increase in the monthly electricity bill by $ 24.375 per person/month ( 95% C.I.= $19.586$29.164) according to the RPL C model. The mean WTP values for the LCM and RPLC model s are estimated with the same method but based on individual parameter estimates. Compared with the mean WTP values of three policies including ITC, rebate programs, and production incentives in group 1, the mean WTP for ITC is relatively large in the PRL C For example, the mean WTP values in RPL C model was $24.110 (95% C.I.= $21.568 $26.652) for ITC indicating about 2 .5 times higher than rebate programs and 4 times higher than production incentives. For LCM model, the mean WT P value in class 1 was $78.779 (95% C.I.= $78.104$79.453) for ITC, showing 2.4 times higher than production incentives and 4.1 times than rebate programs. T he standard deviation of mean WTP in the RPL C are larger than those of LCM in every class To get more insight into the distribution of WTP values, the kernel densi ties of the WTP are plotted in F igure s 6 1 to 6 6 The purpose of kernel distribution is to reveal the distribution of a parameter nonparametrically without considering the assumption of distribution (Greene and Hensher 2003). These figures describe the WTP distribution for attributes, including policies to
85 promote solar energy by kernel density plots. The density plot is a very useful method to explore the distribution of WTP for each attribute (Gree ne and Hensher 2003). We can build the empirical shape of each distribution. This information can guide the analyst to receive the domain of this function. For the group 1, the WTP distribution for production incentives and ITC show bimodal distributio n and t he former has a much large peak. The WTP distribution for Job shows a much spread out distribution compared with other attributes in group1. For models of group 2, we estimated the mean and s tandard deviation of WTP (Table 6 21) for the MNL, MNLIN, and RPL C. Table 622 shows the means and standard deviations of the mean WTP values for the LCM model from every class For WTP, a new job opportunity cr e ated by solar energy industry by one unit allows an increase in the mo nthly electricity bill by $ 16.719 (95% C.I.= $ 14.240 $19.198) according to the RPL C model. For the RP L C model s, the mean WTP values for property tax incentives is relative ly large r compared to other attributes. For inst ance, the mean WTP values in RPL C model was $24.691 (95% C.I.= $22.511$26.871) for property tax incentives, indicating 1.16 times higher t han sales tax incentives and 2.8 times higher than net metering. The LCM model for every class shows a smaller standard deviation than the RPL C mode l In class 1, t he mean WTP value for property tax incentives is $ 48.937 (95% C.I.= $48.919--$48.995) higher than sales tax incentives and net metering. For class 2, the mean WTP value for sales tax incentives is $2.075 (95% C.I.= $1.861-$2.290), higher than property tax incentives and net metering. For class 3, property tax
86 incentives also have a relatively large mean WTP value compared to sales tax incentives and net metering. For the distribution of WTP values, the kernel densities of th e WTP are plotted in F igure s 6 7 to 6 12. These figures describe the WTP distributions for attributes, including policies to promote solar energy by kernel density plots. Further, the individual WTP for every attribute is a unimodal distribution. 6 .10 Estimate a Bid C urve In this section we estimate the bid curve for the WTP from the RPL C model for grou p s 1 and 2. Table 623 shows the estimation of the bid curves using OLS for group 1 The reason we make use of OLS is that the coefficient estimators are best linear unbiased estimators (BLUE) if the error has zero expected value and constant variance. T he coefficients of age are positive and statistically significant for P ercentage of Solar E nergy attribute and the Production incentives attribute, meaning that the older respondents are will ing to pay an extra $0.075 per month to increase the percentage of solar electricity target by 2020 and $0.04 per month to support the production incentives. The coefficients of gender are negative and statistically significant in both equations for production incentives attribute, implying that the male respondents are willing to pay an extra $0.991 per month to support production incentives and $2.298 per month to support the rebate programs The coe fficient s of rent ing a house are negative and statistically significant for P ercentage of solar energy attribute, P roduction incentives attribute, Rebate program attribute and Job attribute. T his implies that respondents who own a house are willing to pay an extra $3.602 per month to increase the percentage of solar electricity target by 2020, $1.423 per month to support
87 production i ncentives, $2.614 per month to supp ort rebate programs, and $12.509 per month to support new job opportunities by solar energy industry For climate change awareness the respondents who believe in clim ate change will pay an extra $0.748 per month t o support production incentives and $7.078 per month to support new job opportunities created by the solar energy industry Comparing across the columns of estimates in Table 623, renting a house is the most important determinant of WTP for group 1. This result is expected as house renter wont receive many benefits from pol icies that promote solar energy. Similarly, Table 62 4 shows the estimation of the bid curves using OLS for group 2. The parameters of the solar sourc e are negative and statistically significant for the S ale s tax incentives attr ibute, meaning that the respondents living at low solar radiation are willing to pay an extra $0.023 per month to support sales tax incentives The coefficients of higher income household are positive and statist ically significant, meaning that high income respondents are willing to pay an extra $14.524 per month to supp ort property tax incentives.
