Household Behaviour and Environmental Policy: Residential Demand ...

Household Behaviour and Environmental Policy: Residential Demand for Renewable Energy

by

Bengt KRISTRÖM SLU University, Sweden Paper prepared for the:

OECD Conference on „Household Behaviour and Environmental Policy’ organised by the

Environment Directorate

3-4 June 2009 OECD Headquarters, Paris

TABLE OF CONTENTS ABSTRACT ....................................................................................................................................................3 1. 2. 2.1 2.2 2.3 2.4 3. 4. 5. 5.1 5.2 5.3 6. 7.

Introduction .......................................................................................................................................4 Updated literature review ..................................................................................................................5 Voluntary behaviour and demand for green energy ..........................................................................6 Studies on the price premium ............................................................................................................6 The price premium across institutional arrangements.......................................................................7 Studies of the price premium in related fields...................................................................................8 Data description and corroboration ...................................................................................................9 Methodological aspects and estimation techniques ........................................................................14 Results .............................................................................................................................................17 On household decisions to undertake special measures to buy renewable energy ....................17 Willingness to pay for renewable energy ..................................................................................23 The decision to enter the market................................................................................................31 Policy implications ..........................................................................................................................34 Summary and Conclusions..............................................................................................................35

REFERENCES ..............................................................................................................................................37 LIST OF TABLES Table 1 - Which of the following sources of energy do you use in your primary residence? ..........9 Table 2 - Does your household take special measures to buy renewable energy from your electricity provider? ........................................................................................................10 Table 3 - Please state why you do not buy renewable energy........................................................11 Table 4 - What is the maximum percentage increase on your annual bill you are willing to pay to use only renewable energy? * ..............................................................................12 Table 5 - Has your household installed any of the following items over the past ten years in your current primary residence? .....................................................................................14 Table 6 - Econometric Models .......................................................................................................16 Table 7 - “Estimation results for Q67: Does your household take special measures to buy renewable energy from your electricity provider?” .................................................19 Table 8 - Estimation results for individual countries, Logit model for Q67: Does your household take special measures to buy renewable energy from your electricity provider? .........................................................................................................................22 Table 9 - Parameter estimates Weibull model for WTP > 0 ..........................................................25 Table 10- Confidence intervals for significant (5%) parameter estimates Weibull model for WTP > 0*...................................................................................................................26 Table 11 - Conditional Mean and Median WTP estimates by country ...........................................28 Table 12 - Model summaries for Weibull model applied to each country* ....................................30 Table 13 - WTP for the three Swedish regions ...............................................................................31 Table 14 -Logit model for the decision to enter the market w/wo the income variable ...........32 Appendix ............................................................................................................................................40 A1. Econometric approach ...............................................................................................................40 A2. Economic model ........................................................................................................................40

2

Residential Demand for Renewable Energy

RESIDENTIAL DEMAND FOR RENEWABLE ENERGY

ABSTRACT

This paper sheds some empirical light on residential demand for renewable energy, using the OECD 10 country web survey. I also update of my previous survey of household energy use, now focussing on “green” electricity. I explore two sets of questions. First, how much are households willing to pay to use only renewable energy? Does willingness-to-pay (WTP) vary significantly across household groups (such as over households in different income groups)? And secondly, how do general attitudes towards the environment (environmental awareness; membership in environmental organization, etc) influence demand for renewable energy? I find that, while there is significant variation across countries, the respondents display a price premium of less than 4% of their current electricity bill, as an average. If those who do not wish to pay are excluded, the average is about 7%. These findings are well within the range of results reported in recent research. Regarding determinants of the price premium, a consistent message across the models used is that environmental awareness and activity/membership in environmental organizations affects the demand for “green” energy. This relationship is found in several other studies as well. Income positively affects the probability of entering the market. There is a stronger statistical link between income and the decision to enter the market, compared to the link between income and the level of the price premium. The policy implications are mainly three. First, the significant support given to the introduction of renewable energy in many countries contrasts in an interesting way with the fairly weak demand reported in this and several other studies. Second, the literature on residential energy demand suggests an important role for incentive-based policy instruments, yet I find a more substantial role for “softer” policy instruments relative to what earlier assessments of policy instruments have found (my previous survey included). This conclusion is based on the literature reviewed and my empirical results. Third, a recurrent theme in this cross-country analysis is country heterogeneity, which here means that there are differences between households that cannot be attributed to explanatory variables in the econometric analysis. I also explore potential “within-country” heterogeneities, using Sweden as an illustrative example. Data suggests an intriguing “urban/rural” (more precisely mid-South/North) asymmetry regarding the reported price premium, portraying a higher/lower price premium for urban/rural group, all else equal. I argue that the revealed type of heterogeneities have some bearing on the shaping of future energy-environmental policies.

3

1.

Introduction

The Household Behaviour and Environmental Policy project is policy-oriented, attempting to inform the design of environmental policies. The sub-project on residential energy demand attempts to further our understanding of fundamental drivers of demand in this sector. In the previous empirical review (Kriström (2008)) I argued that households invariably respond to economic incentives; at the same time, the response is highly variable (in terms of price- and income elasticities). The review also suggested that regulatory instruments indeed can affect residential energy demand. Finally, it concluded that the importance of “softer” instruments for residential energy demand is unclear. This paper reports upon the next step in the project and sheds some empirical light on residential demand for renewable energy, now focussing on “green” electricity. I use the OECD 10 country web survey for the empirical analysis and update my previous survey of household energy use with the new focus in mind. The main questions addressed are1: 1. How much are households willing to pay to use only renewable energy? Does willingness-to-pay (WTP) vary significantly across household groups? 2. How do general attitudes towards the environment (environmental awareness; membership in environmental organization; …) influence demand for renewable energy? The paper is structured as follows. Section 2 has an updated literature review, mainly addressing willingness to pay for “green” electricity. Section 3 begins the empirical analysis by data description and corroboration. Section 4 provides an overview on the econometric approach. Section 5 has estimation results. Section 6 discusses policy implications and section 7 concludes. A technical appendix provides additional details about the methods used. Box 1: Definition of renewable energy

The survey defines renewable energy as energy from renewable sources. There are many other definitions, e.g. those used in some electricity certificate systems that excludes old hydropower and certain bioenergy. There are also a number of definitions on “green electricity”, by e.g. NGOs or sellers of electricity. The survey uses the most general definition of renewable energy.

1

There is too limited information in the returned surveys to be able to make much progress on the question about who takes advantage of grants to install/use renewable energy.

4


2.

Updated literature review

The previous survey (Kriström (2008)) summarized a number of salient points about residential energy demand, which for convenience are repeated here: 1.

Demand for energy is generally quite price-inelastic. There is some consensus on the short-run price-elasticity being about 0.3. The long-run price elasticity might be 0.7. Thus, over the longrun energy demand responds to price in a non-negligible manner. Economists have been more optimistic than many other researchers about the price response.

2.

Demand for energy responds to income, but the response varies substantially across studies. If a number must be singled out, a not unreasonable choice would be close to unity and lower in the short-run. More recent estimates tend to push these figures downwards.

3.

Price- and income elasticities vary across datatype (time-series,cross-section,panel), methodology, time-period and short-run vs long-run. Thus, it might be dangerous to use average elasticities when trying to judge a demand elasticity in a particular case.

4.

According to the mainstream economic view, income encompasses a large number of factors that superficially seems to affect demand. While additional appliances increase energy demand, they were bought because of income increases.The same mechanism might well explain variations across demographic factors. For example, because age and income are usually strongy positively correlated, it is difficult to disentangle their respective impact.

5.

Demand for energy depends on a host of exogenous factors, most importantly temperature.

6.

Attitudes such as “feelings of obligation” , “importance of conservation” as well as “comfort and health concerns” have been found to correlate with energy conservation behavior. Psychologybased studies show mixed results; in some studies attitudinal variables are important, in others not. An open question is still whether the "green consumer" will dominate consumption patters to such a degree than environmental policies are not needed.

7.

Mass information (e.g. “please turn off the lights”) have limited success. See, however, Reiss & White (2008). There is a large literature in psychology on targeted information again with varying success rate

8.

Insofar as the impact of demographic variables on energy consumption can be detached from income influence, empirics suggest that energy consumption varies over the life-cycle, between ethnic groups and cultural practices.

9.

Energy policy tends to have regressive impacts, because energy budget-shares decrease with income. To substitute away from higher energy prices, households may have to install capital-intensive equipment. The extent to which households across income-groups can use the capital markets have implications for the regressivity of energy policy.

With these points in mind, let us now turn to the literature on “green” electricity. At the outset, I should make it clear that the subject is studied in several disciplines and I cannot, for many reasons (including space), provide a fair summary of what has been done in all literatures. My approach is rather to discuss a number of studies in more detail. I proceed as follows: first, I briefly discuss some evidence on voluntary behavior and the extent to which the market economy can solve environmental problems without governmental intervention. I then present evidence from the empirical literature on the price premium for “green electricity”. I continue by

5

looking at this price premium across institutional arrangements (hypothetical vs. real payments) and round off the updated review by adding insights from other social sciences. 2.1

Voluntary behaviour and demand for green energy

Beginning the literature review at a general level, there is a natural connection between the (environmental) economics literatures on voluntary behaviour and the demand for green energy; under which conditions does the (residential) energy market “go greener” on its own? The idea that consumers voluntarily accept extra costs for “going green” contradicts a basic tenet of environmental economics, namely that externalities require government intervention in the market. 2 Yet, some of the earlier research showed that there are conservation efforts beyond those that are solely explained by price/income changes (hence behaviour is probably related to attitudes or some other „non-economic‟ factors).3 An example of recent conceptual papers on voluntary behaviour is Kotchen & Moore (2007). They present a model that seeks to explain two types of “deliberate green behaviour” within a utility-theoretic framework. First, they consider the case when a person voluntary restricts his consumption of a good that causes a negative externality. Second, they explore a model that sheds some light on why a person voluntarily would pay more for a good that is “greener”, a good example being “green electricity”. Thus, there are types of voluntary “green” behaviour that is not inconsistent with neoclassical economic theory. 2.2

Studies on the price premium

A significant number of papers try to place a value on the price premium for renewable electricity.4 The price premium is a natural concern for those who sell electricity; it is imperative to know something about the location and slope of the demand curve. This knowledge is presumably considered a business secret in many cases, so there might well be a significant company knowledge base. I limit myself to some of the scientific literature and begin by looking at research from the US. Farhar (1999) use surveys of US utility customers to derive a kind of demand curve that we will estimate later in section 5. Briefly, her data suggests that some 70%, 38% and 21% of the utility customers would pay at least USD 5, 10 and 15 more per month for “green” electricity. Zarniku (2003) looks at willingness to pay for utility investments in Texas, focussing demographic parameters. He finds that age, education and salary are significant explanatory factors (using bivariate analysis and a Tobit model). Jensen et al (2004), using data from Tennessee, notes that households typically have different preferences across types of renewable energy. Their study tried to unlock those preferences and to estimate the price premium on renewable electricity. To estimate demand, they sold blocks of “green” power in a hypothetical referendum. Out of 421 respondents, about 38% displayed a positive premium. The respondents‟ value wind/solar energy higher than electricity generated from bio energy (12-15 USD per month versus 6 USD per month on the average). Roe et al (2001) use a sample of 1001 respondents from eight US cities in a choice experiment that gave 835 usable responses to a question about the choice between two energy programs. The programs are described by attributes price, contract length and fuel mix; thus marginal prices for each attribute can be estimated. Information from the supply side regarding 2

Coase argued that the problem of externalities can be solved via negotiations between the affected parties, without the need for governmental intervention.

3

In my earlier survey, I discussed the research by Reiss & White, who used modern tools to disentangle the heterogeneous response to price signals in the energy market. They also shed some light on the impact of other signals, in particular public appeals (Reiss & White (2005)). See Reiss & White (2008) for additional analysis of the California energy crisis and the impact of public appeals.

4

A survey in 1996 reported 700 willingness to pay studies on renewable energy only in the US in the period 1973-1996 (Longo, Markandya & Petrucci (2006, p. 8)).

