Willingness-to-pay for renewable energy - NYU

Energy Economics 32 (2010) 129–136

Contents lists available at ScienceDirect

Energy Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e n e c o

Willingness-to-pay for renewable energy: Primary and discretionary choice of British households' for micro-generation technologies Riccardo Scarpa a,1, Ken Willis b,⁎ a b

Department of Economics, Waikato Management School, University of Waikato, Hamilton, New Zealand Centre for Research in Environmental Appraisal & Management, School of Architecture, Planning and Landscape, University of Newcastle, New castle upon Tyne NE1 7RU, U.K.

a r t i c l e

i n f o

Article history: Received 30 January 2009 Received in revised form 2 June 2009 Accepted 8 June 2009 Available online 13 June 2009 Jel classification: Q42, alternative energy sources D12, consumer economics empirical analysis Keywords: Renewable energy Micro-generation Choice modelling Mixed logit

a b s t r a c t This paper documents the policy context of renewable energy production in the European Union. The research adopts a choice experiment approach to investigate households' WTP for these renewable energy technologies in the UK. The micro-generation technologies comprise solar photovoltaic, micro-wind, solar thermal, heat pumps, and biomass boilers and pellet stoves. The study compares the results from conditional and mixed logit models, which estimate the distribution of utility coefficients and then derives WTP values as a ratio of the attribute coefficient to the price coefficient, with a model in which the WTP distribution is estimated directly from utility in the money space. The results suggest that whilst renewable energy adoption is significantly valued by households, this value is not sufficiently large, for the vast majority of households, to cover the higher capital costs of micro-generation energy technologies. © 2009 Elsevier B.V. All rights reserved.

1. Introduction The European Union (EU) Directive on production of electricity from renewable energy sources 2001/77/EC (OJ L 283, 27.10.2001) sets an indicative target of 21% of electricity to be produced from renewable energy sources by 2010, with indicative targets for each member state. The UK target is 10% of electricity from a renewable source by 2010. With current EU policies and progress, it is estimated that renewable electricity sources will account for 19% production in the EU by 2010. In 2008 the EU proposed a binding target of 20%, as the renewable energy source share of all energy consumption by 2020 (Commission of the European Communities, 2008a). Again, national targets vary depending on past progress and feasible achievements by 2020. In 2005 UK renewable energy sources accounted for 1.3% of final energy consumption, compared to 39.8% in Sweden (the highest share for all 27 EU countries). The UK target for 2020 is 15%, compared to 49% for Sweden. Renewable energy policy is driven by the projected damage costs of continued carbon emissions into the atmosphere (i.e., climate change), the real rise in carbon based domestic energy prices over the ⁎ Corresponding author. Tel.: +44 191 222 7807; fax: +44 191 222 8811. E-mail addresses: [email protected] (R. Scarpa), [email protected] (K. Willis). 1 Tel.: +64 7 8384848. 0140-9883/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.eneco.2009.06.004

last few years, and the perceived need for security of supply (i.e., to reduce dependence on a limited number of energy sources: oil and gas). To achieve energy policies of sustainability and security of supply, governments have used instruments to correct for market failure, particularly negative externalities associated with fossil fuels. These instruments can be categorised into awareness measures (i.e., making consumers aware of the benefits of energy efficiency, and attempting to change their behaviour with respect to fossil fuel consumption); command and control instruments (e.g., changing building regulations to require more insulation and energy efficiency measures to be incorporated into new buildings); and market based instruments. Market based instruments themselves can be classified into investment support (capital grants, tax exemptions, and reductions in the purchase price of goods e.g., energy saving light bulbs) and operating support (price subsidies, green certifications, tender schemes, and tax exemptions on the production of electricity) (Commission of the European Communities, 2008b). Priced based market instruments in the UK include feed-in tariffs and premiums, providing a guaranteed price for renewable electricity fed into the national grid. Fiscal incentives in the UK include taxes (the Climate Change Levy (CCL))2, and cap and trade instruments (the Renewables Obligation). 2 This is a charge on non-domestic electricity supply in the UK, i.e., it applies to supplies to industrial, commercial, agricultural, and public service, users.

130

R. Scarpa, K. Willis / Energy Economics 32 (2010) 129–136

The UK Energy Act 2008 provides enabling powers for the introduction of feed-in tariffs (FITs) for small scale low carbon electricity generation, up to a maximum of 5 MW. These provisions are intended to encourage the uptake of small sale low carbon technologies; while the Renewables Obligation (RO) continues to be the main support mechanism for large scale renewable projects. The UK Government is committed to having FITs in place by 2010.3 The CCL is set at GBP4.56/MWh for electricity (2008 UK Budget4) (equivalent to GBP31 per tonne carbon), with exemption for sources that do not contribute to climate change.5 The Renewables Obligation (RO) is administered by OFGEM.6 The RO requires power suppliers to derive a proportion of the electricity supplied to consumers from renewable sources.7 Renewable electricity generators receive Renewable Obligation Certificates (ROCs) for each MWh of metered production. These ROCs can be sold to other electricity suppliers to allow them to fulfil their obligation.8 The RO was designed as a market mechanism to increase electricity production sourced from renewable sources. ROCs have increased the profitability of renewable energy generation, since ROCs have a sale value for renewable producers in addition to the price of electricity generated (see Simpson, 2004; and Mitchell et al, 2006). In addition to promoting larger scale renewable energy production with the CCL and the RO, the UK government's micro-generation strategy envisages micro-generation technologies making a significant contribution to supplying energy close to the point of consumption, and also reducing CO2 emissions (DTI, 2006). The built environment accounts for approximately 47% of CO2 emissions in the UK. Micro-generation technologies can potentially reduce this figure by providing renewable energy, or carbon neutral, sources of heat and electricity to houses, buildings, and small commercial premises (DTI, 2006). Renewable energy use in houses, community and commercial premises is encouraged by BERR9 grants administered through the Energy Saving Trust (EST). Householders can apply for grants up to GBP2500 per property towards the cost of installing a certified microgeneration product by a certified installer. Micro-generation technologies include biomass heaters and stoves (automated wood pellet feed); ground source heat pumps; heat pumps (air); small scale hydro; solar thermal hot water; wind turbines; wood fuelled boiler systems.