88 Table 61. Summary statistics of survey respondents Variable Description Group 1 (%) Group 2 (%) Whole (%) U.S. Population Gender (%) 1 if female 0 if male 51.82 48.18 50.80 49.20 51.63 48.37 49% 51% Age (%) 18 to 24 25 to 34 35 to 44 45 to 54 55 to 64 More than 65 14.85 19.47 19.14 13.20 15.18 18.15 11.25 22.51 18.97 15.43 16.40 15.43 12.87 21.01 19.06 14.66 15.80 16.61 18 to 45 47% 45 to 64 34% Education (%) Less than high school High school Some college Associates degree Bachelors degree Masters degree Professional degree Doctorate degree 3.63 19.81 25.41 13.86 25.08 7.92 2.64 1.65 2.25 14.47 23.47 10.29 30.23 12.54 4.50 2.25 3.09 17.10 24.59 12.05 27.52 10.26 3.58 1.79 High school or higher 86% Bachelors degree and higher 29% Household Income (per year) (%) Under 19,999 20,00024,999 25,00049,999 50,00074,999 75,00099,999 100,000124,999 125,000149,999 150,000174,999 175,000199,999 200,000249,999 250,000499,999 More than 500,000 Prefer not to answer 15.18 14.52 16.17 18.48 15.18 5.61 2.64 0.99 0.99 1.65 0.99 0.66 6.93 9.32 18.33 17.36 19.61 14.47 6.75 1.93 3.22 1.29 1.61 2.57 0.64 2.89 12.38 16.45 16.78 19.06 14.66 6.19 2.28 2.12 1.14 1.63 1.79 0.65 4.89 Median Household income (per year) is 53,482 Employment Status (%) Full time job Part time job Retired Student Unemployed Others 35.31 11.88 20.46 7.26 19.14 5.94 45.98 11.90 17.68 3.86 15.43 5.14 40.88 11.73 19.06 5.54 17.26 5.54 In labor force: 50.7% Unemployed 6.9%
89 Table 61. Continued Variable Description Group 1 (%) Group 2 (%) Whole (%) U.S. Population Race (%) American Indian Asian Black or African American White Native Hawaiian Other 1.65 6.27 5.94 81.19 0.99 3.96 1.29 6.11 10.29 77.81 0.32 4.18 1.47 6.19 8.14 79.64 0.65 3.91 White 77.4% Black or African American 13.2% Live in apartment or house (%) Own or rent your home (%) Apartment/Condominium House Own Rent 25.51 74.49 69.64 30.36 25.72 74.28 69.77 30.23 25.57 74.43 69.71 30.29 64%
90 Table 62 Environmental concerns that people worry about in the U.S. Frequency* Percent (%) Natural disasters, including tsunamis, flooding, earthquakes and drought 306 50 Contamination of drinking water 295 48 Incurable pollution (e.g., air and water pollution and soil contamination) 276 45 Biological pollution, including bacteria, viruses, molds, mildew, dander, mites, pollen, ventilation and infection 265 43 Persistent extreme weather (e.g., heat and cold waves) 215 35 Energy conservation issues, including renewable energy development, energy efficiency, and fossil fuel depletion 303 49 Land mismanagement, including urban sprawl, lack of free space, and habitat destruction and fragmentat ion 260 42 *Number of the total sample is 614
91 Table 63. The reasons that prevent people from using solar energy in 2015. Frequency Percent (%) Cleaning the solar panel is hard The most important 77 13 Important 136 22 Neutral 227 37 Not important 87 14 The least important 87 14 Weather and location of house could affect the reliability of solar energy The most important 101 16 Important 139 23 Neutral 202 33 Not important 96 16 The least important 76 12 I dont really know how a solar system really work The most important 63 10 Important 75 12 Neutral 246 40 Not important 98 16 The least important 132 21 Installing a solar system requires a large area of house/roof The most important 73 12 Important 132 21 Neutral 226 37 Not important 101 16 The least important 82 13 Installation cost is too high The most important 231 38 Important 123 20 Neutral 155 25 Not important 52 8 The least important 53 9 It could increase home owners insurance The most important 57 9 Important 124 20 Neutral 252 41 Not important 89 14 The least important 92 15 It is not efficient enough The most important 68 11 Important 103 17 Neutral 258 42 Not important 93 15 The least important 92 15
92 Table 64 The motivation to use the solar energy systems Frequency Percent (%) When solar panels are installed, they dont need to be switched out or repaired it. The most important 114 19 Important 150 24 Neutral 246 40 Not important 71 12 The least important 33 5 Solar energy provides energy independenc e and reduces crude oil consumption. The most important 161 26 Important 204 33 Neutral 163 27 Not important 57 9 The least important 29 5 Environmental benefits The most important 173 28 Important 172 28 Neutral 181 29 Not important 48 8 The least important 40 7 Additional a source of income from selling excess energy The most important 114 19 Important 178 29 Neutral 222 36 Not important 59 10 The less important 41 7 Reduced electric bills The most important 258 42 Important 159 26 Neutral 125 20 Not important 44 7 The less important 28 5 Reliable energy source The most important 146 24 Important 159 26 Neutral 199 32 Not important 66 11 The least important 44 7
93 Table 65. Willingness to use the solar energy system Frequency Percent (%) I have a solar energy system now 46 7.5 I have a solar energy system and plan to install more in the next two years 35 5.7 I do not have a solar energy system and plan to install in the next two years 140 22.8 I do not have any solar energy system and do not plan to install it in the next two years I do not know 289 104 47.1 16.9
94 Table 66 Variable definition of the individual characteristics Variable Description Job Impact on job Price Additional amount of electricity prices (per month) Production incentives Policy that pay residents for every kWh from solar electricity Rebate programs A discount for solar energy installation ITC A tax credit for 30% of the cost of a residential and commercial solar system Percentage of solar energy Percentage of solar electricity by 2020 Net metering Net metering lets the customers who generate their own electricity from solar energy to feed electricity they do not use back into the grid. Sales tax incentives Exemption from state sales tax for an installing solar syst em. Property tax incentives Exemption from state property tax for installing a solar system. Gender 1 if the gender is female; zero otherwise Higher income household 1 if the household income before tax is larger than $100,000; zero otherwise Lower income household 1 if the household income before tax is smaller than $35,000; zero otherwise Age Age Renting a house 1 if the respondents rent a house; zero if respondent own a house Education 1 for bachelor s degree or higher; zero otherwise Climate change awareness 1 if the respondent believes that climate change is an issue; zero otherwise Having children 1 if the respondent has children; zero otherwise Having a solar energy system 1 if the respondent has a solar energy system; zero otherwise