6


premium, fuel mix, and certification was used in a hedonic regression to “explain” the variation of the charged premiums between suppliers. The regression equation is used to predict the marginal price premium in the market as new renewable sources are added. Interestingly, the estimated marginal price is in the middle of the ranges obtained from the hypothetical survey experiment. See also Byrnes et al (1999), Champ & Bishop (2001), Ethier et al (2000), Farhar & Coburn (1999), Whitehead & Cherry (2007) and Wiser (2003) for further examples from the US. There are also quite a few studies on the subject from the UK. Diaz-Rainey & Ashton (2007) use a sample of the UK population to explore demographic, attitudinal and behavioural characteristics. They find that there is a correlation between WTP for “green” electricity and attitudinal factors (as well as income). Interestingly, one significant variable is “willingness to invest in energy efficiency appliances”. A similar variable is used in our analysis, see section 5. Other studies from the UK include Fouquet (1998), who finds a 20 % participation rate in a “green power” program, while Batley et al (2001) report a 34% participation rate. Hanley & Nevin (1999) presents a significant value of wind mill electricity. Longo, Markandya & Petrucci (2006) use choice experiments to investigate household preferences across electricity generation technologies; their main focus is, however, on preferences across renewable energy policy packages. The packages are assumed to: help internalizing external costs; affect energy supply security and employment in the energy sector. Each package comes at a particular increase of the electricity bill.5 One of the hypotheses tested is whether or not jobs creation matters when choosing between renewable energy packages. The authors find that the jobs creation issue is important to respondents (this link is rejected in other studies; see Longo, Markandya & Petrucci (2006) for further discussion). A final example of the literature on valuation of green electricity is taken from Sweden. Kristina Eks (2005) doctoral dissertation provides five papers on supporting renewable energy in Sweden. Among other things, she tries to expand upon the conventional economic analysis by integrating norms into behavioural models. In one of the empirical analyses, Ek & Söderholm (2008), the authors look at willingness to pay for wind power using data on 655 Swedish households sampled from 4 municipalities. The authors find that income is a key variable in determining the decision to “go green”. Still, attitudinal variables such as “perceived responsibility” also play an important role. The analysis gives little support for the hypothesis that the perception of other‟s behaviour affects individual moral norms and behaviour. Although the survey is potentially marred by a low response rate, it is found that there is a “general lack of trust” in the green electricity scheme. This is an interesting and potentially useful finding.6 2.3

The price premium across institutional arrangements

Given information about the price premium, a next useful subdivision of this literature is the analysis of price premiums across institutional arrangements. The most interesting comparison is between hypothetical and real payments (or intentions). There is a very large amount of research on this issue, not the least in environmental economics. A very substantial advantage here is that the market continually works as an error-correction mechanism, so that we continuously obtain information about how much households are observed to pay. In most cases in environmental valuation, there is no such error-correction mechanism; the estimated valuations must be assessed face value. An important paper in the empirical analysis of price premiums across payment institutions is Cameron et al (2002). The authors combine a telephone survey with six mail surveys in cooperation with a power company in New York State. Respondents were offered, in a real or hypothetical setting, a price premium for the company to plant trees and/or provide energy from renewable sources. An innovative 5

Their sample is restricted to a part of southern England, which makes aggregation an issue.

6

For an example from Australia, see Ivanova (2005).

7

feature of this research is to include several different elicitation methods within a single framework that allows real value elicitation. The results support the notion that there is a difference between real and hypothetical behavior, a point buttressed by several other papers (e.g. Roe et al (2001), Rose et al (2002), Kotchen & Moore (2007), Wiser et al (2001) and Ek & Söderholm (2008)). My impression is that this point is also well-known among power companies. 2.4

Studies of the price premium in related fields

“Green” electricity and consumer behaviour has been studied by several other disciplines and I provide a very small sample of the activity. Pichert & Katsikopoulos (2007, p.63) propose that “…although green electricity is available in many markets, people do not generally buy it”, the reason being a kind of status quo bias. Thus, if “grey electricity” is offered as a default, few would switch to “green electricity”. Pichert & Katsikopoulos (2007) give the interesting example of Schönau, a small German town where virtually everybody uses “green” electricity, following a referendum decision on whether or not (the use of) nuclear-power should be abolished.7 According to the authors, a Schönau citizen will need to exercise some effort in finding “grey” electricity; indeed, 8 years after the referendum decision, very few customers have made the switch. The authors present events in another German town, Wustenhagen, as another case in point. In a survey, 150,000 Energidienst GmbH customers were asked to make a choice between (slightly cheaper) “grey” electricity, substantially more expensive “super-green” electricity, and the status quo (a default “green” alternative). Two months after the request to make a choice, 94% preferred the status quo. Two small laboratory experiments are also offered to further drive home the point. An attempt to integrate theories from economics and psychology is presented in Clark, Kotchen and Moore (2003). They use a mail survey to extract information about participants in a premium priced “green” electricity program. The authors contrast “internal variables” that essentially measure attitudes, and what they call “external” variables such as income and other socio-demographics. They find that two variables in each category are significant in explaining the entry decision into the green market. The results are based on a mail survey to 281/619 participants/non-participants in the Detroit SolarCurrent program. Respondents were asked about their motives for enrolling/not enrolling in the program, as well as a set of questions addressing altruism and attitudes towards environmental issues using the so-called New Ecological Paradigm (NEP) scale. A logit model is employed to explore determinants of the participation decision. The authors are able to rank the importance of motives for participating, finding that egocentric motives rank lower than altruistic ones. This partly concludes the literature review. I will prolong the review of earlier research by comparing the results obtained from the web survey. Thus, we now turn to the empirical analysis, beginning with simple summaries of the data and then turn to econometric modelling.

7

See http://www.ews-schoenau.de/ews.html for current information (in German) about the growth of the local utility “Elektrizitätswerke Schönau”, now providing nuclear-free power (“rebel strom”) to around 80.000 customers.

8


3.

Data description and corroboration

I begin the analysis with some simple summaries of the data at hand.8 This will be followed in section 5 by more detailed analysis, adding structure by means of a sequence of econometric models. Together, these two sections paint one out of many possible pictures of the responses garnered in the survey. Let us first consider energy demand in terms of question Q64 (“Which of the following sources of energy do you use in your primary residence?”) Table 1 - Which of the following sources of energy do you use in your primary residence?

Canada Netherlands France Mexico Italy Czech Republic Sweden Norway Australia Korea

N 1003 1015 1075 1009 1417 701 1006 1019 1006 1000

El 0.87 0.99 0.85 0.99 0.95 0.94 0.77 0.96 0.99 0.73

NGas 0.49 0.92 0.49 0.38 0.77 0.62 0.01 0.02 0.50 0.36

Fueloil 0.10 0.00 0.13 0.42 0.13 0.01 0.02 0.10 0.01 0.08

Wood 0.09 0.04 0.15 0.04 0.19 0.24 0.14 0.58 0.12 0.01

Coal 0.00 0.00 0.00 0.03 0.02 0.14 0.00 0.00 0.01 0.00

Distr.Heat 0.01 0.07 0.06 0.02 0.04 0.44 0.33 0.05 0.00 0.32

OECD-10

10251

0.90

0.46

0.11

0.16

0.02

0.12

The answers to Q64 provide an interesting overall picture of residential energy demand. At one extreme, 99.4% of respondents in Mexico marked electricity as one possible source, while at the lower extreme, 72% made this statement in Korea. As expected, electricity is a dominating energy source in all surveyed countries. We also expect to find that natural gas is used extensively in the Netherlands and in Italy, which indeed is the case. Fuel oil seems mainly to be an important source of residential energy demand in Mexico, while wood is mostly popular in Norway (presumably this mostly pertains to summer houses). Coal is not used much, except in Czech Republic, where 14% of the respondents claimed to use this energy source. Note also the popularity of district heating in the Czech Republic. District heating is extensively used in Sweden, but the reported numbers are smaller than expected, as noted above. Regarding the number of different energy sources used (not shown), Swedish households dominates in terms of single-sourced primary residencies. Respondents in other countries typically used two main two energy sources (Korea being the other exception). Looking more closely at the data, I find 180 Swedes reporting that both electricity and district heating is used.9 This suggests that Q64 has been interpreted in different ways, because electricity and district heating is not used as primary energy sources in combination. It seems as some Swedes interprets the question as asking for the heating source of energy, while others report their total energy use. At any rate, the reported 77% on electricity use in Sweden is arguably due to the formulation of Q64 (South Korea‟s

8

All computations are done using R: R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org.

9

205 Korean respondents and 293 Czech Republic respondents made the same statement.

9

73% also seems low).10 Out of the 233 who did not mark electricity as an energy source, the overwhelming majority reports using district heating. Furthermore, a common answer is that a heating pump is used, an answer supplied by 57 Swedes.11 Consequently, in a future survey, Q64 should be re-worded and contain two alternatives (electricity used for heating/cooling and for other purposes). For the record, 91 respondents in total choose heat pump; the only other country where it is used (in a notable way) is Norway. Of course, this pump requires electricity, but the respondent crossed the box “Other, please explain”, rather than both electricity and “Other, please explain” in Q64. The survey provides no quantitative information on renewable energy demand in the residential sector, but contains other information that allows detailed analysis. The principal information comes from Q67, Q68 and Q69. We look at these in turn. Consider first whether or not special measures has been undertaken by the household, as per Q67 (“Does your household take special measures to buy renewable energy from your electricity provider?”). Observe that the response to Q67 is conditioned on the use of electricity, which considerable reduces the number of answers from Sweden and South Korea.12 1480 (out of the 9265 usable responses) of the respondents claimed to have undertaken special measures; an overwhelming majority (6280) denied such activity (the remaining 1505 “did not know”). There is considerable variation across countries and section 5 explores these patterns further, within an econometric model. The data on activity is summarized in Table 2. Table 2 - Does your household take special measures to buy renewable energy from your electricity provider?

Canada Netherlands France Mexico Italy Czech Republic Sweden Norway Australia Korea OECD-10

Yes

No

Do not know

0.10 0.43 0.05 0.11 0.09 0.04 0.16 0.08 0.21 0.33 0.16

0.67 0.47 0.85 0.71 0.83 0.76 0.57 0.73 0.65 0.43 0.67

0.23 0.09 0.10 0.18 0.08 0.20 0.27 0.18 0.14 0.24 0.17

Thus, the respondents from the Netherlands are most active, with 43% responding in the affirmative on Q67. This is consistent with a finding reported in Bird et al. (2002).13 Interestingly, 24% of the Korean respondents (only topped by the 27% of the Swedes) do not know if they have taken special measures to

10

With hindsight, the translation of Q64, which I did myself, into Swedish should have added extra information to separate out heating from energy use. At some level, the Q64 alternative “electricity” is redundant, but it is useful e.g. in that many Swedes heat their houses directly with electricity radiators.

11

I thank Dr. Milan Scasny for compiling this data.

12

This is hard-coded in the survey, so recoding Q64 does not help. Those who responded “Other, please specify”, did not get to answer Q67.

13

For a discussion about the policies employed in Netherlands in this context, see e.g. van Rooijen, & van Wees (2006).

10


buy renewable energy. France tops the scale at the other end, with 5% of the respondents claiming that they are taking special measures to buy renewable energy. Q68 (“Please state why you do not buy renewable energy”) lets us understand in more detail why a respondent said “no” to Q67 and provides some additional insights into the potential “information gap” (or ignorance, as the case may be). Table 3 has the summary of Q68 across countries. Table 3 - Please state why you do not buy renewable energy.