3 Currently (as of May 2009) some companies offer feed-in tariffs. British Gas will purchase electric power from solar photovoltaic (PV) units up to 5 kW in size at 5.0 pence per kWH; whilst E-On pays between 8.89 pence and 11.27 pence per kWH (depending on the geographical region) for each unit exported by customers from solar PV units up to 5 kW. See Energy Saving Trust website: http://www. energysavingtrust.org.uk/Generate-your-own-energy/Sell-your-own-energy/Searchfor-buy-back-tariffs/(size)/5/(filter)/max (accessed 20/05/2009). 4 The CCL is set periodically in the UK government annual budget, rather than having a built in cost escalator. The CCL of £4.56/MWh was set in the 2008 budget to apply to 2008/09, and subsequent years and until revised. 5 Electricity from nuclear power stations, and hydro schemes larger than 10 MW, do not qualify for exemption. This is illogical, but it is used to stimulate ‘additionality’ in new renewable energy sources, since it is argued that all large scale hydro developments in the UK have already been developed. 6 OFGEM is the Office for Gas and Electricity Markets: the Gas and Electricity Markets Authority, which regulates energy markets ion the UK. 7 This obligation was 3% in 2003, rose to 6.7% in England Wales and Scotland in 2006/07, and is rising to 10.4% by 2010 and 15.4% by 2015. 8 Suppliers can meet their obligation by (i) generating renewable energy; (ii) buying an equivalent amount of ROCs in the trading market and presenting ROCs to cover the required percentage of their output, or (iii) by paying a ‘buy-out’ price for any shortfall. The buyout price is set each year by Ofgem, and in 2007/08 was £34.30/MWh. 9 BERR is the UK government Department for Business, Enterprise and Regulatory Reform. For grants under the low carbon programme see http://www.lowcarbonbuildings.org.uk/home (accessed 20/05/2009). The budget for these grants is limited, and from the start of this policy in April 2006 to May 2009 only GBP10 million was paid to households in grants to assist with the cost of installing micro-generation technologies.

A number of studies have explored the preference for and use of renewable energy by households using binary or multinomial logit (MNL) models. Ek (2005) found households in Sweden generally had a positive attitude towards wind power. Bergmann, Hanley and Wright (2006) assessed the landscape, wildlife, air pollution, and employment impacts of large wind turbines in Scotland; whilst Aravena, Martinsson and Scarpa (2006) explored willingness-to-pay (WTP) in Chile for different wind-farm locations, as well as their extension and the damage caused to the population of migrating birds. Longo, Markandya and Petrucci (2006) investigated WTP by households in Bath for a hypothetical program promoting the production of green energy. Rivers and Jaccard (2005) found capital cost and fuel cost, to be significant determinants of boiler choice by firms, compared to non-fuel operating costs. Borchers, Duke and Parsons (2007) estimated the premium households were willing to pay for on their electricity bills for ‘green energy' purchased from specific sources: wind, biomass, solar, and farm methane. However, fewer studies have estimated WTP for energy saving measures in residential and small commercial buildings. Banfi, Farsi, Filippini, and Jakob (2008) explored willingness to pay for energysaving measures (various types of window insulation, types of façade insulation, and ventilation) in residential buildings in Switzerland. And a study by OXERA (2006) revealed that many people perceived loft insulation and cavity wall insulation to be higher than actual costs; but that perceptions of the costs and benefits of energy efficient light bulbs were significantly more accurate. This study adopts a choice experiment approach to investigate the determinants of the adoption of micro-generation technologies by households in the UK; and their WTP for these renewable energy technologies. The micro-generation technologies comprise solar photovoltaic,10 micro-wind11, solar thermal,12 heat pumps,13 and biomass boilers and pellet stoves.14 The remainder of the paper is structured as follows. The next section outlines stated preference theory, choice experiments, and methodology adopted in the paper. This is followed by sections on the experimental design of the study, and the questionnaire survey. This is followed by the choice experiment (CE) willingness-to-pay (WTP) results for households contemplating renewing their heating system; and then the CE WTP results for households thinking about adding micro-generation to their existing conventional heating system. Finally, some conclusions are drawn about the two approaches of parameterizing models in ‘preference (utility) space’ with those in ‘WTP space’; and also some conclusions on the barriers to the installation of micro-generation technologies.

2. Stated preference theory and choice experiments Choice experiments are based upon consumer demand theory, particularly the theory of consumer behaviour following Lancaster (1966) and Rosen (1974). This assumes that utility to customers e.g., micro-generation technologies, derives from the characteristics of these complex goods. It is based on the notion that consumers are not only interested in energy, but also in the modes that energy is produced.

10 Solar roof panels, comprising thin layers of semiconductor material, convert sunlight to electrical energy. Output is determined by the area of the panels, their efficiency, and the brightness of natural light available. 11 A roof or pole mounted turbine converts kinetic energy of wind to electrical energy. Output is determined by turbine size and wind speed. 12 A roof or pole mounted turbine converts kinetic energy of wind to electrical energy. Output is determined by turbine size and wind speed. 13 Thermal roof panels use sunlight to heat water. Output is a function of available sunlight, panel area, and panel efficiency. 14 Usually wood chips or pellets, and used for space heating and hot water needs. Requires space for boiler and fuel storage.