95 Table 67 Estimation results of the MLN model for group 1. Attributes Coeff. Std. err. Sig. Percentage of solar energy 0.168 0.033 *** Production incentives 0.055 0.030 Rebate programs 0.108 0.024 *** ITC 0.233 0.034 *** Job 0.089 0.037 *** Price 0.007 0.003 ** No. of persons 303 No. of observations 3939 Log likelihood 4224.54 Pseudo R 2 0.0095 AIC 8463.1 Note: Sig. = Significance level: ***0.01, **0.05, *0.1, Table 68 Estimation results of the MNL model for group 2. Attributes Coeff. Std. err. Sig. Percentage of solar energy 0.127 0.025 *** Net metering 0.096 0.030 *** Property tax incentives 0.229 0.034 *** Sales tax incentives 0.227 0.026 *** Job 0.145 0.032 *** Price 0.019 0.003 *** No. of persons 311 No. of observations 4043 Log likelihood 4194.89 Pseudo R 2 0.0143 AIC 8403.8 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
96 Table 69. MNL IN estimate results for group 1. Attributes Coeff. Std. err. Sig. Percentage of solar energy 0.316 0.120 *** Production incentives 0.047 0.114 Rebate programs 0.190 0.098 ITC 0.061 0.111 Job 0.001 0.114 Price 0.008 0.003 ** Percentage of solar energy*Gender 0.026 0.062 Percentage of solar energy*Age 0.003 0.002 Percentage of solar energy*Climate change awareness 0.057 0.060 Percentage of solar energy*Higher income household 0.071 0.093 Percentage of solar energy*Lower income household 0.031 0.069 Percentage of solar energy*Having children 0.074 0.062 Percentage of solar energy*Education 0.004 0.066 Percentage of solar energy*Renting a house Percentage of solar energy*Having a solar energy system 0.065 0.055 0.068 0.092 ITC*Gender 0.041 0.056 ITC*Age 0.001 0.002 ITC*Climate change awareness 0.111 0.055 ** ITC*Higher income household 0.057 0.084 ITC*Lower income household 0.052 0.063 ITC*Having children 0.050 0.056 ITC*Education 0.057 0.060 ITC*Renting a house 0.073 0.062 ITC*Having a solar energy system 0.086 0.085 Rebate programs*Gender 0.045 0.051 Rebate programs *Age 0.004 0.002 ** Rebate programs *Climate change awareness 0.031 0.050 Rebate programs *Higher income household 0.074 0.077 Rebate programs *Lower income household 0.071 0.057 Rebate programs *Having children 0.052 0.051 Rebate programs *Education 0.038 0.054 Rebate programs *Renting a house 0.056 0.055 Rebate programs*Having a solar energy system 0.171 0.075 ** Production incentives*Gender 0.017 0.059 Production incentives *Age 0.000 0.002 Production incentives *Climate change awareness 0.069 0.057 Production incentives *Higher income household 0.006 0.088 Production incentives *Lower income household 0.009 0.066
97 Table 69 Continued Attributes Coeff. Std. err. Sig. Production incentives *Having children 0.033 0.058 Production incentives *Education 0.007 0.063 Production incentives *Renting a house 0.009 0.064 Production incentives*Having a solar energy system 0.072 0.088 Job *Gender 0.074 0.057 Job *Age 0.000 0.002 Job *Climate change awareness 0.225 0.056 *** Job *Higher income household 0.047 0.087 Job *Lower income household 0.055 0.065 Job *Having children 0.038 0.057 Job *Education 0.025 0.062 Job *Renting a house 0.156 0.063 ** Job *Having a solar energy system 0.361 0.085 *** No. of persons 303 No. of observations 3939 Log likelihood 4156.37 Pseudo R2 0.0255 AIC 8416.7 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
98 Table 610. MNLIN estimate results for group 2. Attributes Coeff. Std. err. Sig. Percentage of solar energy 0.170 0.107 Net metering 0.047 0.121 Sales tax incentives 0.189 0.113 Property tax incentives 0.071 0.113 Job 0.116 0.132 Price 0.020 0.003 *** Percentage of solar energy*Gender 0.075 0.051 Percentage of solar energy*Age 0.001 0.002 Percentage of solar energy*Climate change awareness 0.107 0.050 ** Percentage of solar energy*Higher income household 0.086 0.068 Percentage of solar energy*Lower income household 0.053 0.058 Percentage of solar energy*Having children 0.067 0.051 Percentage of solar energy*Education 0.047 0.051 Percentage of solar energy*Renting a house 0.119 0.056 ** Percentage of solar energy*Having a solar energy system 0.039 0.075 Net metering*Gender 0.078 0.058 Net metering *Age 0.001 0.002 Net metering *Climate change awareness 0.164 0.057 *** Net metering *Higher income household 0.009 0.079 Net metering *Lower income household 0.084 0.067 Net metering *Having children 0.042 0.058 Net metering *Education 0.018 0.059 Net metering *Renting a house 0.024 0.064 Net metering* Having a solar energy system 0.076 0.088 Sales tax incentives*Gender 0.048 0.053 Sales tax incentives *Age 0.001 0.002 Sales tax incentives *Climate change awareness 0.003 0.052 Sales tax incentives *Higher income household 0.011 0.071 Sales tax incentives *Lower income household 0.043 0.062 Sales tax incentives *Having children 0.048 0.053 Sales tax incentives *Education 0.051 0.054 Sales tax incentives *Renting a house 0.023 0.058 Sales tax incentives *Having a solar energy system 0.183 0.081 ** Property tax incentives*Gender 0.010 0.052 Property tax incentives *Age 0.004 0.002 **
99 Table 610. Continued Attributes Coeff. Std. err. Sig. Property tax incentives *Climate change awareness 0.084 0.051 Property tax incentives *Higher income household 0.303 0.070 *** Property tax incentives *Lower income household 0.041 0.059 Property tax incentives *Having children 0.147 0.052 *** Property tax incentives *Education 0.033 0.053 Property tax incentives *Renting a house 0.031 0.057 Property tax incentives *Having a solar energy system 0.049 0.079 Job *Gender 0.034 0.062 Job *Age 0.001 0.002 Job *Climate change awareness 0.118 0.061 Job *Higher income household 0.157 0.082 Job *Lower income household 0.212 0.071 *** Job *Having children 0.061 0.062 Job *Education 0.061 0.062 Job *Renting a house 0.099 0.068 Job Having a solar energy system 0.069 0.093 No. of persons 311 No. of observations 4043 Log likelihood 4088.21 Pseudo R2 0.04 AIC 8280.4 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