Canada Netherlands France Mexico Italy Czech Republic Sweden Norway Australia Korea OECD-10

NA, not intr 0.09 0.04 0.11 0.09 0.11 0.07 0.09 0.10 0.07 0.07

NA,intr 0.47 0.03 0.37 0.74 0.46 0.41 0.22 0.32 0.27 0.67

Avail., not intr 0.04 0.53 0.03 0.01 0.04 0.05 0.12 0.01 0.22 0.02

Already Avail. 0.09 0.19 0.04 0.02 0.02 0.04 0.05 0.25 0.08 0.03

Do not know 0.30 0.21 0.44 0.15 0.37 0.42 0.51 0.32 0.36 0.21

0.08

0.40

0.11

0.08

0.33

Consider the first row in the table, which corresponds to the alternative: “service not available and our household is not interested”. The proportion is remarkably consistent across countries. In the next row I collect respondents who have no access, but are interested in using the service. The numbers harmonize rather well with the literature reviewed above, i.e. there is substantial “hypothetical” interest. Or, if taken face value, we can interpret the numbers as suggesting a latent demand. Consistent with the results in the Bell et al (2002) survey, the two next rows show that Dutch respondents score high on availability, but also high on “not interested”14. Because Norwegian electricity generation is essentially only from hydropower, Norwegian respondents could also score high on row 4 in the table, but note that many Norwegian respondents considered renewable energy not available.15 The last row in the table shows that there may be an information gap to be filled. Many respondents claim that “I don‟t know anything about these kinds of services”. The highest percentage is reported for Swedish households. Given the fact that the discussion has been rather lively in Sweden, this is curious16. I will return to Q68 when I look at the price premium in section 5. We now switch to one of the key questions of this report. Question 69 asks “What is the maximum percentage increase on your annual bill you are willing to pay to use only renewable energy? Please assume that your energy consumption remains constant” As we have just seen, there are households who seem unaware of renewable energy services, yet others who are not interested. One might argue that without knowledge of the good in question, a person cannot possibly state a meaningful WTP. This concern is all the more legitimate when trying to estimate a “demand curve” for green electricity in section 5. Consequently, I will return to this issue in section 5.

14 15 16

For further discussion about the Dutch experience, see Arkesteijna & Oerlemans (2007). Cross-border trading means that the electricity actually used can come from other sources. I offer no empirical support for my sentiment. Still, the comparatively harsh winters and the significant share of energy-intensive industry places energy issues high on the table.

11

In the appendix, I present a simple theoretical model that interprets Q69. The model highlights the fact that the individual pays for an implied environmental improvement. It is to be noted that this improvement is not described in any detail in Q69. The question is not intended to mimic the valuation questions used in a rigorous contingent valuation survey. Rather, I consider Q69 as portraying a widely available good (“green electricity”) and a scenario of intense current discussion in many countries. It could provide useful insights into sentiments about the future development of the energy systems held by households in the survey. Figure 1 summarizes the responses to Q69. Figure 1 – WTP for renewable energy

I have drawn the curve so that it resembles a demand curve for a public good. Thus, I linearly interpolate between the observed proportions (denoted by circles in the figure). It helps interpreting the coming econometric results, by thinking about the estimated parameters in terms of shifters of this kind of demand curve. The findings in Figure 1 are somewhat similar to Jensen et al (2004, p.13), their main point being that previous studies show a significant market participation rate in a hypothetical setting, while they obtain a much more conservative estimate (38%). In this survey, we obtain a market participation rate of about 54% (if the “don‟t know “responses are deleted). For completeness, table 4 provides the proportions broken down by countries.17 I now include the “don‟t know” answers, although they will be deleted in the econometric analysis below. Table 4 - What is the maximum percentage increase on your annual bill you are willing to pay to use only renewable energy? *

Canada Netherlands France 17

Zero

< 5%

5 - 15%

16 - 30%

> 30%

Do not know

0.33 0.64 0.43

0.23 0.16 0.26

0.16 0.07 0.11

0.03 0.00 0.02

0.01 0.00 0.01

0.24 0.13 0.16

In a loose sense, this is a non-parametric estimator of the survival distribution for WTP.

12


*

Mexico Italy Czech Republic Sweden Norway Australia Korea

0.20 0.37 0.30 0.47 0.43 0.37 0.29

0.28 0.26 0.27 0.13 0.17 0.27 0.34

0.29 0.18 0.19 0.16 0.19 0.18 0.17

0.10 0.04 0.02 0.03 0.03 0.03 0.02

0.03 0.01 0.01 0.01 0.02 0.01 0.01

0.10 0.15 0.22 0.22 0.16 0.15 0.16

OECD-10

0.38

0.24

0.17

0.03

0.01

0.17

Please assume that your energy consumption remains constant.

The tendency is clear also at the country-level; the higher the price, the lower the proportion that accepts to pay it. There are some minor deviations from this pattern, but overall the responses are consistent and according to expectation. By way of comparison, Diaz-Rainey & Ashton (2007) find for their UK sample that “In 2003 42% of respondents either strongly agreed or agreed with the dependent variable statement that they would be willing to pay a premium of between 5 to 10% for green energy… … from the survey it could be implied that only 0.3% of respondents has actually switched to a green tariff “(Diaz-Rainey & Ashton (2007, p.14). We can compare these results with the proportions in the 5-15% range. We see that there is significant variation across countries, with Dutch respondent displaying a low 7%, the corresponding number for Mexican respondents being a high 29%. It might also be of interest to compare these results with a Financial Times/Harris poll conducted in February 2008.18 The poll used an online survey of 6,448 adults 16 - 64 years old in France, Germany, Great Britain, Spain and the United States, and adults aged 18 to 64 in Italy. The question asked was similar to Q69: “How much of an increase would you be willing to pay at the most for energy if it were from renewable sources?" The Harris poll reports 43% zero WTP for the French respondents and we obtain a very similar number here.. For the Italian respondents, the corresponding numbers are also rather similar (44% Vs 37%). It is somewhat more difficult to compare the other brackets, because the Harris poll uses the construction “5% more”, “10% more” and so on.

18

The Harris Poll® #21, February 26, 2008. http://www.harrisinteractive.com/harris_poll/index.asp?PID=875.

13

Let us now turn to energy savings behaviour and consider one of the questions that shed some light on this. Q73 asks “Has your household installed any of the following items over the past ten years in your current primary residence?” The responses are collected in table 5. Table 5 - Has your household installed any of the following items over the past ten years in your current primary residence?

Yes

No

Already installed

Not possible

Energy saving apparel

0.54

0.26

0.15

0.05

Energy saving lamps

0.69

0.19

0.11

0.02

Heat insulation

0.26

0.39

0.20

0.15

New heating pump

0.14

0.53

0.11

0.22

Installed renewable energy

0.05

0.66

0.02

0.27

This table hides significant country heterogeneity. For example, Low-energy light bulbs (compact fluorescent) have been installed over the past decade in 84% of the Italian respondents‟ homes, while the corresponding percentage for Korea is 28% (the difference may be due to the fact that 37% of the Korean‟s have already installed such lamps). Our main focus, of course, is on renewables and we simply note that the proportion of households that has installed solar panels, wind turbines and other renewable energy technologies is in the order of 5% (Norwegian respondent‟s tops at 15%). The follow-up question is of much interest to us here, namely whether or not the household has benefited from government support (for instance grants, preferential loans, energy audits). There are 1-2 % of the respondents in each country that has received support for renewable energy. I turn now to the econometric modelling in an attempt to uncover additional pieces of information about the residential energy market. The next section discusses some methodological aspects of the modelling, followed by estimation results in section 5. 4.

Methodological aspects and estimation techniques

The formal econometric modelling is used to add additional information about the questions I set out to address. A word of caution is in order; we have a very large data set with many potentially correlated variables. Correlation does, of course, not imply causation. More importantly; in a data set of this size we are almost certainly bound to find “significant” variables.19 We have also seen that there is room for misunderstanding (c.f.the discussion of Q64) and there are undoubtedly information gaps (as the analysis of Q67 suggested). Conventionally, an econometric specification search embodies a search across different models, probability structures and so on and so forth. Here, however, I will confine the analysis to a small subset of possible models. The approach used here, essentially the standard in social sciences, have been forcefully critiqued by some statisticians (and defended by still others). I let the data speak, as many social scientists are wont to. In astronomy, physics leads to precise relationships and fairly well understood error terms. For many good reasons, we do not have the same situation in the social sciences. I mention these issues, so that

19

A recent paper indirectly heeding this warning is Heckman, J. (2008) “The Effect of Prayer on God's Attitude Toward Mankind”, Institute for the Study of Labor (IZA), Bonn, August 2008.

14


the limits of our empirical analysis are squarely put on the table, notwithstanding the use of sophisticated statistical models.20 Furthermore, a specification search typically involves selecting a parsimonious model. This is a complex process with several dimensions. For example, should we prefer a small model that fits the data worse than a more complex larger model? Ideally, we would have a strong theoretical model that pins down all variables and subject it to the data. As noted, I do not follow this path in all cases reported below. Rather, my preference is to find the smallest model that can usefully explain salient features of the data. This leads me to using a particular model reduction approach, described in section 5. I consider three types of models that heuristically can be written as: 1. Response = Country independent constant and marginal effects + error 2. Response = Country specific constant + country independent marginal effect +error 3. Response = Country specific constant +country specific marginal effect+error Because the models that I actually employ are inherently non-linear, this discussion is just a simple way of loosely retaining the intuition from the usual linear regression analysis. I only consider models that have linear response functions, but allow progressively more complex models in terms of allowing for differences between countries. In the first type, I assume homogeneity. In the second, I allow for what in the econometrics literature is called “fixed effects”, or constant country-specific differences. The most general type of model is type 3, where I allow for fixed effects and country specific marginal effects in all countries. Furthermore, given the way I carry out the econometric analysis, there is a subtle difference between type 3 and the other two types. In type 3 models, I estimate one equation separately for each country, while in type 1 and 2 only one equation is estimated. Keeping the linear regression analogy, this is equivalent to making different assumptions about the error term. Thus, in models 1 and 2, a standard linear regression entails assuming that the error term has the same constant variance in all ten countries. In model 3, this assumption is relaxed. It might seem that type 3 models should be preferred, because the simpler versions are subsets of that type. However, if ones place significant weight on model parsimony, model type 2 could be preferable to model type 3; a simpler model that can explain about as much of the variation in the data compared to a more complex model is usually preferred. The sense in which a shorter (fewer parameters) model can be preferred to a longer model will be further discussed below, but before summarizing salient features of the models I use, there is an additional issue of importance to discuss, namely missing data. If there are missing values in the data matrix, the order of deleting “insignificant variables” can make a difference. Model shortening is less of a problem in an orthogonal experiment, where there is no missing data. In the present case, the data-matrix is not orthogonal in this sense; there are missing values on some variables (although this is a smaller problem than I expected). This means that the variable selection process in our case is somewhat more complex than in an orthogonal experiment. The following table summarizes the econometric models, their primary objective and the basic probability model used. 20

For an alternative view, supporting data-mining (carefully defined) see Greene, C. (2000) “I am not, nor have I ever been a member of a data-mining discipline”, Journal of Economic Methodology 7:2, 217–230.

15

Table 6 - Econometric Models

Question Q67 "Does your household take special measures to buy renewable energy from your electricity provider?"

Objective Explore potential drivers of the decision to be active in the renewable energy market (electricity)

Response Binary (yes/no)

Q69. I "What is the maximum percentage increase on your annual bill you are willing to pay to use only renewable energy?

Estimate the "demand curve" for renewable energy (electricity)

Interval censored (willingness to pay belongs to known intervals)

Q69.II " Are you willing to pay anything?" (inferred from the zero alternative in Q69.I)

Analyze the decision to enter the market for renewable energy

Binary (willingness to pay zero/ greater than zero)

Model Type 1. Logistic ("Standard 'bell'-shape type ") 2. Cauchy ("Fat-tailed") 3. Complementary log-log ("Asymmetric")

Weibull =

Pr(T  t )  exp( exp(

log(t )  X



))

1. Logistic ("Standard 'bell'-shape type ") 2. Cauchy ("Fat-tailed") 3. Complementary log-log ("Asymmetric")

Thus, the response variable in the econometric models used below is either binary or interval censored. The binary data is primarily analyzed with a standard logistic model using maximum likelihood. It is well-known that the logistic gives almost the same result as the Normal distribution in binary response models, the difference being that the logistic has slightly fatter tails. The use of the logistic is based on a widely accepted idea; while there is no particular reason why the logistic (Normal) should fit any particular data-set well, there is no strong a priori reason to choose any other single distribution. Still, because maximum likelihood is sensitive to distributional assumptions (see e.g. Cosslett (1983)), it is prudent to widen the class of probability models. I have added two distributional assumptions. The Cauchy distribution is symmetric with fatter tails than the logistic, while the complementary log-log is asymmetric.21 By using these additional models, we will obtain some information about how “robust” the standard logistic model is towards perturbations.22 For the interval censored data I use survival analysis techniques. The data provides information about whether or not WTP is in a given interval. This is analogous to the information structure available in certain medical studies, in which it is known whether or not an event occurred in specific intervals of time. One could use the mid-points of each interval, but this technique cannot, inter alia, handle open-ended intervals in a satisfactory manner. It is convenient to assume a distribution of the unknown (interval censored) variable. I assume a Weibull-type distribution. There are many alternatives, including nonparametric approaches and extensions such as mixing several distributions. For a recent example of the latter approach in a similar application, see Belyaev & Kriström (2009).