The choice experiments (CEs) presented customers with sets of alternative combinations of attributes (or characteristics) of microgeneration technologies, and asked them to choose their most preferred alternative. Repeated choices by customers from sets of alternatives reveal the trade-offs customers are willing to make between attributes and micro-generation technologies. Each individual was asked to choose one alternative from each choice set. This choice is modelled as a function of the attributes of that microgeneration using Random Utility Theory. Random Utility Theory is based on the hypothesis that individuals will make choices based on the characteristics of a good (an objective component) along with some degree of randomness (a random component) which helps the analyst reconcile theory with observed choice. The random component arises either because of randomness in the preferences of the respondent or the fact that the researcher does not have the complete set of information available to the respondent. Consider a respondent facing a set of K alternatives, denoted by j = 1,…,K. The utility function for respondent i related to alternative j is specified as: Uij ¼Vij + eij

ð1Þ

where Vij is a non-stochastic utility function and ɛij is a random component. If it is assumed that Vij is a linear utility function then Vij = β'xij. The conditional logit (CL) model is derived by placing restrictive assumptions on the random component of the utility: error disturbances are assumed to be independently and identically distributed according to a Type 1 extreme value distribution with the cumulative distribution function: exp −exp −μeij ;

ð2Þ

where μ is the scale parameter that remains unidentified in estimation. The indicator function 1i(.) for the selection of a generic alternative j may be expressed as: 8 n o < 1 iff U = max U j = k ij ij j=1 1i ð jÞ = : 0 otherwise:

ð3Þ

Differences from independent and identically distributed (iid) errors from a Type 1 extreme value distribution are distributed logistically. As a consequence, assuming a linear indirect utility, the probability that the individual utility of alternative j is the largest among those in the choice set C facing individual i is conditional logit (McFadden 1974): 0 0 Pið jÞ = exp β xij =Σk exp β xik ; j; kaC

ð4Þ

A property of the conditional logit (CL) model is the independence of irrelevant alternatives (IIA). IIA implies that all crosseffects are equal; so, for example, if one micro-generation technology gains in utility it draws shares from other micro-generation technologies and the status quo energy provision in proportion to their current market share. The IIA property derives from the assumption of an independent and identical distribution (iid) in the random utility function: independence of utility across alternatives and across choice contexts. This is clearly a strong assumption that is worth relaxing. In fact other forms of choice modelling relax the IIA assumption, and adopt different distributions for the error term, and different structures in decision-making. The nested logit (NL) model assumes that decisions are taken sequentially following a decision tree. NL models assume a generalised extreme value distribution for the error term εij, where the distribution of εij is correlated across alternatives

131

in the same nest, with the IIA property retained within nests but not between nests. 0

Uij = β xij + eij

ð5Þ

There is support for this decision structure from behavioural observation: customers appear to decide whether to stick with the status quo position or seek a change (Samuelson and Zeckhauser, 1988; Scarpa et al., 2007); and if they decide to adopt micro-generation, then they consider the various micro-generation technologies available in the set of alternatives. Thus, respondents might be assumed to consider whether they are satisfied with their current energy supply, and if not then to consider what micro-technologies they wish to adopt. The panel mixed logit (MXL) model was developed to account for the intuitive fact that decision agents differ from each other. The common formulation is that they differ in terms of taste intensity (Train, 1998), leading to the following utility specification: 0 Uij = β˜ i xij + eij

ð6Þ

the utility function of each customer has some random taste parameters β̃i with values that depend on the values of the parameters θ of an underlying distribution f(β̃|θ). The choice of distribution strongly affects the properties of the model (Hensher and Green 2003). Such random taste parameters β̃i induces correlation across choices made by the same agent but it maintains the advantageous logit probability. In fact εij is independent and identically distributed (iid) Gumbel and so, conditional on the parameter draw, the choice probability is still logit. However, the marginal choice probability must be obtained by integrating the distribution density over the range of parameter values: Z

0 0 ˜ x ; f β˜ jθ dβ exp β˜ i xij = Σk exp β i ik

P ð jÞ =

ð7Þ

Breffle and Morey (2000) are amongst the few who investigated heterogeneity in scale. In their model it is the scale of the Gumbel error, rather than taste intensity that varies randomly across respondents. So that the utility function is: 0

Uij = μ˜ i β xij + μ˜ i eij

ð8Þ

This explicitly accounts for the fact that the same taste intensities might be associated with more or less utility variance (inversely related to error scale) in different choice agents. In our context this may be due, for example, to household differences in the evaluation of the utilities associated with the proposed microgeneration technologies. The utility specifications seen so far are all in the preference space, but often economic investigations focus on WTP estimation. In such contexts a potentially more advantageous reparameterization of utility is the one in the WTP space (Cameron and James, 1987; Train and Weeks, 2005; Sonnier, et al. 2007). In panel mixed logit models these specifications can account for both scale variation and isolate utility coefficients that can immediately be interpreted as marginal WTP effects. Let us denote with αi the cost coefficient and the WTP specification of utility takes the generic form of: 0 Uij = − α i = μ˜ i pij + βi = μ˜ i xij + eij = μ˜ i ˜ p + λ ˜ w 0x + η ; λ ; μ N 0 = −λ i ij i i ij i i ij

ð9Þ

where wi = βi/−αi is a vector of marginal WTP for each of the choice attributes. Note that without random taste or scale variation (9) is a mere reparameterization of Eq. (5). Yet, even in this case the results may be

132


more convenient because coefficient estimates are interpretable in the money space, and so are the estimated standard errors that need not be derived using simulation (Poe et al. 2005) or closed form approximations (e.g., via the delta method (Goldberger, 1991)). In panel mixed logit this specification has the advantage of allowing the researcher to exercise a better control on the distribution of marginal WTP values in the population, which are otherwise jointly determined by the distributional assumptions for the money and attribute coefficients, which give respectively the denominator and the numerator of individual WTP estimates (Scarpa et al. 2008). Note, however, that this specification is non-linear in the parameters and it is hence slower to estimate by maximum simulated likelihood. For this reason some authors advocate the use of Monte Carlo Markov Chains (Train and Weeks 2005; Sonnier, et al. 2007). In our estimation we used BIOGEME (Bierlaire 2003) with the algorithm CFSQP (Lawrence et al. 1997) to avoid the problem of local maxima in maximum simulated likelihood. For the purpose of estimation marginal probability integrals were all approximated numerically by means of simulation methods (Train 2003) based on Modified Hypercube Sampling draws (Hess et al. 2006). Different forms of welfare measure can be calculated from CE model information: compensating variation (the amount of money that can be taken away from (or returned to) a household after a specified micro -generation change improvement (decrement) and still leave the household at its original utility position (the status quo utility position)); and the change in micro-generation cost that would result in X% of households adopting a micro-generation technology. The compensating variation is equivalent to the consumer surplus that a household would receive if the micro-generation change was made, assuming no income effect.