100 Table 611. Basic RPL estimation results for group 1. Coeff. Std. err. Sig. Variable Percentage of solar energy 0.229 0.043 *** Production incentives 0.099 0.037 *** Rebate program 0.113 0.035 *** ITC 0.312 0.047 *** Job 0.160 0.070 ** Price 4.574 0.484 *** Standard deviation parameter Goal STD 0.157 0.078 ** Production incentives STD 0.099 0.123 Rebate programs STD 0.191 0.054 *** ITC STD 0.380 0.055 *** Job STD 0.660 0.053 *** Price STD 1.705 0.477 *** No. of persons 303 Log likelihood 3906.23 Chi squared (13 d.f.) 842.407 Pseudo R 2 0.0973 AIC 7838.5 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
101 Table 612. RPL C estimated result for group 1. Coeff. Std. err. Sig. Variable Percentage of solar energy 0.241 0.048 *** Production incentives 0.105 0.039 *** Rebate programs 0.107 0.036 *** ITC 0.319 0.048 *** Job 0.139 0.061 ** Price 4.642 0.513 *** Standard deviation parameter Percentage of solar energy STD 0.214 0.065 *** Production incentives STD 0.061 0.155 Rebate programs STD 0.125 0.098 ITC STD 0.353 0.255 Job STD 0.534 0.337 Price STD 1.617 0.995 Covariance of random parameters Production incentives: Percentage of solar energy 0.005 0.015 Rebate programs: Percentage of solar energy 0.026 0.016 Rebate programs: Production incentives 0.004 0.049 Job: Percentage of solar energy 0.092 0.039 ** Job: Production incentives 0.002 0.095 Job: Rebate programs 0.053 0.245 ITC: Percentage of solar energy 0.064 0.027 ** ITC: Production incentives 0.011 0.100 ITC: Rebate programs 0.037 0.425 ITC: Job 0.072 1.724 Price: Percentage of solar energy 0.273 0.169 Price: Production incentives 0.009 0.249 Price: Rebate programs 0.158 0.718 Price: Job 0.858 2.219 Price: ITC 0.200 4.521 No. of persons 303 Log likelihood 3822.09 Chi squared (28 d.f.) 1010.686 Pseudo R 2 0.1168 AIC 7700.2 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
102 Table 613. Basic RPL estimation results for group 2. Coeff. Std. err. Sig. Variable Percentage of solar energy 0.173 0.041 *** Net metering 0.135 0.055 *** Property tax incentives 0.319 0.064 *** Job 0.118 0.080 Sale tax incentives 0.322 0.052 *** Price 4.579 0.385 *** Standard deviation parameters Goal STD 0.330 0.049 *** Net metering STD 0.578 0.049 *** Property tax incentives STD 0.655 0.058 *** Job STD 0.930 0.064 *** Sale tax incentives STD 0.507 0.051 *** Price STD 2.372 0.314 *** No. of persons 311 Log likelihood 3451.77 Chi squared (13 d.f.) 1979.82 Pseudo R 2 0.2229 AIC 6929.6 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
103 Table 614. RPL C estimation results for group 2. Attributes Coeff. Std. err. Sig. Variable Percentage of solar energy 0.167 0.038 *** Net metering 0.088 0.052 Property tax incentives 0.311 0.066 *** Job 0.087 0.060 Sale tax incentives 0.293 0.048 *** Price 4.273 0.317 *** Standard deviation parameters Percentage of solar energy STD 0.289 0.053 *** Net metering STD 0.525 0.050 *** Property tax incentives STD 0.690 0.112 *** Job STD 0.744 0.091 *** Sales tax incentives STD 0.465 0.091 *** Price STD 1.965 0.275 *** Covariance of random parameters Net metering: Percentage of solar energy 0.009 0.022 Property tax incentives: Percentage of solar energy 0.111 0.040 *** Property tax incentives: Net metering 0.107 0.065 Job: Percentage of solar energy 0.158 0.037 *** Job: Net metering 0.160 0.090 Job: Property tax incentives 0.322 0.116 *** Sales tax incentives: Percentage of solar energy 0.056 0.029 Sales tax incentives: Net metering 0.106 0.045 ** Sales tax incentives: Property tax incentives 0.010 0.069 Sales tax incentives: Job 0.211 0.058 *** Price: Percentage of solar energy 0.120 0.077 Price: Net metering 0.658 0.170 *** Price: Property tax incentives 0.514 0.229 ** Price: Job 0.692 0.236 *** Price: Sales tax incentives 0.261 0.166 No. of persons 311 Log likelihood 3371.22 Chi squared (28 d.f.) 2140.94 Pseudo R 2 0.2410 AIC 6798.4 Note: Sig. = Significance level: ***0.01, **0.05, *0.1,
104 Table 615. Parameters estimates of the LCM model for group 1 Class 1 Class 2 Class 3 Utility parameters Job 1.195*** (0.182) 0.052 (0.057) 0.578*** (0.136) Price 0.021 (0.012) 0.009 (0.006) 0.089*** (0.009) Production incentives 0.604*** (0.116) 0.110** (0.044) 0.519*** (0.126) Rebate programs 0.320*** (0.079) 0.077** (0.034) 0.113 (0.087) ITC 1.381*** (0.180) 0.038 (0.054) 0.489*** (0.106) Percentage of solar energy 0.989*** (0.149) 0.013 (0.049) 0.336*** (0.104) Class membership parameters Constant 1.06 (0.951) 1.936** (0.794) Fixed parameters Gender 0.346 (0.457) 0.435 (0.382) Higher income household 0.115 (0.633) 0.099 (0.576) Lower income household 0.119 (0.513) 0.590 (0.431) Age 0.001 (0.014) 0.034*** (0.013) Having children 0.242 (0.435) 0.226 (0.359) Education 0.216 (0.451) 0.521 (0.399) Climate change awareness 1.063** (0.424) 0.602 (0.373) Renting a house 0.189 (0.520) 0.250 (0.412) Having a solar energy system 0.417 (1.086) 1.236* (0.686) Log likelihood function 3742.05 AIC 7566.1 N 3939 Chi square (41 df) 1170.76 Pseudo R 2 0.1353 Average class probabilities 0.230 0.539 0.231 Note: Significance at 10% level ** Significance at 5% level *** Significance at 1% level
105 Table 616. Parameters estimates of the LCM model for group 2 Class 1 Class 2 Class 3 Utility parameters Job 0.346*** (0.047) 0.144* (0.075) 0.336 (0.215) Price 0.008 (0.004) 0.034*** (0.006) 0.138*** (0.021) Net metering 0.149*** (0.046) 0.065 (0.065) 0.100 (0.179) Sales tax incentives 0.377*** (0.039) 0.091* (0.054) 0.248 (0.193) Property tax incentives 0.406*** (0.058) 0.051 (0.069) 0.161 (0.167) Percentage of solar energy 0.215*** (0.038) 0.031 (0.054) 0.168 (0.200) Class membership parameters Constant 1.313 (0.881) 2.143** (0.940) Fixed parameter Gender 0.638 (0.434) 0.666 (0.466) Higher income household 1.330 (0.826) 1.389 (0.875) Lower income household 0.895** (0.414) 0.513 (0.457) Age 0.001 (0.013) 0.025* (0.014) Having children 0.757* (0.390) 0.605 (0.447) Education 0.001 (0.428) 0.347 (0.475) Climate change awareness 0.553 (0.418) 0.259 (0.446) Renting a house 0.417 (0.428) 0.321 (0.472) Having a solar energy system 0.334 (0.762) 0.294 (0.785) Log likelihood function 3414.94 AIC 6911.9 N 4043 Chi squared 2053.49 Pseudo R 2 0.2312 Average class probabilities 0.556 0.297 0.147 Note: Significance at 10% level ** Significance at 5% level *** Significance at 1% level