21

See McCullagh & Nelder (1989) for details about these models.

22

There are certainly many other alternative lines of attack, including a family of distributions by estimating an additional parameter (the approach is similar to the Box-Cox transformation used in ordinary least squares problems, see e.g. Arando-Ordaz (1982)).

16


For the models used below, I begin the specification search with a matrix comprising three groups of variables. The list of variables is designed to help shedding some light on the questions set out at the beginning of the report (see Box 2). Box 2: Explanatory variables used in the statistical analysis

Socioeconomics

Household

Household composition Household income Top earner in the household (yes/no)

Residence variables

Residence owner Duration of living in current residence Type of residence Area of residence (town.,village etc) Age of residence

Attitudes

Other

5.

Marital status Gender Age Level of education Employment status

Attitudinal characteristics (various) Environmental concerns Environmental attitudes (secretariats computation, based on Q28)

Voluntary organization work Member of environmental organization Taking energy costs into account when buying/renting current residence

Results

Let us begin by considering individual activity on the energy market of interest, using Q67 (“Does your household take special measures to buy renewable energy from your electricity provider?”). We then turn to the analysis of willingness to pay by using survival analysis. This analysis is, in turn, divided into two parts. I begin by looking at the conditional willingness-to-pay, i.e. the subset of respondents that reported a positive valuation. I then look at the binary decision of being in the market or not. 5.1

On household decisions to undertake special measures to buy renewable energy

We begin the analysis of renewable energy demand drivers by considering individual‟s activity in the market. To repeat, Q67 asks the respondent “Does your household take special measures to buy renewable energy from your electricity provider?” We focus on the 7760 (1480+6280) “yes” and “no‟s” and drop the 1505 “don‟t know” answers. I thus include the 2538 respondents who said that “service not available, but our household would be interested”. A strong case can be made for deleting those observations. After all,

17

the response is not all that interesting if the choice itself is infeasible. However, it is possible that such a response does represent a deliberate choice, if, say, the service is available in area A but not in area B. The household could move to area A, but chooses area B in selecting where to live. I readily concede that this is farfetched in countries where the option is scarcely available, but I will retain these responses in the analysis. There are thus 986 missing values, corresponding to those respondents that did not mark “electricity” as an energy source on Q64. As noted above, the Swedes and the Koreans may be underrepresented, given their responses to Q64. I use a logistic model to analyze the probabilities of “yes” and “no”. The initial, or full, model includes the variables in the Box 2 above, i.e. the stepping stone for further analysis of the data. For model selection, I primarily use an approach based on Lawless and Singhal (1978) using the Akaike Information criterion23. This criterion is based on the likelihood function and imposes a penalty (on the likelihood) for adding variables. The Akaike criterion helps finding the best fitting model with the minimum number of parameters. Importantly, the criterion as used here is based on factor inclusion, not on single dummy variable inclusion. For example, the age variable is a factor with several age classes. The criterion then works out if age, viewed as a factor, should be included in the final model, and not whether or not a particular age class should be included. As noted, I compare this inclusion approach with the standard approach of including variables that are significant at the 5% level. Thus, while my conclusions are based on the objective of finding a parsimonious model, I include information that allows comparison with the more standard line of attack. I estimate the three model types described above: (i) assumes full homogeneity (ii) allows country fixed effects and (iii) allows country fixed effects and country specific marginal effects. The first model is rejected (using a likelihood test) compared to model type (ii), and I will not discuss it further. The test suggests that we must consider country heterogeneity. As a first step, I estimate a fixed effects model (type (ii)) and then turn to the type (iii) models. I apply the Akaike type criterion to arrive at the final model. Table 7 (below) has the final logistic model and comparative analysis with the two distributions mentioned above. I have estimated the model so that the parameters are to be interpreted in terms of the probability of a “no” to Q67. The right-hand variables selected in the final specification for correspond to: a dummy variable for the different countries, an attitudinal question on the importance of economic concerns (Q22.2), activity in environmental organizations (Q25.2), the environmental concern index and the energyrelated question Q70 (“Did you take energy costs into account when purchasing or renting your current primary residence?”). Observe that neither income, details about the residence etc. and other variables in the questionnaire were selected in the preferred model. Gender (Q3), household age (45-55 years of age), number of small children in the household (Q6) and membership in environmental organizations (Q27) and an environmental concern index is added to the model, if the model selection procedure is based on significance of factors, rather than the AIC. Adding these variables to the model does not add much to the explanatory power, as shown in Table A-7 in the appendix. Thus, for this discussion I will stick to the more parsimonious model.

23

The criterion is 2*log-likelihood - k*npar, where k is a constant (k=2 in the standard case) and npar is the number of parameters.

18


Table 7 – “Estimation results for Q67: Does your household take special measures to buy renewable energy from your electricity provider?” glm.q67.logit (Intercept) Netherlands/Canada France/Canada Mexico/Canada Italy/Canada Czech R./Canada Sweden/Canada Norway/Canada Australia/Canada Korea/Canada Ec. Concern= Not important

3.03*** (0.27) -2.48*** (0.24) 3.65*** (0.93) 0.26 (0.29) 0.63** (0.32) 3.39*** (1.04) -1.03*** (0.26) 0.52 (0.38) -1.23*** (0.24) -2.07*** (0.24)

0.95*** (0.08) -1.15*** (0.07) 0.32*** (0.07) 0.12* (0.07) 0.16** (0.06) 0.38*** (0.08) -0.34*** (0.07) 0.04 (0.07) -0.43*** (0.07) -0.92*** (0.08)

-0.39***

-0.38***

-0.21***

(0.10)

(0.04)

-0.29***

-0.10***

(0.09)

(0.03)

-0.02***

-0.01***

(0.00)

(0.00)

-1.04***

-0.43***

(0.10)

(0.13)

(0.05)

0.46***

0.41***

0.24***

(0.07)

(0.08)

(0.03)

0.26*

0.22

0.15**

-0.22*** (0.07)

Environmental concern (index)

-0.01*** (0.00)

Env Org Support (Time) = Yes Energy Cost Concern = No Energy Cost Concern = Not sure

glm.q67.cloglog

2.49*** (0.17) -2.10*** (0.14) 0.92*** (0.20) 0.26* (0.16) 0.36** (0.15) 1.01*** (0.23) -0.71*** (0.16) 0.13 (0.17) -0.88*** (0.14) -1.67*** (0.15)

(0.09) Ec. Concern= Not so important

glm.q67.cauchy

-0.85***

McFadden R-sq. Likelihood-ratio p Log-likelihood

(0.15) 0.15 1160.97 0.00 -3200.71

(0.18) 0.15 1103.39 0.00 -3229.50

(0.07) 0.15 1165.57 0.00 -3198.41

AIC

6433.42

6491.01

6428.83

19

To evaluate the chosen model specification, consider its fit and the significance of the parameters. The Mc-Fadden R2 is similar for the three models. The p-value (which is 0) and the likelihood ratio chi-square statistic suggest that the model “significantly” explains variation in the data, relative to a model without explanatory variables. Because the parameters are related to a probability, their particular values are not of primary interest. More important are their sign and we find that several parameters are significant at the 1% level (denoted by ***). 24Also note that other two distributions give rather similar results. Thus, the logistic model seems “robust” to a particular perturbation of the underlying assumptions. The dummy-variables for each country (Canada is absorbed into the constant) can be interpreted in this way: if the country-effect is positive, it is more likely that a person from such a country would say “no” to Q67 all else equal, compared to a Canadian. The coefficient for other countries in the table is to be interpreted in terms of the difference in probabilities for “no” relative to the Canadian baseline. The country-specific dummies shows if there are significant differences between countries, given the background variables. With this interpretation, we thus say that there are positive country-effects differences for Czech Republic, France, Italy and Norway and Mexico, but observe that the differences for Norway and Mexico are not significant.25 Conversely, the model strongly suggests that a respondent from the Netherlands, Sweden, Australia and Korea is more likely to respond “yes” to Q67, all else given, relative to a Canadian. In short, there are heterogeneities -- differences between countries -- that are not picked up by the background variables in the model. “Economic Concern” in table 7 is derived from a survey question (Q22) that asks the respondent to rank the importance of various issues. The responses are originally coded 1 for “most important” to 6 for “least important”. I merged these into three categories “important” (1-2), “not so important” (3-4) and “not important” (5-6). The results suggest that those respondents who are less concerned about economic conditions are less likely to say “no” to Q67. “Env Concern” is based on the answer to: “In the past 24 months have you given any of your personal time to support or participate in activities of .[an Environmental Organisation] ”. The coefficient is significantly negative, which means that being active in such an organization increases the probability of being active in the renewable electricity market (in the sense of Q67). Membership/activity has been found in earlier research to be important for understanding the residential market for renewable energy. The coefficient on the environmental concern index is negative and statistically significant. This index is based on Q23, and the higher the index, the “more environmental concern”. Overall, the mean of this index is about 46, the minimum of about 36 is recorded among respondents from the Netherlands, while the Mexican respondents tops at about 57. Whether or not it is meaningful to compare this index across countries is a contentious issue, but I will proceed on the assumption that “very concerned” (say) means the same to a Dutch and a Mexican, after translation. At any rate, the estimated coefficient suggests that the “more concerned” an individual is, the more likely that he is actively searching for “green” energy alternatives. The positive coefficient on “no” to Q70 (“Did you take energy costs into account when purchasing or renting your current primary residence?”), tells us that such respondents are less likely to be active in the sense of the answer to Q67 (this is also true for “not sure” group, although this coefficient is not significant). Consequently, more active respondents belong to a category of respondents that considered their energy costs when purchasing/renting the residence. 24

I used bootstrapping to buttress the t-values for the logistic model and found that t-values were close to the original estimates.

25

The latter result could be guessed from the raw data, as the raw percentages of “no” (deleting the “don‟t knows”) are 0.87, 0.86 and 0.9 for Canada, Mexico and Norway, respectively.

20


A capsule summary of the above results could be the following. A respondent who is active in the market for renewable energy is; more likely to be active in an environmental organization; more likely to be “environmentally concerned” and is more often considerate of energy costs in renting/buying residence. These results are based on my objective of finding a parsimonius model. If our focus is on significance of parameters then gender (women have a higher estimated probability of saying “no”) and number of children in the households are potential candidates. In addition, there is some indication that the age-group 45-55 has a significantly higher probability of saying “no”. Most importantly, there are significant differences between several countries unrelated to the sets of explanatory variables. To cap this section, it might be useful to return to Box 2 with the information from the econometric analysis added. Variable Socioeconomics

Marital status Gender Age Level of education Employment status

Significant (5%)

Gender Age:45-55

Household

Household composition Household income Top earner in the household (yes/no)

Residence variables


Attitudes

Attitudinal characteristics (various) Environmental Concern Environmental attitudes

Economic a Concern Pers. Safe b Concern Environmental Concern Environmental attitudes

Voluntary organization work Member of environmental organization (Q27) Taking energy costs into account when buying/renting current residence

Charitable Org. time

Other

Country dummies

Included in preferred model

# Children

Economic Concern Environmental Concern

Env.Org Support (Time) Env.Org Support (Time) Energy Cost Energy Cost ca,nl,it,cz,se,au

All

Note: Results are for the logistic model. Significant dummy variables: a) “Not important” and “Not so important” b) “Not important”.