experiment can also be designed to allow for choices between specific micro-generation technologies, as well as a choice of not to install such a technology but to continue space heating and electricity consumption from existing non-renewable sources. Two choice experiments (CEs) were designed. The first CE investigated respondents' choice in a situation where their existing heating system was no longer functioning and had to be replaced. The second CE investigated respondents' choice in a discretionary situation, in which their existing heating and electricity supply was functioning normally but where they had the discretion to supplement it with a renewable energy micro-generation source. The micro-generation technologies for the discretionary CE were selected on the basis of those technologies which households would be most likely to install, namely solar thermal (hot water), solar voltaic and wind turbine (electricity). The attributes covered a range of prices which encompassed the capital costs of these technologies, savings in energy bills, maintenance costs, and sources of recommendation on these technologies. The primary CE included the alternative systems such as biomass boilers and supplementary heat pumps with their associated attributes (with space requirements for fuel storage and hot-water storage tanks), compared to combi-gas boilers which deliver central heating and hot water on-demand without the need for hot water storage or fuel storage or the inconvenience associated with tending solid fuel boilers. The primary heating CE asked respondents to

3. Experimental and survey design

In this CE respondents traded-off 6 attributes: (1) capital cost of the new system, (NEWBOI); (2) energy bill (per month/year), (ENERGY); (3) maintenance cost (per year), (MAINT); (4) recommended by (none, friend, heating engineer, friend and heating engineer (FP)); (5) contract length (unspecified, 1, 2 and 4 years), (CL1,CL2 and CL3); and (6) the inconvenience of the system [none; requires garden to be dug up during installation (GARDUG), requires refueling and space (REFSP) for fuel storage, requires cupboard space for hot water tank (CUPTNK)]. Attribute (6) explored aversion to replacing an existing combi-gas boiler15 either with a new combi-gas boiler plus a supplementary micro-generation system such as ground source heat pump, or with a biomass boiler which would require extra space for fuel storage. Ground source heating pumps, solar thermal, and biomass boilers would require space for a hot water tank compared to a combi-gas boiler. Capital cost, energy bill, and maintenance cost were interval coded; with the other attributes being dummy coded in the analysis. This coding allows the identification of the effect on utility of a qualitative attribute from an established baseline. It has the advantage that the utility coefficients have a more intuitive and immediate interpretation than those obtained with effect-coding. In the primary heating CE, it was assumed that the respondent's primary heating system had ceased functioning and needed to be replaced. Respondents therefore had to chose one of two alternatives: they could not opt for a ‘no choice’ option. A ‘no choice’ option in the primary CE (where the existing heating system was no longer useable and had to be replaced) would have effectively meant the respondent preferred to have no heating in the house! Thus in primary CE respondents had to trade-off capital cost of a new system, with energy savings, maintenance cost, information about new systems, and possible inconvenience of a heating system in making a choice

The choice and WTP for micro-generation can be assessed using either observed behavior or a stated preference method. The number of different types of micro-generation technologies installed in the UK, are outlined in Table 1. It proved difficult to identify all households in England with microgeneration installation, and to obtain detailed data on these households and properties; and also obtain a set of households without microgeneration technologies as a control set, to estimate a hedonic price function for micro-generation technologies. A hedonic price model requires information on price, the specification of the micro-generation technology installed, the alternative ‘without' micro-generation system, the characteristics of the property, and the characteristics of the household, to estimate WTP and choice of micro-generation technology. A stated preference choice experiment approach was therefore adopted, which allowed for (a) more variation in price and (b) better control over information provision and choice set consideration. The

Table 1 Installations of micro-generation technologies. Technology

Number of installations up to 2004a

Number of installations up to 2007b

Photovoltaic Micro-wind Solar thermal Ground source heat pumps Biomass boilers Micro-combined heat and power Micro-hydro Air source heat pumps Fuel cells Total

1301 650 78,470 546 150 990 90

2300 1100 90,000 745–2000 500–600 200–1000 65–75 N 150

a

5 82,202

“Please imagine that your current heating system needs replacement. I would like you to think about some alternative heating systems for your home. All of the following systems would fully replace your current system. For example, if you had a gas boiler, it would be taken out and replaced by the new system. The rest of your heating system, such as the radiators, would not need to be changed.”

95,000–98,000

Source: DTI (2006). b Source: Element Energy (2008a,b). Element Energy (2008a,b) estimates for PV, Micro-CHP, Micro-wind, and Micro-hydro are for grid connected installations.

15 Combi-gas boilers do not require hot water storage tanks. Space heating and hot water is provided ‘on demand’. Combi-gas boilers are now the usual standard boiler installed in domestic properties.

R. Scarpa, K. Willis / Energy Economics 32 (2010) 129–136 Table 2 Primary heating choice experiment: attributes and levels. Attribute

Level 1 Level 2

Level 3

Level 4

Capital cost Energy bill per month (year) Maintenance cost (£/year) Recommended to you by Contract length (years) Inconvenience of system

£3000 £25 (£300) £50

£3500 £50 (£600) £100

£4000 £75 (£900) £150

£4500 £100 (£1200) £200

None

Friend

Plumber

0

1

2

Friend plus plumber 4

None

Requires garden Requires refuelling Requires cupboard to be dug up and space for fuel space for boiler during installation storage

£ = GBP.

about replacement. In the discretionary CE respondents could opt for a ‘no choice’ which meant that they could opt not to install a supplementary micro-generation system to their existing system, but to continue using just their existing (working) system. The CE dealing with the discretionary choice stated: “Now I would like you to imagine that your current heating system is functioning completely normally, and to think about supplementing your existing system with an additional system.” The discretionary CE presented respondents with three microgeneration systems that are typically installed to run along side an existing system: solar thermal (hot water), wind turbine, and photovoltaic (solar electricity). Respondents were shown illustrations for each technology. They were then told that: “All of these technologies produce energy for your home, and this reduces the amount of energy that must be bought from your current supplier, leading to lower bills. Because the energy produced is from renewable sources, these systems also reduce the greenhouse gas emissions of your home.” In this CE respondents traded-off 5 attributes: (1) the type of technology (solar power, solar water, wind turbine); (2) capital cost of the new system (in GBP); (3) maintenance cost (in pounds per year); (4) the source of the recommendation (none, friend, plumber, friend & plumber); and (5) the energy saved by the technology (in GBP per year). In this discretionary CE respondents were given a ‘no purchase’ option, that is the respondent could reject both hypothetical alternatives on the choice card and elect to continue with their existing heating/power system without any additional micro-generation technology. The attributes and attribute levels of the two CEs are listed in Tables 2 and 3. A full factorial experimental design for the six attributes and four levels comprising the primary heating CE provided 46 = 4096 combinations of alternatives; while that for the discretionary technology CE was 1024 (=45). A full factorial design permits the main effect of each attribute, as well as the effect of interactions between all attributes, to be identified. However, since the focus of the study was on the main effect of each attribute (the utility of each attribute irrespective of changes in other attribute levels), a fraction of the full factorial design was adopted. A good fraction of the full factorial design is one that minimises the variance of the estimates one wishes to obtain or conversely maximises the information in the data matrix. Other properties are also of interest, such as orthogonality,16 and balancedness.17 The experimental design was based on minimising the standard error for the utility coefficients (D-efficiency) for the MNL model. In particular,