106 Table 617. Comparison of model based on statistical goodness of fit for group 1. Model Log likelihood AIC Pseudo R 2 MNL 4224.54 8463.1 0.0095 MNL IN 4156.37 8416.7 0.0255 RPL (Basic) 3906.23 7838.5 0.0973 RPL C 3822.09 7700.2 0.1168 LCM 3742.05 7566.1 0.1353 Table 618. Comparison of model based on statistical goodness of fit for group 2. Model Log likelihood AIC Pseudo R 2 MNL 4194.89 8403.8 0.0143 MNL IN 4088.21 8280.4 0.0400 RPL (Basic) 3451.77 6929.6 0.2229 RPL C 3371.22 6798.4 0.2410 LCM 3414.94 6911.9 0.2312
107 Table 6 19. Summary statistics of individual WTPs from group 1. Attribute Mean ($) Standard deviation 95% C.I. MNL Job 12.920 8.190 11.998 13.842 Production incentives 8.059 5.818 7.404 8.714 Rebate programs 15.734 8.254 14.805 16.663 ITC 33.858 16.821 31.964 35.752 Percentage of solar energy 24.407 12.577 22.991 25.823 MNL IN Job 0.125 14.250 1.480 1.730 Production incentives 5.875 14.419 4.251 7.499 Rebate programs 23.750 15.145 22.044 25.455 ITC 7.625 14.167 6.030 9.220 Percentage of solar energy 39.500 21.080 37.126 41.873 RPL C Job 24.375 42.534 19.586 29.164 Production incentives 6.889 4.187 6.417 7.360 Rebate programs 9.631 10.599 8.437 10.824 ITC 24.110 22.573 21.568 26.652 Percentage of solar energy 17.216 12.286 15.832 18.599
108 Table 620. Summary statistics of individual WTP for LCM from group 1 Attribute Mean ($) Standard deviation 95% C.I. Class 1 Job 69.115 4.978 68.555 69.676 Production incentives 32.915 1.533 32.742 33.087 Rebate programs 19.243 2.353 18.978 19.507 ITC 78.779 5.991 78.104 79.453 Percentage of solar energy 58.015 4.749 57.480 58.550 Class 2 Job 1.292 7.514 0.446 2.139 Production incentives 2.160 2.253 1.906 2.414 Rebate programs 4.732 4.052 4.276 5.189 ITC 3.878 2.713 4.183 3.572 Percentage of solar energy 0.756 1.578 0.579 0.934 Class 3 Job 6.712 0.196 6.734 6.690 Production incentives 5.939 0.119 5.926 5.953 Rebate programs 1.283 0.139 1.299 1.267 ITC 5.503 0.034 5.499 5.507 Percentage of solar energy 3.875 0.098 3.864 3.886
109 Table 621. Summary statistics of individual WTPs from group 2. Attribute Mean ($) Standard deviation 95% C.I. MNL Job 7.603 2.057 7.486 7.720 Sales tax incentives 11.909 2.284 11.779 12.039 Property tax incentives 12.009 2.557 11.864 12.154 Net metering 5.055 1.773 4.954 5.156 Percentage of solar energy 6.651 1.668 6.556 6.746 MNL IN Job 5.800 6.657 5.060 6.540 Sales tax incentives 9.450 5.825 8.803 10.097 Property tax incentives 3.550 5.675 4.181 2.919 Net metering 2.350 6.060 1.676 3.024 Percentage of solar energy 8.500 5.500 7.889 9.111 RPL C Job 16.719 43.718 14.240 19.198 Sales tax incentives 21.197 29.859 19.504 22.890 Property tax incentives 24.691 38.450 22.511 26.871 Net metering 8.804 27.414 7.249 10.359 Percentage of solar energy 11.123 16.689 10.177 12.069
110 Table 622. Summary statistics of individual WTPs for LCM from group 2. Attribute Mean ($) Standard deviation 95 % C.I. Class 1 Job 41.637 0.154 41.620 41.665 Sales tax incentives 45.385 0.143 45.369 45.401 Property tax incentives 48.937 0.159 48.919 48.955 Net metering 17.974 0.054 17.968 17.980 Percentage of solar energy 25.901 0.084 25.891 25.910 Class 2 Job 2.148 3.120 2.499 1.796 Sales tax incentives 2.075 1.908 1.861 2.290 Property tax incentives 0.567 2.199 0.319 0.815 Net metering 1.201 1.405 1.043 1.359 Percentage of solar energy 0.314 1.252 0.173 0.455 Class 3 Job 3.921 2.087 4.156 3.686 Sales tax incentives 1.570 0.317 1.605 1.534 Property tax incentives 2.621 2.069 2.388 2.854 Net metering 2.445 2.437 2.170 2.719 Percentage of solar energy 2.429 2.429 1.855 2.402
111 Table 623. Estimated a bid curve for Group 1 Percentage of solar energy Production incentives Rebate program Impact on job ITC Constant 14.400*** (3.304) 5.450*** (1.107) 9.016*** (2.940) 27.828** (11.405) 21.528*** (6.166) Solar resource 0.005 (0.012) 0.001 (0.004) 0.003 (0.011) 0.038 (0.043) 0.017 (0.023) Age 0.075* (0.043) 0.040*** (0.014) 0.059 (0.038) 0.032 (0.148) 0.047 (0.080) Gender 1.143 (1.400) 0.991** (0.469) 2.298* (1.246) 1.237 (4.833) 1.376 (2.613) Education 1.475 (1.592) 0.597 (0.534) 0.195 (1.417) 2.445 (5.498) 0.183 (2.972) Higher income household 1.091 (1.909) 0.477 (0.640) 0.224 (1.700) 5.454 (6.592) 2.272 (3.564) Lower income household 0.080 (1.720) 0.143 (0.576) 1.267 (1.531) 1.050 (5.937) 2.616 (3.210) Renting a house 3.602** (1.530) 1.423*** (0.512) 2.614* (1.362) 12.509** (5.194) 2.214 (2.857) Having children 0.568 (1.505) 0.062 (0.504) 0.172 (1.339) 1.607 (5.194) 2.853 (2.808) Climate change awareness 1.682 (1.110) 0.748** (0.372) 0.884 (0.988) 7.078* (3.831) 1.123 (2.071) Note: Significance at 10% level ** Significance at 5% level ***Significance at 1% level
112 Table 624. Estimated a bid curve for Group 2 Percentage of solar energy Net metering Sale tax incentives Impact on job Property tax incentives Constant 11.926*** (4.584) 7.608 (7.561) 34.127*** (8.165) 28.804** (12.023) 34.072*** (10.433) Solar resource 0.004 (0.006) 0.000 (0.011) 0.023** (0.011) 0.010 (0.017) 0.004 (0.015) Age 0.033 (0.059) 0.028 (0.097) 0.140 (0.105) 0.061 (0.155) 0.215 (0.134) Gender 2.752 (1.897) 4.452 (3.128) 3.389 (3.378) 3.824 (4.974) 0.398 (4.316) Education 0.481 (2.028) 0.662 (3.344) 3.146 (3.612) 0.007 (5.318) 6.015 (4.615) Higher income household 1.386 (2.503) 1.753 (4.129) 0.003 (4.459) 5.146 (6.566) 14.524*** (5.697) lower income household 3.832 (2.382) 2.956 (3.929) 6.335 (4.242) 13.63** (6.247) 3.206 (5.421) Renting a house 2.341 (2.184) 2.219 (3.602) 0.575 (3.890) 1,971 (5.728) 1.842 (4.970) Having children 1.081 (2.016) 1.098 (3.326) 3.071 (3.591) 1.374 (5.289) 3.103 (4.589) Climate change awareness 1.289 (1.362) 2.791 (2.247) 2.352 (2.426) 2.744 (3.572) 1.303 (3.100) Note: Significance at 10% level** Significance at 5% level *** Significance at 1% level