21

The above model (type (ii)) is constrained by the fact that the slope parameters in the model are assumed equal for all countries. Because the underlying model is non-linear, the consequence of this simplification is somewhat subtle, but we can loosely interpret it in the same way we would in a linear model. To release this homogeneity constraint, I first re-estimated the “full” model for all separate countries, thus allowing the impact of all the background variables to vary between countries. In terms of the model selection criteria, the results were not very encouraging, with most factors being insignificant. I therefore re-estimated the preferred specification above for each country. Briefly, the estimation results display significant variation (although each individual country model is “significant” in the sense of the p-value being smaller than 0.05) and the model fit is poor (see table 8). Table 8 - Estimation results for individual countries, Logit model for Q67: Does your household take special measures to buy renewable energy from your electricity provider?

Canada Netherlands France Mexico

(Intercept) Ec. Concern=Not important Ec. Concern= Not so important Environmental concern (index) Env Org Support (Time) = Yes Energy Cost Concern= No Energy Cost Concern = not sure McFadden R-sq. Likelihood-ratio p Log-likelihood AIC

3.08*** (0.51)

1.22*** (0.24)

2.98*** 0.78 (0.66) (0.73)

-0.66**

-0.26

0.07

-0.52*

(0.32)

(0.18)

(0.56)

(0.28)

-0.76***

-0.30*

-0.35

-0.01

(0.28)

(0.16)

(0.34)

(0.23)

-0.02**

-0.02***

-0.01

0.02

(0.01)

(0.00)

(0.01)

(0.01)

Italy

Czech R.

1.87*** 2.39*** (0.50) (0.68)

Sweden Norway Australia Korea

2.64*** 1.96*** (0.46) (0.38)

0.06 (0.41)

-0.54* -1.46*** -0.35

-0.66^** -0.02

-0.16

(0.28) (0.52)

(0.30)

(0.29)

(0.22)

(0.31)

-0.36

-0.10

0.13

-0.45**

(0.24)

(0.29)

(0.19)

(0.21)

-0.02**

-0.01

-0.03***

0.00

(0.01)

(0.01)

(0.01)

(0.01)

-1.10*** -0.73*

-0.55**

0.32

(0.25)

(0.40)

(0.28)

(0.33)

0.22

-0.95*

(0.24) (0.49) 0.00

0.02

(0.01) (0.01)

1.98*** (0.37)

-0.26

-1.82***

-0.58

-1.51*** -0.63 0.98***

(0.40)

(0.49)

(0.57)

(0.21)

0.72***

-0.35**

0.58*

0.64*** 1.04*** 0.73

0.45*

0.46*

0.65***

0.45**

(0.24)

(0.17)

(0.33)

(0.21)

(0.23)

(0.27)

(0.19)

(0.19)

-0.40

-0.14

0.14

0.31

0.63

0.64

-0.60

0.28

0.41

(0.51) 0.06 30.37 0.00 -247.99 509.98

(0.41) 0.04 50.89 0.00 -604.44 1222.88

(0.78) 0.02 6.32 0.39

(0.45) 0.06 37.45 0.00

(0.42) 0.10 77.29 0.00

(0.56) 0.07 40.86 0.00 -277.91 569.81

(0.57) 0.04 19.39 0.00 -252.35 518.70

(0.51) 0.04 42.44 0.00 -451.76 917.51

(0.32) 0.01 11.07 0.09 -370.99 755.98

(0.23) (0.67)

(0.21) (0.49) 1.49**

(0.76) 0.08 17.84 0.01 -162.60 -314.01 -353.05 -103.64 339.21 642.02 720.11 221.29

I interpret these results to mean that the sample is rather heterogeneous. When data are pooled, as above, the pooled model masks significant variation between countries. To demystify this statement, let us compare Table 7 and Table 8 and the parameter for “Energy Cost Concern =No”. The pooled model suggests a positive and significant coefficient, suggesting that households (on the average) are more likely to be inactive (“taking special measures”), if they did not consider the energy cost when renting/buying their residence. Table 8 shows, for example, that this coefficient is significantly positive for Canada and significantly negative for the Netherlands. Looking at the raw data, one finds that Netherlands is different

22


from the other countries in one interesting aspect. Thus, except for the Netherlands, there is a preponderance of households that are inactive and said “no” to the question about energy cost concern. In the Netherlands data, 344 households selected the combination (“Active”, “No Concern”), while 352 households choose (“Not Active”, “No Concern”). The corresponding numbers for e.g. Italy is 48 and 740. This is an example of the heterogeneities that are masked by pooling the data. In principle, there could also be relevant within-country differences in certain cases. I consider such differences of less interest when looking at market activity and proceed by analysing the (hypothetical) price premium for “green” electricity. 5.2

Willingness to pay for renewable energy

Q69 asks the respondent about the price premium for green energy, assuming his consumption of electricity fixed. This is a construction that differentiates this survey from others in the literature.26 If consumption is not held constant, the consumer might change his consumption level as the price of electricity increases. The gist of this construction is to avoid confounding an environmental benefit and a consumer surplus loss, the latter being due to a changed level of consumption. The technical appendix develops a simple conceptual model to which the interested reader is referred, in which this idea is conceptualized. It is important to note that WTP is expressed as a percentage increase, relative to current spending on electricity. Below, I will simply call this “WTP”, even though it is a percentage. As noted above, the econometric analysis proceeds in two steps; we begin with those who are in the market (i.e. WTP>0) to explore shifters of the “demand curve”. A particular analysis is then undertaken on the decision to “enter the market”. I could analyze these two decisions in one single model, at the cost of computational complexity. Before proceeding to estimate the demand curve, we need to discuss the responses to Q69, given what the respondent has stated earlier in the questionnaire. In particular, the responses to Q68 regarding the reasons as to why “…you do not buy renewable energy” are particularly relevant. The possible answers to Q68 were (to repeat): 1. “Service not available and our household is not interested” 2. “Service not available, but our household would be interested” 3. ”Service available, but our household is not interested”; 4. ”Energy from electricity provider is already from renewable energy”; 5. “I don‟t know anything about these kinds of services”. This last option may not be an uncommon response, see Zarniku (2003, p. 1664). If the households who are not interested answered that they have zero WTP for renewable electricity, this is arguably a consistent answer. However, there are 295 households who state “service not available and our household is not interested” and simultaneously claim WTP>0. Out of the 2106 ignorant (i.e. “don‟t know”) households, 1225 has WTP>0. This poses a challenge, because one can argue that such households cannot meaningfully state a value. Nevertheless, I will accept the answers to Q69 face value, inconsistent or not. After all, the respondent did furnish the information and disregarding it introduces other issues (such as masterminding the respondent). For example, the scenario posited a world (or rather a country) in which a 26

The theory underlying this reasoning is given in Johansson (1987).

23

complete switch to renewable energy had been undertaken. When the respondent states he is not interested to buy “green” electricity in today‟s circumstances and simultaneously furnish a positive WTP on Q69, this could be a quite consistent sentiment. At any rate, I will assume that the individual‟s answer to Q69 is a fair assessment of his preferences, no matter the underlying motives. Before turning to the results, it might be useful to recap briefly what some other studies have found to be determinants of WTP for “green electricity”. A contribution from Rowlands, Scott & Parker (2003) approaches WTP from a marketing perspective; they try to find variables that can act as segmentation indicators. In a bivariate analysis, they uncover a number of results to “profile” the individual respondents‟ demand for green electricity. They report, for a sample of South-western Ontario residents, that “Attitudinal characteristics – specifically ecological concern, liberalism and altruism best identify the potential purchasers of green electricity” (Rowlands, Scott & Parker (2003, p. 36)). Roe et al (2001, p.919) found that those who are not members of environmental organizations have lower WTP. These studies are representative of studies that find attitudinal and other “non-economic” factors important. As noted, studies that find socio-economic variables significant include Ek & Söderholm (2008) and Zarniku (2003). With this pre-amble, we now turn to the estimation results. I use maximum likelihood methods to fit a Weibull distribution (see Table 6) to the data. In contrast to many other studies in this genre, I carry out a separate analysis of WTP conditional of it being positive and the decision to enter the hypothetical market. First, I consider the level of WTP, conditional on it being positive. The estimated parameters are related to the probability that WTP>x and can loosely be interpreted as shifters of the “demand curve” (c.f. the empirical distribution function displayed in section 3). As shown in the technical appendix, the stated percentage is (to a linear approximation) inversely proportional to the household‟s current electricity spending. There is no information in the questionnaire about how much the household spends on electricity, but note that with this information it would possible to approximately measure the level of WTP. Furthermore, the model in the appendix suggests an econometric specification. Importantly, if the utility function is linear over the relevant interval, income should not enter the econometric model. The linear model is silent on additional parameters (beyond the budget-share), but additional parameters can be added as a way of exploring WTP in different groups. Therefore, I proceed with the same set of variables as used for Q67, but exclude income. This might seem like a particularly curious line of attack, given that a key question in this report is the relationship between WTP and income. Consequently, I will address this relationship separately, and begin by analyzing a model that is consistent with the theoretical model in the Appendix. I arrive at the preferred econometric specification using the same procedures as above. Econometric results from the preferred specification are reported in table 9. To remove some of the mysteries of the model, table 9 also has the formula specifying exactly what is estimated. Recall that I begin with respondents that are “in-themarket”, i.e. WTP >0.

24


Table 9 - Parameter estimates Weibull model for WTP > 0 Parameter Intercept Netherlands France Mexico Italy Czech Republic Sweden Norway Australia Korea Female Env. Concern= Not Import Env. Concern= LessImport Env. Org=Not member Log(scale) Scale = 1.03 2 R = 0.06 L0 = -4618 Lmodel = -4498 *

Value 2.2987 -0.4094 -0.1594 0.3327 0.0118 -0.0313 0.1703 0.2557 -0.0449 -0.1024 -0.1536 -0.1447 -0.1707 -0.3374 0.0301

Std. Error 0.0677 0.0918 0.0757 0.0674 0.0683 0.0807 0.0815 0.0756 0.0733 0.0722 0.0327 0.0446 0.0374 0.0431 0.0171

z 33.974 -4.462 -2.106 4.939 0.173 -0.387 2.090 3.384 -0.612 -1.418 -4.692 -3.245 -4.562 -7.833 1.761

p 5.47e-253 8.11e-06 3.52e-02 7.84e-07 8.63e-01 6.98e-01 3.66e-02 7.14e-04 5.40e-01 1.56e-01 2.70e-06 1.18e-03 5.06e-06 4.77e-15 7.83e-02

All countries in single model. The dummy coefficients are to be interpreted as differences between each level and the baseline.

A Weibull model for WTP > 0

  log t   X  PrT  t  exp exp  1.03   

where

Xˆ  2.3 –0.409{Netherlands} –0.159{France} +0.333 {Mexico} + 0.012 {Italy} –0.031{Czech R} +0.170{Sweden} +0.256{Norway} + 0.045 {Australia} –0.102{Korea} –0.15{Female} –0.14{EnvConcern = Not important} –0.17{EnvConcern = Not so important} –0.34 {no}

The model as a whole is significant relative to a model without explanatory variables. Next, consider the “shifters” of the demand curve. First, note that several of the country effects are significant. Again, Canada is absorbed in the constant and we interpret the coefficients as differences. The model predicts a lower WTP for women, ceteris paribus, which, to some extent, contradicts earlier findings in the literature on the valuation of environmental quality.27 Next, “Env. Concern” (Q22.3) is an attitudinal question regarding how important environmental concerns are to the individual; those who find environmental issues relatively unimportant display a lower WTP, cet. par. This is intuitively plausible. The coefficient for “”Env.Org =Not Member”, indicates a lower WTP (cet. par) for a respondent that was not a member of

27

For a review, see Farreras, V., Riera, P. and J. Mogas (2005).