16

Orthogonality occurs when all estimable effects are uncorrelated. Balance occurs when each level occurs equally often within each factor or alternative. 17

133

we used Dp-efficiency (Ferrini and Scarpa, 2007) under uninformative priors for the indirect utility coefficients and preserved balancedness. This produced 48 alternative choice combinations or ‘attribute bundles’. From these 48 alternatives 24 choice sets were produced, by pairing choice alternatives without replacement, subject to nondominance. Each choice set thus contained two of the generated alternatives, plus the status quo which allowed respondents to continue with their existing energy system without adopting a micro-generation technology. In both CEs, each respondent was provided with four choice tasks and asked to state their preferred option for each. Both experimental designs adopted allowed for main effects to be identified and estimated without confounding. 4. Questionnaire and data The sample comprised a stratified random sample of households across England, Wales, and Scotland. The questionnaire was a computer assisted personal interview (CAPI), administered by TNS well known a market research firm. The survey was conducted in late 2007. The sample was chosen to reflect the age, gender, social class, employment status, income and location of the population in Britain. It also covered different dwelling types, housing ages, and heating configurations. A total of 1279 completed questionnaires from households were obtained. The response rate was high.18 Each respondent completed a total of eight choice cards: four primary CE cards, and four discretionary technology heating CE cards. The sample was stratified across regions, but in the estimation the contribution of each observation to the sample-likelihood was scaled to account for stratification weights. The results can therefore be generalised to the UK as a whole (excluding Northern Ireland). 5. Specification search and results For both datasets we follow the same specification search and present multinomial logit estimates for model in the preference space and then in the WTP-space, so that coefficient estimates for this and other WTP-space models can be interpreted in the money metric, up to some scale. For example, in this case all WTP coefficients are in GBP 1000. To account for taste heterogeneity and avoid the restrictive IIA assumption we then present two panel mixed logit estimates in the WTP-space. In the first where only λi varies, accounting for scale and price coefficient variation. This first model accounts for the intuitive fact that marginal utility of money varies across respondents and so does the scale of the Gumbel error. However, they are assumed to vary jointly. In the second panel RPL model we present a similar model in which there is also some variation in WTP for some attribute. This last model accounts for the fact that taste for some attribute of microgeneration solutions, and therefore the respective intensity of marginal WTP, may also vary across respondents. Various tests (unreported here) were carried out to determine which random WTP attribute to include in this latter specification for both datasets. We assume λi to be distributed log-normal to respect the nonnegativity constraint and WTP to be distributed normally. We report goodness-of-fit measures along with the values of coefficient estimates. Statistical significance is reported in terms of |t-values| for the preference-space model and st. errors for the WTP-space models, along with the p-values for the null of the coefficient to be equal to zero. The variance-covariance estimator is always the robust sandwich estimator. The panel RPL models in WTP space always outperform the fixed effect MNL counterpart indicating that unobserved heterogeneity is at play. Nevertheless the results of these models confirm most of the findings of the MNL models thereby 18 Although TNS was not asked to record response rates, for a random probability sample with return visits to households, TNS estimates its response rates are generally between 60 and 70%.

134


Table 3 Discretionary technology choice experiment: attributes and levels. Attribute

Level 1

Level 2

Level 3

Type of system

Solar hot water

Capital cost Annual energy saving (£) Maintenance cost (£/yr) Recommended by

£500 £100 £30 None

Wind turbine £1000 £200 £60 Friend

Solar electricity £2000 £300 £90 Plumber

Level 4

£5000 £400 £120 Friend plus plumber

£ = GBP.

showing that the main results are robust to specification choice, and more precisely to whether one accounts for the natural issue of taste variation across the sampled population. 6. Primary heating choice Table 4 presents the model estimates for the primary choice of heating for Great Britain. These were obtained from binary responses collected from choice contexts with two alternatives, in groups of four choice sets presented to each respondent. Three factors were dummy coded across their respective levels. “Recommended by” was dummy coded for friend, heating engineer, and friend and heating engineer. The sign and size of the coefficients on these factors are with reference to “no recommendation” and are all positive and significant. A recommendation by a “friend”, or a “heating engineer”, or a “friend and heating engineer” for a system, increased the probability of a respondent choosing such a heating system. Similarly, “inconvenience” was dummy coded for in respect of “refueling and fuel storage space”, “garden being dug-up”, and “requires space for hot water tank”, with respect to no inconvenience (e.g., as exemplified by a combi-gas boiler). These effects are all negative implying that these factors decreased the probability of a respondent choosing a system with this attribute, and reduced the utility of these systems to households. Contract length in years was also dummy coded, with reference to “no contract length”, and all coefficients are negative and significant. If a contract prevented a change in energy supplier for 1, 2, or 4 years, then this reduced the probability of such a heating system being chosen and reduced utility to the respondent. The goodness-of-fit of the MNL model is quite good, with a pseudoR2 = 0.1248. Breffle and Rowe (2002) report that a pseudo-R2 of 0.12 is typical for cross sectional data. In addition, all the factors had their a