113 Figure 61 Kernel density for Percentage of Solar Energy from group1. Figure 6 2 Kernel density for Production Incentives from group 1 Figure 63 Kernel density fo r Rebate Programs from group 1 RWTPG .0056 .0112 .0167 .0223 .0279 .0000 0 10 20 30 40 50 60 70 -10 Kernel density estimate for RWTPG Density RWTPP .016 .032 .048 .064 .079 .000 4 8 12 16 20 0 Kernel density estimate for RWTPP Density RWTPR .0083 .0167 .0250 .0333 .0416 .0000 0 10 20 30 40 50 60 -10 Kernel density estimate for RWTPR Density
114 Figure 64 Kern el density for Job from group 1. Figure 65 Kern el density for ITC from group 1. Figure 6 6 Kernel de nsity from group 1. RWTPJ .0021 .0042 .0063 .0083 .0104 .0000 0 100 200 -100 Kernel density estimate for RWTPJ Density RWTPI .0032 .0065 .0097 .0130 .0162 .0000 -25 0 25 50 75 100 125 -50 Kernel density estimate for RWTPI Density .016 .032 .048 .064 .079 .000 0 100 200 -100 RWTPG RWTPP RWTPR RWTPJ RWTPI Density
115 Figure 67 Kernel density for Percentage of Solar Energy from group 2 Figure 68 Kernel density for Net Metering from group 2. Figure 69 Kernel density for Proper ty Tax Incentives from group 2 RWTPG .0061 .0121 .0182 .0242 .0303 .0000 -50 0 50 100 -100 Kernel density estimate for RWTPG Density RWTPN .0039 .0078 .0117 .0156 .0194 .0000 -100 -50 0 50 100 150 200 250 -150 Kernel density estimate for RWTPN Density RWTPP .0025 .0050 .0075 .0101 .0126 .0000 -50 0 50 100 150 200 -100 Kernel density estimate for RWTPP Density
116 Figure 610. Kernel density for Job from group 2 Figure 611. Kernel density for Sale Tax Incentives from group 2 Figure 612. Kernel density from group 2. RWTPJ .0022 .0044 .0066 .0088 .0110 .0000 -150 -100 -50 0 50 100 150 200 250 -200 Kernel density estimate for RWTPJ Density RWTPS .0032 .0064 .0096 .0128 .0160 .0000 -100 -50 0 50 100 150 -150 Kernel density estimate for RWTPS Density .0061 .0121 .0182 .0242 .0303 .0000 -150 -100 -50 0 50 100 150 200 250 -200 RWTPG RWTPN RWTPP RWTPJ RWTPS Density
117 CHAPTER 7 CONCLUSION AND DISCUSSION Expanding solar energy capacity in the U.S is a good option to address the issue of climate change and global warming The inherent sustainability of solar energy should provide significant motivations for making a transition away from fossil fuel s and toward solar energy. In this study, we applied choice exper iment s to examine the ef fect of different policies to promote solar energy to the US household. The data were collected by an online survey that was distributed to national representative consumer panels in the U.S. Two choice experiments were designed ac cording to different solar energy policies: t he first choice experiment ( g roup 1 ) include d the ITC, production incentives, and rebate program s, and t he second choice experiment (group 2 ) included the net metering, sale tax incentive, and property tax incentives. Both groups considered three additional important attributes such as the percentage of solar energy, change in electricity prices and impact on the job Several discrete choic e models are estimated and compared including MNL, MNL IN, RPL, RPL C, and LCM models The r esults of RPL, RPL C, and LCM model take account of the heter ogeneity in consumer preference. Howe ver, it is inconclusive regarding the performance of RPLC and LCM; for group 1 RPL C performs better and LCM has slight advantages for group 2. This result is consistent with previous research that cannot give a clear conclusion regarding the performance of RPL and LCM models. For example, Kosenius (2010) showed th at RPL model fit the data better than LCM model, however Shen (2009) and Beharry Borg and Scarpa (2010) demonstrated that LCM models result in more efficient estimates than RPL model
118 RPL C results in group 1 show that if respondents pay more attention to new jobs opportunities created by the solar energy industry, they could care less about the percentage of solar energy target for solar energy installation. On the other hand, if the respondents care more about the goal for solar energy installation, they could not care about the rebate programs and ITC. For the LCM estimation results in group 1, class 1 was named as Prosolar energy class and respondents with climate change awareness are more like to be in class 1. Only the reducing solar energy installation cost can increase respondents preference to adopt solar energy and young respondents are more like to be classified into class 2. For the RPLC estimation results in group 2, if respondents care more about property tax incentives, they could not focus on the goal for solar energy installation. Otherwise, if the respondents care about the new job opportunities from the solar energy industry, they also support the property tax incentives and sales tax incentives. For the LCM estimation results in group 2 class 1 was named as Prosolar energy class and the respondents who had children are more like to be in class 1. For class 2, younger respondents are more like to be classified in it but the new job opportuniti es created by solar energy industry cannot increase respondents preference to adopt solar energy. Last, class 3 was named as No solar energy class. Next, compared with policies, people cared more about ITC in group 1. In group 2, respondents cared about property tax incentive and sale tax incentives. Another finding is that people with climate change awareness care significantly about the pol icies to promote solar energy consistent with Bergmann et al. (2006) study, which also concluded that envir onmental attributes significantly impacted public acceptabil ity