25

an environmental organization. Hence, there is a significantly positive effect on the demand curve from those who are members. For completeness, I discuss the results from the standard inclusion approach; if the model selection procedure instead was based on including significant coefficients at the 5% level, then we get a slightly different picture. Significant fixed effects differences (in the above interpretation) are found for Netherlands (negative), Mexico and Norway (both positive) “living alone” and “sharing a house/flat with non-family members” both have significantly positive impact. There is also an effect of significant employment status; the coefficient on “retired” is negative. If a person is not the one who earns most in the household, then this affects WTP positively. Environmental preferences, represented by the environmental concern index and membership in environmental organizations both have positive impact on WTP. Table 10 has details about these results. Table 10 – Confidence intervals for significant (5%) parameter estimates Weibull model for WTP > 0 Parameter

2.5 %

97.5 %

1.5866

2.2857

-0.5595

-0.1793

Mexico

0.1556

0.4564

Norway

0.0528

0.3628

Primary residency status = Living alone

0.0301

0.2544

Primary residence status = sharing a house/flat with non-family members

0.0031

0.3682

Female

-0.2747

-0.1275

Current employment status= Retired

-0.2839

-0.0227

0.0260

0.1957

Intercept Netherlands

Earning most in HH=no Economic concern = Not important

0.0455

0.2325

-0.1873

-0.0219

Personal safety=Not important

0.0127

0.1853


0.0011

0.0067

Charitable Org Time ="yes"

0.0243

0.1769

-0.3411

-0.1555

Environmental concern =Not so important

Env Org Member="No"

* All countries in single model. Dummy variables are to be interpreted in terms of differences to the baseline.

26

*

For convenience, I summarize the above discussion, by returning to Box 2 and inserting the econometric results for Q69.

Socioeconomics

Household

Residence variables

Variable Marital status Gender Age Level of education Employment status Household composition Household income Top earner in the household (yes/no)

Significant (5%)


Gender (Q3)

Gender(Q9)

Retired (Q9)

Top earner (Q12)


1. Living alone 2.sharing (Q1)

Attitudinal characteristics

(Q22.2) Economic Concern b (Q22.3) Environmental concern c (Q22.6) Personal safety

Environmental Concern

a

Environmental Concern (Q22.3)

Attitudes Environmental Concern (index)

Other

Environmental attitudes Voluntary organization work Member of environmental organization (Q27) Taking energy costs into account when buying/renting current residence

Country dummies a) “Not important” b) “Not so important” c) “Not important”

Charitable Org. time (Q25.4) Member of environmental organization (Q27)

Nl,fr,mx,no

Member of environmental organization (Q27)

All


Overall, the picture that emerges from the econometric analysis of Q69, using model type (ii), is the following. First of all, there are some significant country-effects. They can be interpreted as summarizing a set of unmeasured variables and portray differences between countries. Second, WTP is clearly related to sentiments about the environment, either via attitudinal variables or variables that signal activities in environmental organizations. I also find a gender effect, but with opposite sign to the one usually found. WTP and income The theoretical model shows that income should not be included in the econometric model, at least not to a first order approximation. Yet, if we do include income in the model, it is not significant at the 5% level. Re-estimating the full Weibull model for each country shows that the income coefficient is significant (at the 10% level) in the case of the Netherlands, but not for the other countries. I conclude that there is no strong statistical relationship between WTP, conditional on it being positive, and income. I will buttress this finding by exploring the link between income and the decision to “enter the market” below. Furthermore, it will be useful to further investigate the “between” and “within” country differences. But before carrying out this analysis, I turn to estimating mean WTP. Mean WTP Let me begin with a rudimentary calculation of mean WTP, using class midpoints. Because the last class is open-ended, we need to assume an upper bound. I put this at 30% to get a conservative estimate and for an upper bound I put it, arbitrarily, at 100%. Thus, I simply multiply the class midpoints by the proportions in each class, using 30 and 100 as upper bounds and then add up. Straightforward calculations show that the conditional mean is in the interval 7.4-9.1%. Factoring in the 46.2% zero WTPs, we obtain the bounds 4-4.9 %. I then estimate conditional median and mean WTP using a Weibull model without covariates. The results for the respective countries are in table 11. Table 11 – Conditional Mean and Median WTP estimates by country

*

Intercept

Scale

Mean

Median

Canada

1.9

1.05

6.8

4.7

Netherlands

1.5

1.14

4.9

3.4

France

1.5

1.31

6.1

4.2

Mexico

2.2

0.97

8.9

6.2

Italy

1.8

1.05

6.7

4.6

Czech Republic

1.8

1.03

6.2

4.3

Sweden

2.1

0.83

6.9

4.8

Norway

2.1

0.96

8.1

5.6

Australia

1.8

1.09

6.6

4.6

Korea

1.6

1.21

6.2

4.3

Given Pr(T  t) = exp(exp([(log(t)x0)/()])), where x0 is the intercept and  the scale, E(WTP)=x0 and Median(WTP)=log(2)x0.

The results show some variations across the countries, with the Dutch respondents showing the lowest mean WTP and the Mexican respondents the highest mean WTP. Thus, the country with the highest penetration of renewable energy has respondents displaying the lowest WTP and the country with the lowest average income has the highest. These facts make the results seem rather counterintuitive, but consider the following three arguments as to why such a disparity may exist.


First, one could argue that the Netherlands market is saturated, so that WTP at the margin is not very high. There is no data to corrobate this assertion, except the data cited above on the market penetration in the Netherlands. Of course, there could be differences between countries regarding how respondents answer hypothetical questions, so that the Dutch have a different approach to hypothetical questions. At any rate, I offer the saturation hypothesis as one possible story. Second, the theoretical model in the appendix suggests that the stated WTP is inversely proportional to the amount spent on electricity. Thus, information about the actual budget shares would be useful. Lacking such information, I consider the structure of electricity pricing. In Mexico, the pricing structure is rather complex. Residential consumers obtain electricity from two state owned companies: Comisión Federal de Electricidad (Federal Power Commission) and Luz y Fuerza del Centro (LFC). LFC provides electricity in Mexico City and nearby areas while CFE is the main utility for the rest of the country. The residential rate is divided into 7 different rates according to average temperature and location. There is a special rate for high consumption called the DAC rate (De Alto Consumo).28 It is therefore possible that the actual budget shares in the Mexican data are smaller than in the other countries. Again, there is no hard information in the data whether or not this is true. Third, a rudimentary analysis of the Mexican data suggests to me that there are considerable representativity issues; after all, the internet penetration rate is only slightly above 20 % (according to www.internetworldstats.com). On this ground, it is possible that the Mexican respondents actually have a relatively lower expenditure share of electricity (and/or different preferences). Let us proceed by calculating the sample average and contrast this with what is known in the scientific literature and elsewhere. Mean conditional WTP over the countries using the Weibull is 7.1%. The corresponding median is 4.9 %. If we factor in the fact that 46.2% of the sample has zero WTP (this deletes the “don‟t know” answers), mean WTP is about 3.8%. Observe that this number is close to the simple calculation above. The bottom line is that WTP for renewable energy in the sense of Q69 is a couple of percentage points above the current bill. This is a result that concords with some of the folklore in the power business and, as we will see, it concords with recent findings in the scientific literature. The “average” WTP reported for France and Italy in the Harris (2008) survey (see above), is 4.7% and 5.1% respectively. Thus, the numbers reported in the Harris survey are close to the numbers found here.29 A few additional examples from the literature might be useful: Farhar and Coburn (1999) [Colorado home owners, 1998] found a median WTP between 2% - 5% increase of the monthly electricity bill. Champ and Bishop (2001) [Wisconsin residents] obtained WTP for wind-generated electricity at USD 101/59 for a hypothetical/real scenario. Wiser (2003) [US residents] estimated a WTP for the switch to renewable energy at approximately US$3 a month. In short, the results reported here are not markedly different compared to similar studies. Again, because the countries in the survey comprise a rather diverse set across economic, environmental and energy system parameters, I find the small variation in mean WTP fascinating. Let us now return to a further analysis of heterogeneity and proceed in two steps. First I look at between country differences and then one case of within-country differences, namely Sweden.

28

I am grateful to Dr. Diego Arjona, Deputy Undersecretary for Research, Technology Development and Environment Ministry of Energy Mexico, for supplying this information.

29

The “averages” reported in the Harris survey seem to be just the weighted proportions. This is not average WTP in the sense we use “average” here, because we are using the Weibull.

29

Country analysis The analysis of Q67 suggested country heterogeneities and it is prudent to analyze the econometric model for each country also here for WTP. Fitting the full model for each country results in a bewildering number of coefficients (570), so it is necessary to try to distill key features of the results. First, consider model summaries in Table 12, where I fit the full Weibull model for each country. Table 12 – Model summaries for Weibull model applied to each country*

surv.c

P

R

2

Obs

Model L.R.

d.f.

Canada

410

111

57

2.7e-0.5

0.27

Netherlands

231

116

56

4.2 e-06

0.50

France

420

83

57

1.5 e-02

0.21

Mexico

668

86

56

5.7 e-03

0.14

Italy

637

87

57

6.5 e-03

0.15

Czech R.

307

78

57

3.4 e-02

0.26

Sweden

305

90

56

2.5 e-03

0.29

Norway

408

78

57

3.6 e-02

0.20

Australia

454

104

57

1.5 e-04

0.24

Korea

532

71

57

9.4 e-02

0.15

* The estimated model for each country is based on Pr(T  t) = exp[exp( [(log(t)X)/()])]  is the scale parameter. X include all variables listed in Box 2. R2 = (1 exp(Model L.R./N)/( 1  exp(2 L0)/N), where L0 is the likelihood with the constant only.

“Model L.R.” is 2*difference between the likelihood of the full model and a model with no explanatory variables. “d.f.” is the number of fitted coefficients, which together with the “Model L.R” values produces the column “P”, when inserted in to a Chi-square distribution. The final column “R2” is an indicator of model fit that is related to the standard measure of fit in linear regression. Overall, the models are ”significant” for all countries, as indicated by the “P” statistic. Furthermore, the best fit is found for the Netherlands data. When we turn to the significance of the parameters, the overall fit is not very impressive. As expected, the model for the Netherlands has the largest number of significant parameters, given table 12. Membership in environmental organizations and attitudes towards the environment are the most consistently significant parameters across models by country. Other than that, there seems to be no particular pattern. Next, I considered the preferred model that was used above for all countries. If this model is run for each country, I find that membership in environmental organizations is significant, with expected sign, in all country-models, except Norway. Other than that, there is no particular pattern. I conclude that environmental attitudes/activities in environmental organizations is the most important determinant of WTP, To finish the discussion about determinants of the level of WTP, I take a brief look at within-country differences. Within country analysis: Sweden In the case of Sweden, the only within-country analysis undertaken in this report, there are some potentially interesting regional differences (such as the stronger tendency in the North to use biofuels and

30


less district heating). A question is if these and other differences map into differences regarding WTP. I thus divided Sweden into three regions (corresponding very roughly to North, Middle and South) and explored the data. A problem with this (common division) of Swedish regions is that only 10% of the Swedish population live in the “North”, reducing the number of observations to N=92 in this region. Nevertheless, consider Table 13. Table 13 – WTP for the three Swedish regions Gotaland

Svealand

Norrland

I would not pay anything additional

0.47

0.48

0.42

less than 5%

0.12

0.13

0.12

5%-15%

0.17

0.14

0.18

16%-30%

0.03

0.02

0.02

more than 30%

0.01

0.00

0.00

don’t know

0.21

0.22

0.25

The proportions across the three regions are remarkably similar. I applied the Weibull model along the same lines as above. As expected, the model for the North does not fit the data all that well. Interestingly, while the coefficient for households “living alone” in the sparsely populated North is significantly negative, this coefficient is significantly positive for the Svealand region (which includes Stockholm). The population densities in these two areas are 4.4 and 39.6 persons/km2, respectively (according to Wikipedia). Is this an example of an urban/rural asymmetry regarding preferences for renewable energy? For all the three regions, I find that at least one of the variables signalling environmental attitudes is significant, with expected sign. Furthermore, variables signalling activities in environmental or voluntary organizations are not significant for the North, but for the other two regions. Consistent with the results above, women are expected to have a lower WTP (cet.par) (but this coefficient is not significant for the North). Thus, there are some interesting within-country heterogeneities also within a country. To summarize the econometric analysis of conditional WTP: the analysis this far suggests that beyond country idiosyncrasies, gender matters, as does variables representing attitudes towards the environment. We also find that the WTP for the good described in Q69 is rather low, a number of percentage points of the current bill (for those in the market). This is consistent with (soft data on) the experience of electricity retailers in Sweden and several other recent findings. There is, however, no strong correlation between the level of WTP and income. While this result is not inconsistent with a rudimentary microeconomic model, some relationship is still expected. I would also like to highlight an interesting within-country finding, suggesting diametrically opposing preferences for green electricity in two parts of Sweden. Singlehouseholds in the “densely” populated Svealand are expected to have a higher WTP than the average in this region, while the converse is true in the sparsely populated Norrland. Because there is quite a large number of respondent that are “not-in-the-market”, one might suspect that this group is different. Therefore, I now consider the binary decision of entering the market for “green electricity”. 5.3