priori expected sign. Factors such as capital cost (new_boil), energy bill per year (energy), and maintenance cost per year (maint) have correct expected signs: as these factor levels increase the probability of choosing a system declines and utility decreases. The disutility from higher energy bills is nearly 3 times more important to respondents than the change in the capital cost of a new boiler (per unity change in the cost of each factor). All factors are, with the exception of a 2 year contract length, highly statistically significant. The MNL model in WTP-space shows that amongst the inconveniences explored, the “garden dug-up” is the most dreaded (GBP545 ± 78), followed by the need for fuel storage (GBP446 ± 74), and storage space for a hot water tank is last (GBP221 ± 31). Advice from the heating engineer is worth GBP224, and substantially more than the GBP144 associated with advice from a friend. If both are advising the same solution then the worth of this joint advice is not much greater than that by the heating engineer alone, which is plausible. The panel WTP-space models have a higher fit, but show more or less similar results, with the only difference in the implication that advice from friends is not valued. Allowing the marginal WTP coefficient of such attribute to vary in the last model (WTP-space 2) produces a significant standard deviation for the WTP distribution on the advice by a friend. This is consistent with the existence of heterogeneity of preference with respect to this form of advice, which is plausible. However, the underlying mean is not significantly different from zero and overall this extension does not improve the model fit significantly, nor does it modify the conclusions. 7. Discretionary micro-generation adoption CE The CE for discretionary adoption of micro-generation solutions presented respondents with four choice situations, each of which comprised of three alternatives: the status-quo and two experimentally designed alternatives. Because 38 respondents did not provide complete choices the number of respondents in this analysis was reduced to 1241. In Table 5 we report MNL with utility specifications in the preference- and in the WTP-space in the first two sets of columns, as we did for the previous CE. We then present estimates of a NL model in the third set of columns and a RPL model with utility specifications in the WTP-space in the last set of columns. This last model is the result of a specification search that tested for randomness of taste in various attributes. The only attribute that when associated with the hypothesis of a random coefficient was consistently supported by

Table 4 Estimates of models for primary heating choice: Britain. Attributes Advised by heating engineer Advised by friend St. dev. advised by friend Advised by both Garden dug-up Room for hot water tank Room for fuel storage Contract length 1 year Contract length 2 years Contract length 4 years Energy bill⁎ /mean ln(λ) St. dev. energy bill Maintenance costs Cost of new boiler Final log-likelihood Likelihood ratio test Adjusted rho-square Number of choices Number of individuals

MNL preference-space

MNL WTP-space

Panel RPL WTP-space 1

Panel RPL WTP-space 2

Coefficient

|t-values|

p-values

WTP

St. err.

p-values

WTP

St. err.

p-values

WTP

St. Err.

p-values

0.3141 0.2024

4.84 2.55

b0.001 0.0109

0.2241 0.1445

0.0479 0.0537

b0.001 0.0071

0.1977 0.0622

0.0541 0.0643

0.0003 0.3335

0.3467 − 0.7643 − 0.3103 − 0.6255 − 0.2749 − 0.1380 − 0.2938 − 1.4011

7.02 8.39 7.27 6.57 3.68 2.04 5.35 21.12

b0.001 0.0109 b0.001 b0.001 0.0002 0.0411 b0.001 b0.001

0.2475 − 0.5455 − 0.2214 − 0.4464 − 0.1962 − 0.0985 − 0.2097 0.3373

0.0367 0.0781 0.0307 0.0748 0.0523 0.0489 0.0397 0.0475

b0.001 b0.001 b0.001 b0.001 b0.001 0.0438 b0.001 b0.001

− 0.0026 − 0.4500

6.65 6.78 − 4643.4 1351.6 0.1248 7674

b0.001 0.0109

− 0.0019 − 0.3212

0.0003 0.0413 − 4643.4 1351.6 0.1248 7674

b0.001 b0.001

0.0947 − 0.5186 − 0.2411 − 0.4210 0.0462 − 0.0150 − 0.1508 0.1745 1.5598 − 0.0019 − 0.2011

0.0414 0.0938 0.0409 0.0924 0.0710 0.0616 0.0480 0.0988 0.2047 0.0003 0.0519 − 4535.4 1567.6 0.1449 7674 1279

0.0222 b 0.001 b 0.001 b0.001 0.5152 0.8076 0.0017 0.0772 b 0.001 b0.001 b0.001

0.1785 − 0.0054 0.2474 0.0918 − 0.6235 − 0.2531 − 0.5291 0.0661 − 0.0059 − 0.1615 0.1468 1.7372 − 0.0017 − 0.1823

0.0541 0.0623 0.0343 0.0354 0.0892 0.0406 0.0958 0.0549 0.0621 0.0460 0.0995 0.2037 0.0003 0.0430 − 4535.3 1567.9 0.1448 7674 1279

0.0010 0.9309 b 0.001 0.0095 b 0.001 b 0.001 b 0.001 0.2286 0.9240 b0.001 0.1402 b 0.001 b0.001 b0.001


135

Table 5 Estimates of models for discretionary micro-generation choice: Britain. MNL preference-space Solar electricity Solar hot water Wind turbine Capital cost/mean ln(λ)a St.dev. capital cost Friend Heating engineer Both Maintenance cost Energy savings st.dev. Energy savings Inclusive valueb Final log-likelihood Likelihood ratio test Rho-square Adjusted rho-square Number of observations Number of individuals a b

MNL WTP-space

NL preference-space

Panel RPL WTP-space

Coefficient

|t-values|

p-values

WTP

St. err.

p-values

Coefficient

|t-values|

p-values

WTP

St. err.

p-values

0.9312 0.9547 0.4236 − 0.3288

11.01 10.84 5.15 24.13

b0.001 b0.001 b0.001 b0.001

2.8316 2.9032 1.2882 − 1.1122

0.2441 0.2555 0.2408 0.0415

b 0.001 b 0.001 b 0.001 b 0.001

0.9460 0.9428 0.7797 − 0.1030

24.28 23.94 13.52 3.69

b 0.001 b 0.001 b 0.001 b 0.001

− 0.0697 0.0864 0.1820 − 0.0303 0.0973

1.31 1.43 3.52 5.08 5.20

0.191 0.1513 b0.001 b0.001 b0.001

− 0.2120 0.2626 0.5534 − 0.0922 0.2957

0.1627 0.1834 0.1575 0.0184 0.0590

0.1927 0.1521 b 0.001 b 0.001 b 0.001

− 0.0085 0.0452 0.0748 − 0.0010 0.0003

0.49 1.91 2.73 3.06 2.64

0.626 0.056 0.006 0.002 0.008

2.9357 3.0341 2.4618 −0.1397 1.8162 − 0.2310 − 0.0495 0.3787 − 0.0047 − 0.0011 0.0198