119 for renewable energy in S cotland. On the other hand, the higher income household would be willing to pay more to increase the percentage of solar energy, consistent with Batley et al. ( 2001) and Zarcnikau (2003) studies Last, respondents cared about t he new job opportunities created by the solar energy industry, consistent with Longo et al. (2008) and OKeeffe (2014) study. For the mean WTP, our results consistently present the heterogeneity in WTP for policies in groups 1 and 2. The heterogeneity in WTP can be regard as a tool to build political supports for policies that promote the sol ar energy. For the mean WTP in g roup 1, t he mean value for ITC was higher than these for rebate programs and production incentives For the mean WTP in group 2, the mean value of property tax incentives was higher than these for sales tax incentives and net metering. W e estimate the bid curves for WTP s for group s 1 and group 2 For group 1, the older respondents are willing to pay more to increas e the percentage of solar elec tricity target by 2020 and to support the production incentives. The male respondents are also willing to pay more to promote production incentives and rebate programs The respondents who own a house will pay more to support the percentage of solar electricity target by 2020, production incentives, rebate programs, and the job market in the solar industry For climate change awareness, the respondents who concern in climate change will pay more to support production incentives, and create new job opportunities by the solar energy industry Renting a house for respondents is an important determinant of WTP for group 1. For group 2, the respondents living at low solar radiation are willing to pay more to support sales tax incentives H igher income respondents are willing to pay more to support property tax incentives. Lower income
120 respondents are not willing to pay more for the new job created by the solar energy industry. F or the development policy, ITC and property tax incentives need to be maintained because these policies have consistent and stable effects on the adoption of solar energy. Many r espondents care about the job opportunities created by solar energy industry Therefore, policies for solar energy may need a specific requirement for new job s created by the solar industry. Last, a lthough this study shows that U.S. solar policies have some effect on consumers preference, the conclusions are drawn based on hypothetical choice experiment surveys. A potential hypothetical bias may exist (Carlsson and Martinsson 2001; Fifer et al. 2014) Therefore, future research can compare our results to those from revealed preference methods
121 REFERENCE LIST Adua, L. (2008). Alternative Energy: Political, Economic, and Social Feasibility, by Christopher A. Simon, Lanham, MD: Rowman. Rural Sociology, 73(4), 684686. Alnatheer, O. (2005). The potential contribution of renewable energy to electricity supply in Saudi Arabia. Energy Policy, 33(18), 22982312. Barbose, G., Darghouth, N. R., Weaver, S., Feldman, D., Margolis, R., & Wiser, R. (2015). Tracking US photovoltaic system prices 1998 2012: a rapidly changing market. Progress in Photov oltaics: Research and Applications, 23(6), 692704. Batley, S. L., Colbourne, D., Fleming, P. D., & Urwin, P. (2001). Citizen versus consumer: challenges in the UK green power market. Energy policy, 29(6), 479487. Beck, F., & Martinot, E. (2004). Renewable energy policies and barriers. Encyclopedia of energy, 5(7), 365383. Beharry Borg, N., & Scarpa, R. (2010). Valuing quality changes in Caribbean coastal waters for heterogeneous beach visitors. Ecological Economics, 69(5), 11241139. Bergmann, A., Hanley, N., & Wright, R. (2006). Valuing the attributes of renewable energy investments. Energy policy, 34(9), 10041014. Berman, E. (1985). U.S. Patent No. 4,510,344. Washington, DC: U.S. Patent and Trademark Office. Berry, T., & Jaccard, M. (2001). Th e renewable portfolio standard: design considerations and an implementation survey. Energy Policy, 29(4), 263277. Borchers, A. M., Duke, J. M., & Parsons, G. R. (2007). Does willingness to pay for green energy differ by source? Energy policy, 35(6), 33273334. Bos, I. D., Van der Heijden, R. E., Molin, E. J., & Timmermans, H. J. (2004). The choice of park and ride facilities: an analysis using a context dependent hierarchical choice experiment. Environment and Planning A, 36(9), 16731686. Boxall, P. C., Adamowicz, W. L., Swait, J., Williams, M., & Louviere, J. (1996). A comparison of stated preference methods for environmental valuation. Ecological economics, 18(3), 243253. Carlsson, F., & Martinsson, P. (2001). Do hypothetical and actual marginal will ingness to pa y differ in choice experiments? Application to the valuation of the
122 environment. Journal of Environmental Economics and Management, 41(2), 179192. Cascetta, E. (2009). Random utility theory. In Transportation systems analysis (pp. 89167). Springer US. Caussade, S., de Dios Ortzar, J., Rizzi, L. I., & Hensher, D. A. (2005). Assessing the influence of design dimensions on stated choice experiment estimates. Transportation research part B: Methodological, 39(7), 621640. Champ, P. A., Boyle, K. J., & Brown, T. C. (Eds.). (2012). A primer on nonmarket valuation (Vol. 3). Springer Science & Business Media. Chapin, D. M., Fuller, C. S., & Pearson, G. L. (1957). U.S. Patent No. 2,780,765. Washington, DC: U.S. Patent and Trademark Office. Cicia, G., Cembalo, L., Del Giudice, T., & Palladino, A. (2012). Fossil energy versus nuclear, wind, solar and agricultural biomass: Insights from an Italian national survey. Energy Policy, 42, 5966. Duke, R., Williams, R., & Payne, A. (2005). Accelerati ng residential PV expansion: demand analysis for competitive electricity markets. Energy Policy,33(15), 19121929. Ek, K. (2005). Public and private attitudes towards green electricity: the case of Swedish wind power. Energy Policy, 33(13), 16771689. Farhar, B. C. (1999). Willingness to pay for electricity from renewable resources: a review of utility market research. National Renewable Energy Laboratory, NREL/TP, 550. Fifer, S., Rose, J., & Greaves, S. (2014). Hypothetical bias in Stated Choice Exper iments: Is it a problem? And if so, how do we deal with it? Transportation research part A: policy and practice, 61, 164 177. Gao, Z., & Schroeder, T. C. (2009). Effects of label information on consumer willingness to pay for food attributes. American Journal of Agricultural Economics, 91(3), 795 809. Gnansounou, E., Dauriat, A., & Wyman, C. E. (2005). Refining sweet sorghum to ethanol and sugar: economic tradeoffs in the context of North China. Bioresource technology, 96(9), 9851002. Goodrich, A., Jam es, T., & Woodhouse, M. (2012). Residential, commercial, and utility scale photovoltaic (PV) system prices in the United States: current drivers and cost reduction opportunities. Contract, 303(2012), 2753000.