The decision to enter the market

While it is possible to statistically analyze the WTP statement simultaneously with the decision to “enter the market”, i.e. WTP>0, I will simplify by analyzing those decisions separately. Note that this

31

separation is seldom analyzed in the literature on green price premiums. I use the same set of explanatory factors as in the previous models, using a logit model. The model reduction procedure is based on the Akaike criterion. In Table 14, I present the final logit specification with and without income, for comparative purpose. Table 14 - Logit model for the decision to enter the market w/wo the income variable model.lgt.final (Intercept) Netherlands/Canada France/Canada Mexico/Canada Italy/Canada Czech R./Canada Sweden/Canada Norway/Canada Australia/Canada Korea/Canada age= 25-34/18-24 age =35-44/18-24 age =45-55/18-24 age =55+/18-24 high school graduate some post-secondary education (tafe qualification, diploma) bachelor's degree (ba) post graduate degree (master or phd) prefer not to answer Ec. Conc. = Not important Ec. Concern= Not so important Env. Concern= Not important Env. Concern= Not so important Personal safety = Not important Personal safety = Not so important

32

model.lgt.final.w.income

-0.22 (0.19) -1.09*** (0.11) -0.13 (0.11) 0.78*** (0.12) 0.15 (0.10) 0.41*** (0.12) -0.57*** (0.11) -0.13 (0.11) 0.10 (0.10) 0.42*** (0.11) -0.42*** (0.09) -0.62*** (0.08) -0.73*** (0.09) -0.72*** (0.08) 0.07 (0.09) 0.11

-0.42** (0.20) -1.01*** (0.12) -0.09 (0.11) 0.99*** (0.13) 0.21** (0.10) 0.49*** (0.13) -0.50*** (0.11) -0.23** (0.11) 0.13 (0.11) 0.49*** (0.12) -0.42*** (0.09) -0.67*** (0.09) -0.77*** (0.09) -0.75*** (0.09) 0.03 (0.09) 0.08

(0.09) 0.36*** (0.09) 0.47*** (0.10) -0.40 (0.24) 0.27*** (0.07) 0.12** (0.06) -0.42*** (0.07) -0.22*** (0.06) 0.32*** (0.06) 0.15** (0.06)

(0.09) 0.28*** (0.09) 0.39*** (0.11) -0.57** (0.28) 0.27*** (0.07) 0.13** (0.06) -0.39*** (0.07) -0.20*** (0.06) 0.35*** (0.07) 0.15** (0.06)


model.lgt.final

model.lgt.final.w.income


0.01*** (0.00)

0.01*** (0.00)

Charitable Org Time

0.28*** (0.06) 0.04*** (0.00) -0.40*** (0.07)

0.29*** (0.06) 0.04*** (0.00) -0.37*** (0.07) 0.01*** (0.00) 0.11 -4948.38 9956.75

Environmental attitude (index) Env Org Member = no income McFadden R-sq. Log-likelihood AIC

0.11 -5268.95 10595.90

It is interesting to note that age is a significant predictor for the entrance decision, but not for the level of WTP (according to the models used here). Thus, increasing age seems to have a negative impact on the decision to enter. Q8 describes education status. The significant categories (higher education) suggest a positive effect of education on the probability of entering the market, ceteris paribus. The remaining variables (except personal safety (Q22.6) and income) in Table 14 are related to attitudes towards the environment, and activities/membership in environmental organizations. The signs of those coefficients are congenial to intuition. Q22.6 deals with personal safety concerns and is not straightforward to interpret. Interestingly, we find a positive significant of income on the entrance decision. This means that income (according to the models used) has a positive impact on the decision to enter the market, but not on the level of WTP. Let us apply Box 2 again to summarize this discussion. Socioeconomics

Household

Residence variables

Attitudes

Variable Marital status Gender Age Level of education Employment status Household composition Household income Top earner in the household (yes/no) Residence owner Duration of living in current residence Type of residence Area of residence (town.,village etc) Age of residence Attitudinal characteristics Environmental Concern

Other

Country dummies

Environmental attitudes Voluntary organization work Member of environmental organization (Q27) Taking energy costs into account when buying/renting current residence

Significant (5%)


Age (all groups) Bachelor Post-graduate Income

Age Education

(Q22.2) Economic Concerna (Q22.3) Environmental concernb (Q22.6) Personal safetyc

(Q22.2) Economic Concern (Q22.3) Environmental concern (Q22.6) Personal safetyc

Environmental Concern (index)

Environmental Concern (index) Environmental Attitude (index)



cz,it,nl,mx, ko,no,se

All

Income

a) “Not important” “Not so important” b) “Not Important” “Not so important” c) “Not important” “ Not so Important”.

33

In short, I find that the variables driving the decision to enter the market are not exactly the same as those driving WTP. The most interesting variable from an economic point of view is income. It affects the probability that WTP>0, but not the level of WTP. One interpretation of this result is heterogeneity; the two groups (positive/zero WTP) are dissimilar along a number of relevant dimensions. Country level analysis To explore these findings further, I re-ran the full model at the country-level. As before, the results suggest heterogeneity. The most stable factor is household age, being significant in all sub-models except two30. Furthermore, the variables depicting environmental attitudes are often significant. Income is significant (at the 5% level) in the submodels for Australia, France and the Netherlands. Consequently, the full model does hide significant heterogeneity between countries and we can conclude that the link between income and the entry decision depends on how we model country differences. Still, the link between income and the decision to enter the market is stronger (in a statistical sense), than the link between income and WTP. This might also be the case within a country and only a few brief remarks are in order. I find that the “within-sweden” model for the North does not fit the data well, in terms of significance of parameters. For the two other regions, age is, again, the most stable factor, suggesting that higher age is negatively related to the probability of entering the market. I also find that the variables measuring environmental attitudes positively affect the probability to enter, while income is not significant at this level of aggregation in any of the three regions. In the case of “Norrland” this could be due to the low number of observations. To sum up: I used a “Russian doll” approach to progressively fine-grain the analysis over regions (all countries, individual countries, within a country). I find that there are potentially different drivers behind the decision to enter the market for “green” electricity and how much to contribute, given entry. The analysis suggests heterogeneity, perhaps signalling that we are unable to pin down the data-generating process in any detail. Still, there is one consistent finding: environmental attitudes/activities in environmental organizations play a role in determining demand for “green electricity”. Finally, it may seem odd that we find significant relationships for a type 2 model, but when we use a more general type 3 model, the result is no longer as strong, A good example is income. When assuming a type 2 model (where then intercepts are allowed to vary by country, but not the slope coefficients, to continue the regression analogy), income significantly “explains” the entry decision. However, when we allow the income effect to vary by country, the effect is no longer as strong. There are a number of possible reasons for such results, but consider the most intuitive. If we only had two countries, and the correlation is very strong in one case and zero for the other, it can certainly happen that the amalgamation of these two countries into one single model results in a “significant coefficient”. The aggregation masks the heterogeneities. 6.

Policy implications

The analysis leads to three broad conclusions on the residential demand for renewable energy. First of all, this and several other studies send a strong message regarding household sentiments. While the analysis uncovers differences between countries, the demand for renewable energy is not very significant in terms of willingness to pay, a few percent above the current bill on the average. It may be unnecessary to read too much in the results; indeed, the questionnaire did not intend to mimic a rigorous contingent valuation study, and, after all, the question about WTP is hypothetical. On the other hand, we might have expected that individuals‟ overstate their WTP, just because the question is hypothetical. If this is the case, then the numbers reported here are upper bounds, and thus re-inforcing the point that demand is not overwhelmingly strong for “green” electricity. Therefore, the results contrast in a rather interesting manner 30

There are some variations regarding the significance of age-categories.

34


with the substantial support given to renewable energy in many countries today. I have not considered the costs, nor does the benefit estimates presented here allow a rigorous comparison with the costs, but the sentiments reported by survey respondents certainly give food for thought. Secondly, a compilation of the literature, combined with the empirics, allows a number of comments on policy instruments. While incentive-based instruments remain preferred instruments in economic analysis, it is interesting to note that “softer” instruments may play a more important role than previously thought. This is borne out by the literature reviewed in this report and, in particular, by the companion paper (Kriström (2008)). Furthermore, the empirical results suggests that “non-economic” variables, such as environmental attitudes and activities in environmental NGOs has an impact on household activity in the residential energy market, in particular regarding the demand for “greener” alternatives. Thirdly, a consistent message throughout has been heterogeneity. There is substantial between and within country variation in many relevant dimensions. The policy import of this finding could well be that “one size fits all” is not necessarily the best policy. A prime example is the energy & climate directive recently established by the EU. We have presented heterogeneities on the demand side and only a cursory look at the supply side shows that countries are dissimilar. Consequently, a concerted action imposes different costs and benefits across Member states. Needless to say, this is not the place for normative statements regarding EU-policy; rather it would appear that heterogeneity should be considered when developing energy policies. As the example from Mexico on electricity pricing showed, countries do often streamline energy policy to cater for distributional concerns, etc. There are countless other examples, so from a within country perspective the heterogeneity message may be moot. Yet, we do see many examples of where policies must be co-ordinated across countries to be effective, another important example being “climate change” policy. How to map these findings into suggestions for an efficient shaping of, let us say, the next Kyoto round is beyond the scope of this report. Yet, it seems to me that at least we can learn something from an economic theory standpoint and further consider the heterogeneity issue. Currently, these are conveniently swept under the carpet by appealing to the separation between efficiency and equity. As I have noted elsewhere, Hourcade (2001, p. 1) sums up the difficulties of the Kyoto-procol and the policyimport of heterogeneities nicely: “[the] delay in „finishing the Kyoto business‟ reveals additional fundamental difficulties stemming from the fact that the cap and trade approach was too often interpreted as an „open sesame‟ solution. This would be the case if the world was a homogenous „tabula rasa‟ as in the simple models for the first year economic students. But it is increasingly obvious that the real world is full of complexities in the form of sectoral heterogeneities and country specifics.” 7.

Summary and Conclusions I set out to address two key questions: 1. How much are households willing to pay to use only renewable energy? Does WTP vary significantly across household groups? 2. How do general attitudes towards the environment (environmental awareness; membership in environmental organization; …) influence demand for renewable energy?