0.4343 0.4494 0.3699 0.1393 0.2549 0.1291 0.1238 0.1221 0.0012 0.0009 0.0013

b 0.001 b 0.001 b 0.001 0.316 b 0.001 0.074 0.689 0.002 b 0.001 0.224 b 0.001

3.5635

2.67 − 7281.07 1386.048 0.08691 0.08566 7280 1241

0.008

− 7328.88 1290.438 0.08091 0.07979 7280 1241

− 7328.88 1290.438 0.08091 0.07979 7280 1241

− 5744.01 4460.17 0.27967 0.27829 7280 1241

Variable scaled by 1000. Test for the null of equal to 1.

the data was “energy saving”, all the other attributes of the microgeneration solution where found not to be significantly subject to random taste variation. In the context of discretionary choices we notice that the source of advice does not significantly impact on utility, regardless of whether the source is a friend or a heating engineer. However, when the suggestion combines the two sources the effect is significant and positive, and worth GBP 553(±157). An energy saving device also produces a significant increase in the probability of choice with a WTP equivalent of GBP 296(±59). In terms of the type of micro-generation technology, respondents are WTP approximately the same for solar plants (electricity GBP 2831 ± 244 and hot water GBP 2903 ± 255), but less than half the amount for wind turbines (GBP 1288 ± 241). The nested logit model increases fit substantially, but gives an estimate for the inclusive value coefficient that is not consistent with random utility theory. Its marginal WTP estimates must be weighted by the probability of selecting a change from the status quo, which is 0.712. Accounting for this in the marginal WTP estimation gives a value estimate for combined advice of a similar magnitude, but the estimates for the type of micro-generation technology appear over twice the figures for the MNL model in the case of solar, and more than three times for the wind turbine. Finally, the panel RPL model with utility in the WTP-space accounts for correlation across choice by the same respondents and for variation of scale and taste intensity. In terms of taste intensity we find that apart from the marginal utility of money–as expressed in terms of capital investment–only the attribute coefficient for “Energy savings in GBP” varies across respondents and with a very small variance, which is nevertheless significantly different from zero. The results are mostly in line with those obtained from the MNL model in the WTP-space, with the only noticeable change in the implied marginal WTP for the wind turbine which is GBP1,173 higher in this model (nearly double), and for the value associated with the joint advice from friend and heating engineer, which is GBP174 smaller (two third of the MNL estimate). 8. Conclusions The results suggest that whilst renewable energy adoption is significantly valued by households, this value is not sufficiently large, for the vast majority of households, to cover the higher capital costs of micro-generation energy technologies, and in relation of annual savings in energy running costs.

The average installation costs for a 25 kWh condensing gas boiler was GBP 2500 in 2008, and GBP 2000 for 11 kWh direct (grid) electric heating equipment; compared to GBP10,801 for an 11 kWh biomass boiler, GBP 10,281 for an 11 kWh ground source heat pump, GBP 8545 for an 11 kWh air source heat pump, GBP 10,638 for a 2 kWh solar PV unit, GPB 4998 for a 1 kWh micro-wind unit, and GBP 3904 for a 2 kWh solar hot water unit (Element Energy, 2008a, page 11 Table 5). These costs are greater than the respondents' WTP for these technologies (GBP 2831 for solar electricity; GPB 2903 for solar hot water; and GBP 1288 for a wind turbine). The estimates from the CE data also permitted the evaluation of the relative importance consumers attached to capital in relation to ongoing energy savings. Consumers were WTP GBP2.91 ± GBP0.30 in capital costs to reduce annual fuel bills by GBP1. This value also applied to solar electricity, solar thermal, and wind power. Such a value implies that consumers' time horizon for cost is generally less than 3 to 5 years (Element Energy, 2008b). This is considerably shorter than microgeneration technology lifetimes of 10 to 25 years. Conversely it also implies that consumers have high discount rates, of approximately 34%. Either way it is consistent with the very slow uptake of micro-generation energy technologies, in the absence of subsidies and regulatory requirements, observed in the UK. It has been estimated that only 95,000 to 98,000 micro-generation units had been installed in the UK by the end of 2007 (Element Energy, 2008b). The vast majority of these units (90,000) were solar thermal units (note the smaller difference between WTP and cost of this technology in the above CE results). The results of this study suggest that the British government will have to give substantially larger grants than those currently available (in 2009), if it is to induce significantly more households to install micro-generation technologies; or conversely the price of microgeneration technologies will have to fall substantially. Some of the cost of installing micro-generation technologies in existing houses is the utility loss to households caused by space requirements for fuel storage, hot water tanks, and gardens dug up to install ground heat pumps. These costs would be reduced by concentrating policy on new houses, where micro-generation technologies could be designed into the house at construction at a lower cost than trying to modify existing buildings. In terms of estimation we also notice that allowing for models not subject to the IIA restriction and allowing for taste variation makes a difference in terms of the implied welfare estimates. In particular, random parameter models in the WTP-space provide estimates of marginal WTP that differ, albeit not for all attributes, from those