123 Greene, W. H., & Hensher, D. A. (2003). A latent class model for discrete choice analysis: contrasts with mixed logit. Transportation Research Part B: Methodological, 37(8), 681698. Hanley, N., Mourato, S., & Wright, R. E. (2001). Choice Modelling Approaches: A Superior Alternativ e for Environmental Valuation? Journal of economic surveys, 15(3), 435462. Hansen, J., Johnson, D., Lacis, A., Lebedeff, S., Lee, P., Rind, D., & Russell, G. (1981). Climate impact of increasing atmospheric carbon dioxide. Science,213(4511), 957966. Hensher, D. A. (1994). Stated preference analysis of travel choices: the state of practice. Transportation, 21(2), 107133. Hernndez Moro, J., & Martnez Duart, J. M. (2013). Analytical model for solar PV and CSP electricity costs: Present LCOE values and their future evolution. Renewable and Sustainable Energy Reviews, 20, 119132. Herriges, J. A., & Kling, C. L. (1999). Nonlinear income effects in random utility models. Review of Economics and Statistics, 81(1), 6272. Horne, P., Boxall, P. C., & Adamowicz, W L. (2005). Multipleuse management of forest recreation sites: a spatially explicit choice experiment. Forest Ecology and Management, 207(1), 189199. Kammen, D. M., Kapadia, K., & Fripp, M. (2004). Putting renewables to work: How many jobs can the clea n energy industry generate. RAEL Report, University of California, Berkeley, 13. Kosenius, A. K. (2010). Heterogeneous preferences for water quality attributes: The case of eutrophication in the Gulf of Finland, the Baltic Sea. Ecological Economics, 69(3), 528 538. Kuhfeld, W. F. (2005). Experimental design and choice modelling macros. SAS technical support document, SAS website: http://support.sas.com. Kuhfeld, W. F., Tobias, R. D., & Garratt, M. (1994). Efficient experimental design with marketing research applications. Journal of Marketing Research, 545557. Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated choice methods: analysis and applications. Cambridge University Press.
124 Lancaster, K. J. (1966). A new approach to consumer theory. The journal of political economy, 132157. Longo, A., Markandya, A., & Petrucci, M. (2008). The internalization of externalities in the production of electricity: willingness to pay for the attributes of a policy for renewable energy. Ecological Economics, 67(1), 140152. Loomis, J. B. (2014). Strategies for overcoming hypothetical bias in stated preference surveys. Journal of Agricultural and Resource Economics, 39(1), 3446. Loureiro, M. L., & Umberger, W. J. (2007). A choice experiment model for beef: What US consumer responses tell us about relative preferences for food safety, country of origin labeling and traceability. Food policy, 32(4), 496514. Louviere, J. J. (1988). Conjoint analysis modelling of stated preferences: a review of theory, methods, recent developments and external validity. Journal of transport economics and policy, 93119. Louviere, J. J. (1992). Experimental choice analysis: introduction and overview. Journal of business research, 24(2), 8995. Manski, C. F. (1977). The structure of random utility models. Theory and decision, 8(3), 229254. Martinsson, P., Carlsson, F., & Alpizar, F. (2001). Using Choice Experiments for NonMarket Valuation. rapport nr.: Working Papers in Economics, nr (52). McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. McFadden, D. (1980). Econometric models for probabilistic choice among products. Journal of Business, S13S29. Menegaki, A. (2008). Valuation for renewable energy: a comparative review. Renewable and Sustainable Energy Reviews, 12(9), 2422 2437. Navrud, S., & Brten, K. G. (2007). Consumers' preferences for green and brown electricity: a choice modelling approach. Revue d'conomie politique,117(5), 795811. Nomura, N., & Akai, M. (2004). Willingness to pay for green electricity in Japan as estimated through contingent valuation method. Applied Energy, 78(4), 453463. O'Keeffe, L. (2014). A Choice Experiment Survey Analysis of Public Preferences for Renewable Energy in the United States. Journal of Environmental and Resource Economics at Colby, 1(01), 107.
125 Panwar, N. L., Kaushik, S. C., & Kothari, S. (2011). Role of renewable energy sources in environmental protection: a review. Renewable and Sustainable Energy Reviews, 15(3), 15131524. Roe, B., Teisl, M. F., Levy, A., & Russell, M. (2001). US consumers willingness to pay for green electricity. Energy policy, 29(11), 917925. Sagebiel, J. (2011). Comparing the latent class model with the random parameters logit. A choice experiment analysis of highly heterogeneous electricity consumers in Hyderabad, India. In II International Choice Modeling Conference, Leeds. Scarpa, R., & Willis, K. (2010). Willingness to pay for renewable energy: Primary and discretionary choice of British households for micro generation technologies. Energy Economics, 32(1), 129 136. Shen, J. (2009). Latent class model or mixed logit model? A comparison by transport mode choice data. Applied Economics, 41(22), 29152924. Shen, J., & Saijo, T. (2009). Does an energy efficiency label alter consumers' purchasing decisions? A latent class approach based on a stated choice experiment in Shanghai Journal of Environmental Management, 90(11), 35613573. Sherwood, L. (2007). US solar market trends. In Proceedings of the Solar Conference (Vol. 2, p. 909). American Solar Energy Society; American Institute Of Architects Swait, J., & Adamowicz, W. (2 001). The influence of task complexity on consumer choice: a latent class model of decision strategy switching. Journal of Consumer Research, 28(1), 135148. Tingem, M., & Rivington, M. (2009). Adaptation for crop agriculture to climate change in Cameroon: turning on the heat. Mitigation and Adaptation Strategies for Global Change, 14(2), 153168. Tsoutsos, T., Frantzeskaki, N., & Gekas, V. (2005). Environmental impacts from the solar energy technologies. Energy Policy, 33(3), 289296. Wiser, R., & Barbo se, G. (2008). Renewables portfolio standards in the United States. Berkeley, CA: Lawrence Berkeley National Laboratory, 24. Wiser, R. H. (2007). Using contingent valuation to explore willingness to pay for renewable energy: a comparison of collective and voluntary payment vehicles. Ecological economics, 62(3), 419432.
126 Witt, J., Scott, A., & Osborne, R. H. (2009). Designing choice experiments with many attributes. An application to setting priorities for orthopaedic waiting lists. Health economics, 18(6), 681696. Xiarchos, I. M., & Vick, B. (2011). Solar Energy Use in US Agriculture: Overview and Policy Issues. US Department of Agriculture, Office of the Chief Economist, Office of Energy Policy and New Uses. Yoo, J., & Ready, R. C. (2014). Prefere nce heterogeneity for renewable energy technology. Energy Economics, 42, 101114. Yoo, S. H., & Kwak, S. Y. (2009). Willingness to pay for green electricity in Korea: A contingent valuation study. Energy policy, 37(12), 54085416. Zarnikau, J. (2003). Consumer demand for green power and energy efficiency. Energy Policy, 31(15), 16611672.
127 BIOGRAPHICAL SKETCH ChaoLin Lu received his Ph.D. in food and resource economics at the University of Florida in 2016. He also holds a masters degree in economics from University at Albany, SUNY and a masters degree in agricultural economics from National Taiwan University. Chao Lins research interests focus on market research, consumer behavior, survey and experimental design, eco nomic forecasting, and statistic and predictive modeling.