The short answer to the first question is: not much. While there is significant variation across countries, the respondents display a price premium of less than 4%. This finding is well within the range of results reported in recent research. Furthermore, the stated price premium varies across household groups in several ways. However, household income seems to plays little, if any, role, for the level of WTP (expressed as a percentage increase of the current electricity bill). From a theoretical point of view this

35

result is not unexpected, given some simplifying assumptions. But in the general case we expect that income plays a role. Empirically, I find that income plays a more significant role for the binary decision to pay or not. It is more difficult to pin down how much WTP will increase ( as a percentage of the current electricity bill), as income grows. For the second question, a consistent message across the models used here is that environmental awareness and activity/membership in environmental organizations play an important role in determining the demand for renewable energy. I find this relationship to hold across different levels of aggregation and models. This result is found in several other studies as well. The policy conclusions were, in brief, mainly three. First, the significant support given to the introduction of renewable energy in many countries contrasts with the fairly weak demand reported here (and buttressed by other studies). Second, the literature on residential energy demand suggests an important role for incentive-based instruments, yet the literature reviewed together with the empirical results points to a more substantial role for “softer” policy instruments, relative to what earlier assessments of policy instruments have found. Thirdly, a recurrent theme in this cross-country analysis is country heterogeneity. With hindsight it is easy to point to some constructions in the questionnaire that could have been different. There are some easy fixes in any next round of this survey regarding clarifications of energy usage questions, as outlined in the text. However, there are more substantial challenges in a multi-lingual questionnaire of this type, not the least when attitudinal questions are used. The first issue is the mapping between languages and how to secure that the questions are consistent across languages. These issues might deserve additional analysis. Secondly, it is not straightforward to compare sentiments such as “concerned” and “somewhat concerned” in a multi-lingual, multi-country study. Aside from the translation issue, there remains the issue of how such sentiments are parsed by individuals in different cultures. While I am unable to add much substance to the basic issue, it seems intuitively clear that such measurement issues deserve further attention. Needless to say, there is an extensive literature to tap, e.g. the experience harnessed within large multilingual survey research projects such as European Social Survey (www.europeansocialsurvey.org). I close with some more general lessons learned. In the models that have been employed, I find that the country effects are often significant. I interpret this to mean that country idiosyncrasies are often important in understanding residential energy market and the demand for renewables. Respondents in, say, Korea face decisions that are superficially the same as respondents from, say, Sweden. Yet, when considering the choice sets and the constraints (economic and others) these respondents face, the results seem to suggest that there are intrinsic differences that we cannot completely pick up in the models used. Attitudinal factors and memberships in environmental organizations seem to correlate stronger with various decisions, compared to economic parameters. Correlation, it must be stressed, does not imply causation. Finally, I caution that I have explored only a subset of possible econometric models, essentially using a linear predictor within a limited class of probability models. Widening this class to allow for non-linearities and more general probability distributions will be left for future work.

36


REFERENCES

Aranda-Ordaz, F.J. (1982) “On two families of transformations to additivity for binary response data”, Biometrika, 68, 357–63 Arkesteijna, K & L. Oerlemans (2007) „The early adoption of green power by Dutch households: An empirical exploration of factors influencing the early adoption of green electricity for domestic purposes‟, Energy Policy, 33, 2, 183–196 Batley, S, Colbourne, D, Fleming, P. & P. Urwin (2001) „Citizen versus consumer: challenges in the UK green power market‟, Energy Policy, 29,6, 479-487´ Belyaev, Y. & B. Kriström (2009) “Approach to Statistical Analysis of Self-Selected Intervals”, Mimeo, Department of Forest Economics, SLU-Umeå. Bird, L.,Wüstenhagen, R., & J. Aabakken (2002) “A review of international green power markets: recent experience, trends, and market drivers” Renewable and Sustainable Energy Reviews, 2002. Byrnes, B., Jones, C. & S. Goodman, (1999). “Contingent Valuation and Real Economic Commitments: Evidence from Electric Utility Green Pricing Programmes”. Journal of Environmental Planning and Management, 42(2), 149-166. Cameron, T. A. , G. L. Poe, Ethier, R.G. & W.D. Schulze (2002) “Alternative non-market value-elicitation methods: Are the underlying preferences the same?” Journal of Environmental Economics and Management., 44, 3, 391-425. Champ, P.A., & R.C. Bishop (2001) "Donation mechanisms and contingent Valuation: An empirical study of hypothetical Bias." Environmental and Resource Economics 19 383-402. Clark, C.F., Kotchen, M.J., & M.R. Moore “Internal and external influences on pro-environmental behavior: Participation in a green electricity program”Journal of EnvironmentalPsychology, 23,3,237-246.‟ Cosslett, S. R. , (1983) "Distribution-Free Maximum Likelihood Estimator of the Binary Choice Model," Econometrica, Econometric Society, vol. 51(3), 765-82, May. Diaz-Rainey, I, & Ashton, J, (2007) “Characteristics of UK consumers‟ willingness to pay for green energy”, Working Paper. Available at SSRN: http://ssrn.com/abstract=1030530 Dubin, J. and D. McFadden (1984), “An econometric analysis of residential appliance holdings and consumption”, Econometrica, 52, 1, 345-362. Ek, K. (2005) “The economics of renewable energy support”, Doctoral Thesis, 2005:40, University of Luleå, Sweden.

37

Ek, K. & P. Söderholm (2008) Norms and economic motivation in the Swedish green electricity market, Forthcoming, Ecological Economics, DOI:10.1016/j.ecolecon.2008.02.013. Ethier, R.G., Poe, G.L., Schulze, W.D. & J. Clark (2000). “A comparison of hypothetical phone and mail contingent valuation responses for green-pricing electricity programs”. Land Economics 76(1), 5467. Farreras, V., Riera, P. and J. Mogas (2005). “Does gender matter in valuation studies? Evidence from three forestry applications.” Forestry 78(3). Farhar, B. (1999) “Willingness to Pay for Electricity from Renewable Resources: A Review of Utility Market Research.” Golden, CO: National Renewable Energy Laboratory. NREL/TP.550.26148. Farhar, B. & T. Coburn (1999) “Colorado homeowner preferences for energy and environmental policy.” National Renewable Energy Laboratory. Technical Report. NREL/TP-550-25285. June 23, 1999. Fouquet, R. (1998). “The United Kingdom demand for renewable electricity in a liberalized market.” Energy Policy, 26, 3, 281-293. Gan, L., Eskeland, G. & H. Kolshusa (2007)” Green electricity market development: Lessons from Europe and the US “Energy Policy,Volume 35, 1 , 144-155 Hanemann, W. M. (1984) “Welfare evaluation in contingent evaluation experiments with discrete responses”. American Journal of Agricultural Economics, 66, 332–341 Hanley, N. & C. Nevin, (1999) “Appraising renewable energy developments in remote communities: the case of the North Assynt Estate, Scotland.” Energy Policy, 27,9,527-547. Ivanova, I., (2005) “Queensland Consumers‟ Willingness to Pay for Electricity from Renewable Energy Sources.” ANZSEE Conference. New Zeland Centre for Ecological Economics, December 1113,2005. Jensen, K. Menard, J. English,B. & P. Jakus (2004) “An analysis of the residential preferences for green Power – The role of bioenergy” Paper presented at the Farm Foundation Conference on Agriculture as a Producer and consumer of Energy, June 24-25, 2004. Johansson, P.-O, (1987) The Economic Theory and Measurement of Environmental Benefits, Cambridge University Press, Cambridge, UK. Kriström, B. (2008), “Empirics of residential energy demand”. Paper presented at the OECD Workshop on Household Behaviour and Environmental Policy: Empirical Evidence and Policy Issues, 15-16 June 2006, Paris. Kotchen, M. & M. Moore (2007) “Conservation: From Voluntary Restraint to a Voluntary Price Premium”, (published online 17 Juli 2007) Environmental & Resource Economics, 2008 40,195–215 Longo, A., Markandya, A. & M. Petrucci (2006) “The Internalization of Externalities in the Production of Electricity:Willingness to Pay for the Attributes of a Policy for Renewable energy”, FEEM Working Paper, 132, 2006. McCullagh P. and Nelder, J. A. (1989) Generalized Linear Models. London: Chapman and Hall.

38


Pichert, D. & K.V. Katsikopoulos (2007) ”Green defaults: Information presentation and proenvironmental behaviour”, Journal of Environmental Psychology, Reiss, P. & M. White (2005) “Household electricity demand, revisited.” Review of Economic Studies, 72, 3, 853-883, Available at SSRN: http://ssrn.com/abstract=791852 Reiss, P. & M. White (2008) “What Changes Energy Consumption? Prices and Public Pressures” Forthcoming, Rand Journal of Economics, 39, 3, 636-663. Roe, B., Teisl, M.F., Levy, A., & M. Russell, (2001) “US consumers' willingness to pay for green electricity.” Energy Policy, 29, 11, 917-925. van Rooijen, S. & van Wees, .M.T. (2006) “Green electricity policies in the Netherlands: an analysis of policy decisions” Energy Policy, 34, 1, 60-71. Rose, S.K., Clark, J , Poe, G.L., Rondeau, D, & W.D. Schulze (2002) “The private provision of public goods:tests of a provision point mechanism for funding green power programs” Resource and Energy Economics, 24, 131–155. Rowlands, I.H., D. Scott, & P. Parker “Consumers and green electricity: profiling potential purchasers”, Business Strategy and the Environment, 12 Issue,1, 36 – 48. Sundqvist (undated, p.5) “Quantifying Non-Residential Preferences over the Environmental Impacts of Hydropower in Sweden: A Choice Experiment Approach, Mimeo, Department of Economics, University of Luleå. Zarnikau, J. (2003) “Consumer Demand for „Green Power‟ and Energy Efficiency.” Energy Policy 31, 155, 1661-1672. Whitehead, J.C. & T.L. Cherry (2007) “Willingness to Pay for a Green Energy Program: A Comparison of ex-ante and ex-post hypothetical bias mitigation approaches” Resource and Energy Economics 29, 247-261. Winter, J. (2003) “Bracketing effects in categorized survey questions and the measurement of economic quantities”, Sonderforschungsbereich 504, Universität Mannheim & Sonderforschungsbereich 504, University of Mannheim, 02-35. Wiser, R., (2003) “Using Contingent Valuation to Explore Willingness to Pay for Renewable Energy: A Comparison of Collective and Voluntary Payment Vehicles.” LBNL-53239, Lawrence Berkeley National Laboratory, Berkeley, California, August. Wiser, R.H., Fowlie, M., & E. Holt. (2001) “Public goods and private interests: understanding nonresidents residential demand for green power” Energy Policy, 29,13, 1085–1097.

39

APPENDIX

A1.

Econometric approach

The econometric analysis is carried out by using the free package R. To estimate the logit model, I have used the lrm routine in the Design library and the glm routine from the Stats library. The latter routine provides options to estimate the alternative Cauchy and complementary log-log models. For the survival analysis, I used the psm routine in the Design library to estimate the Weibull model. The survival analysis here heuristically amounts to maximize a log-likelihood over a set of parameters, where one sums over the individual contributions (suppressing indeces) Log{F(x_up)-F(x_low)} F is the distribution function of the censored random variable X and x_up and x_low the upper and lower bounds of the intervals. As noted, I choose F to be a one-parameter Weibull distribution. Given an estimate of F, there is sufficient information to calculate any statistics of interest. I return below to how the survival analysis and the welfare analysis is connected. A2.

Economic model

Let z denote environmental quality, q the consumption of electricity, p the price of electricity and m individual income. Let V(p, m, z) denote a standard indirect utility function, decreasing in p and increasing in m and z. I suppress indeces (to identify individuals) for simplicity. Suppose that a switch to “green” electricity induces an improvement in the sense that z_1 > z_0, where z_0 is the status quo quality. Let a be the maximum percentage increase of the current electricity bill (pq_0) the individual is willing to pay for the implied environmental improvement. Thus, a is defined implicitly be the equation V(q_0, m-pq_0, z_0) = V(q_0, m-pq_0(1+a), z_1)

(1)

Consequently, I interpret the status quo as given and consider a rationing version of the indirect utility function (see Johansson (1987) for details about rationing models in this context). Approximating (1) linearly, I obtain c/b= a* pq_0

(2)

where c is the marginal utility of z and b the marginal utility of income. Marginal willingness to pay (c/b) is thus proportional to the current bill. Alternatively, a is, to the first-order, inversely proportional to the existing electricity bill. The higher the bill, the lower is a, ceteris paribus. Note that a does not depend on income in this simple linear model. If, given a, information were available on pq_0, the marginal willingness to pay can be estimated directly. However, we have no information about pq_0 in the survey.

40


Now, Q69 does not ask about a directly, rather a is in 1 out of 5 known intervals, ie a is censored. I assume that a is censored at random. Thus a is independent of the particular intervals, the censoring mechanism, we have chosen. If the individual behaves according to the utility theoretic model, the individual has a fixed a independently of how we have chosen to elicit it. However there is an extensive literature in economic psychology on anchoring that challenge the censored at random assumption. See Winter (2003) for an interesting experiment, showing how the choice of intervals can affect respondents behaviour. Finally, using Hanemann‟s (1984) framework, it is possible to translate the model in equation (1) so that it fits the way Q69 is actually asked. Hanemann (1984) can be also used to show that a linear utility function gives a welfare expression analogous to (2), even if a is censored. Hanemann (1984) illustrates his theory with a simpler type of censoring, but his approach is directly applicable here.

41