136


implied by standard MNL models. From the estimation viewpoint specifications in the WTP-space are obviously more practical when the objective is the derivation of welfare estimates as they offer a more immediate interpretation of the estimated parameters of the utility function. They also account for the issue of inter-personal scale variation that might be confounded with taste variation and with intensity of taste, as pointed out in previous literature (Hausman and Ruud, 1987). Acknowledgements This study was undertaken for the Energy Saving Trust. Helpful comments and suggestions were made by Dr Harsh Pershad and Mr Alex Stewart of Element Energy. All views expressed in this paper are those of the authors alone. References Aravena, C., Martinsson, P., Scarpa, R., 2006. Does money talk? The effect of a monetary attribute on the marginal rate of substitution in a choice experiment. Paper presented at the World Congress of Environmental and Resource Economists, Kyoto, Japan. Banfi, S., Farsi, M., Filippini, M., Jakob, M., 2008. Willingness to pay for energy-saving measures in residential buildings. Energy Economics 30, 503–516. Bergmann, A., Hanley, N., Wright, R., 2006. Valuing the attributes of renewable energy investments. Energy Policy 34 (9), 1004–1014. Bierlaire, M., 2003. BIOGEME: a free package for the estimation of discrete choice models. Proceedings of the 3rd Swiss Transport Research Conference, Monte Verita, Ascona, Switzerland. Borchers, A.M., Duke, J.M., Parsons, G.R., 2007. Does willingness to pay for green energy differ by source? Energy Policy 35 (6), 3327–3334. Breffle, W.S., Morey, E.R., 2000. Investigating preference heterogeneity in a repeated discrete-choice recreation demand model of Atlantic salmon fishing. Marine Resource Economics 15, 1–20. Breffle, W.S., Rowe, R.D., 2002. Comparing choice question formats for evaluating natural resource trade-offs. Land Economics 78 (2), 298–314. Cameron, T., James, M., 1987. Efficient estimation methods for closed-ended contingent valuation survey data. Review of Economics and Statistics, 69, 269–276. Commission of the European Communities, 2008a. Proposal for a Directive of the European Parliament and Council on the promotion of the use of energy from renewable energy sources. 2008/0016, Brussels. 23rd January 2008. Commission of the European Communities, 2008b. Commission Staff Working Document: the support of electricity from renewable energy sources, accompanying document to the Proposal for a Directive of the European Parliament and Council on the promotion of the use of energy from renewable energy sources. 23rd January 2008. Department of Trade and Industry, 2006. Our Energy Challenge: power from the people. Micro–generation Strategy. DTI, London. Ek, K., 2005. Public and private attitudes towards “green” electricity: the case of Swedish wind power. Energy Policy 33 (13), 1677–1689. Element Energy, 2008a. The Growth Potential for Microgeneration in England, Wales and Scotland. Report to the Department for Business, Enterprise and Regulatory Reform, (BERR), London. Appendix. Page 11: Table 5.

Element Energy, 2008b. The Growth Potential for Microgeneration in England, Wales and Scotland. Report to the Department for Business, Enterprise and Regulatory Reform, (BERR), London. Ferrini, S., Scarpa, R., 2007. Designs with a-priori information for nonmarket valuation with choice-experiments: a Monte Carlo study. Journal of Environmental Economics and Management 53, 342–363. Goldberger, A.S., 1991. A course in econometrics. Harvard University Press, Cambridge. Hausman, J., Ruud, P., 1987. Specifying and testing econometric models for rank-ordered data. Journal of Econometrics 34, 83–103. Hensher, D.A., Green, W.H., 2003. The mixed logit model: the state of practice. Transportation 30, 133–176. Hess, S., Train, K.E., Polak, J.W., 2006. On the use of a modified Latin hypercube sampling (MLHS) method in the estimation of a mixed logit model for vehicle choice. Transportation Research B 40, 147–167. Lancaster, K.J., 1966. A new approach to consumer theory. Journal of Political Economy 74, 132–157. Lawrence, C., Zhou, J., Tits, A., 1997. User's Guide for CFSQP Version 2.5: A C Code for Solving (Large Scale) Constrained Nonlinear (Minimax) Optimization Problems, Generating Iterates Satisfying All Inequality Constraints. Institute for Systems Research, University of Maryland. Longo, A., Markandya, A., Petrucci, M., 2006. The Internalisation of Externalities in the Production of Electricity: willingness to pay for the attributes of a policy for renewable energy. Nota di Lavoro 132.2006. Foundazione Eni Enrico Mattei, Venice. McFadden, D., 1974. Conditional logit analysis of qualitative choice data. In: Zarembka, P. (Ed.), Frontiers in Econometrics. Academic Press New York, pp. 105–142. Mitchell, C., Bauknecht, D., Connor, P.M., 2006. Effectiveness through risk reduction: a comparison of the renewable obligation in England and Wales and the feed-in system in Germany. Energy Policy 34 (9), 297–305. OXERA, 2006. Policies for energy efficiency in the UK household sector. Report to Defra, Oxera, Oxford. Poe, G.L., Giraud, K.L., Loomis, J.B., 2005. Computational methods for measuring the difference of empirical distributions. American Journal of Agricultural Economics 87, 353–365. Rivers, N., Jaccard, M., 2005. Combining top-down and bottom-up approaches to energy-economy modelling using discrete choice methods. The Energy Journal, 26 (1), 83–106. Rosen, S., 1974. Hedonic prices and implicit markets: product differentiation in pure competition. Journal of Political Economy 82, 34–55. Samuelson, W., Zeckhauser, R., 1988. Status quo bias in decision making. Journal of Risk and Uncertainty 1, 7–59. Scarpa, R., Willis, K., Acutt, M., 2007. Valuing externalities from water supply: statusquo, choice complexity and individual random effects in panel kernel logit analysis of choice experiments. Journal of Environmental Planning and Management 50 (4), 449–466. Scarpa, R., Thiene, M., Train, K., 2008. Utility in WTP space: a tool to address confounding random scale effects in destination choice to the Alps. American Journal of Agricultural Economics 90 (4), 994–1010. Simpson, D., 2004. Tilting at Windmills: the economics of wind power. Hume Occasional Paper No. 65. The David Hume Institute, Edinburgh. Sonnier, G., Ainslie, A., Otter, T., 2007. Heterogeneity distributions of willingness-to-pay in choice models. Quantitative Marketing Economics, 5, 313–331. Train, K.E., 1998. Recreation demand models with taste differences over people. Land Economics 74 (2), 230–239. Train, K.E., 2003. Discrete Choice Models with Simulation. Cambridge University Press, Cambridge. Train, K.E., Weeks, M., 2005. Discrete choice models in preference space and willing-topay space. Chapter 1 In: Scarpa, R., Alberini, A. (Eds.), Applications of simulation methods in environmental and resource economics. Springer Publisher, pp. 1–16.