Intervention strategy to stimulate energy-saving behavior of local residents

Energy Policy 52 (2013) 706–715

Contents lists available at SciVerse ScienceDirect

Energy Policy journal homepage: www.elsevier.com/locate/enpol

Intervention strategy to stimulate energy-saving behavior of local residents Q. Han a,n, I. Nieuwenhijsen a, B. de Vries b, E. Blokhuis a, W. Schaefer a a b

Construction Management and Urban Development, Department of Built Environment, Eindhoven University of Technology, PO Box 513, 5600 MB Eindhoven, The Netherlands Design Systems, Department of Built Environment, Eindhoven University of Technology, PO Box 513, 5600 MB Eindhoven, The Netherlands

H I G H L I G H T S c c c

A latent class model to identify segments with preferred energy-saving interventions. An integrated energy-saving behavior model of casual relations. A tree structure overview of potential interventions

a r t i c l e i n f o

a b s t r a c t

Article history: Received 2 September 2011 Accepted 11 October 2012 Available online 3 November 2012

This study investigates intervention strategy in stimulating energy-saving behavior to achieve energy neutral urban development. A tree structure overview of potential interventions classified into three categories is revealed. An integrated behaviour model is developed reflecting the relations between behaviour and influence factors. A latent class model is used to identify segments of local residents who differ regarding their preferences for interventions. Data are collected from a sample of residents in the Eindhoven region of the Netherlands in 2010. The results indicate that social-demographic characteristics, knowledge, motivation and context factors play important roles in energy-saving behaviour. Specifically, four segments of residents in the study area were identified that clearly differed in their preferences of interventions: cost driven residents, conscious residents, ease driven residents and environment minded residents. These findings emphasize that the intervention strategy should be focused on specific target groups to have the right mixture of interventions to achieve effective results on stimulating them to save energy. & 2012 Elsevier Ltd. All rights reserved.

Keywords: Energy-saving behavior Intervention strategy Latent class model

1. Introduction Sustainable urban development has developed in the past a few decades in the Netherlands to a mature subject of policy, research and innovation with various titles, such as low carbon city, energy neutral city, etc. The strategy of the local government to realize the energy-neutral target is based on the Trias Energetica: reduce energy demand, use renewable energy resources and use fossil fuels efficiently. The first step in this approach is to reduce energy demand because energy-saving is one of the cheapest ways to reduce CO2 emission (IEA, 2008). More than 25% of residential energy use could be reduced using readily available technologies (Gardner and Stern, 2008). Despite all efforts currently being undertaken, the energy-saving rate of residents is still very low (Laitner et al., 2009). Therefore, it is important to investigate how residents can be encouraged to save energy.

n

Corresponding author. Tel.: þ31 40 2475403; fax: þ 31 40 2478488. E-mail address: [email protected] (Q. Han).

0301-4215/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.enpol.2012.10.031

There are two different types of energy-saving behaviors: investment behavior and curtailment behavior. Investment behavior is about spending money on the improvement of energy efficiency, and consequently saving energy. Curtailment behavior is about reducing energy usage by behavioral changes, such as shortening shower duration, lowering room thermostat settings. Contextual factors, knowledge, motivations, abilities and sociodemographic variables may influence such energy-saving behavior (Lutzenliser, 1993). There are certain interventions that local government can apply to promote energy-saving behavior, such as providing information, demonstration, offering free products, commitment with goal setting, giving feedback, rewards, financial support and legislation. There are a few researches about behavior models with causal relations between influence factors and behavior. Olander and Thøgersen’s (1995) developed Motivation-Opportunities-Abilities model (MOA-model) with the focus on behavior in general. ValueBelief-Norm model developed by Stern (2000) addressed the environmental behavior in particular. However, an integrated behavior model with the focus on interventions and energy-saving behavior of residents is still missing.

Q. Han et al. / Energy Policy 52 (2013) 706–715

There are also researches that investigate household preferences for energy-saving measures using conjoint analysis (Poortinga, 2003), the relative impacts of two social change paradigms on residential behavioral energy-savings using regression model (Tiedemann, 2009), and behavior patterns and household profiles related to energy spent on heating using factor analysis(Guerra Santin, 2011). However, as we believe that people are different, interventions aimed at residential energy-saving may have different influences for different people (Guns, 2007). The intervention strategy that recognizes and accommodates the ways in which people differ will be more effective. In this paper, we propose a latent class model to tackle this problem. This paper is structured as follows: in Section 2 energy use behavior and its influence factors are discussed; Section 3 describes interventions; in Section 4 theoretical model of latent class are proposed; Section 5 provides information of data collection; in Section 6 all the analysis results are reported; and its implications for policy making are discussed in Section 7; finally some conclusions are drawn in Section 8. The main objectives of this study were to: (1) provide a overview of potential interventions, (2) attempt to present an integrated energy-saving behavior model of casual relations, (3) apply a new approach to identify segments with preferred energy-saving interventions. The results of this paper can provide decision support for local government in their policy making to effectively stimulate residents to save energy.

2. Energy use behavior Households in the Netherlands use energy directly in forms of natural gas and electricity and indirectly through the energy that is used to develop the products and foods that households consume (Vringer and Blok, 1995). The amounts of electricity consumed and natural gas used per household are comparable to a total energy of about 73.4 GJ/yr on average. The amount of electricity and natural gas use per household slightly changed over the past 10 years. In comparison to former years more electricity and less gas is used per household (CBS, 2011). Arkel et al. (1999) distinguished energy use into two categories, namely dwelling related energy use and user behavior related (or appliances related) energy use. The dwelling related energy use consists of space heating (which is influenced by insulation and ventilation) and electricity consumption for mechanical ventilation is present. The user behavior related energy use consists of using all kinds of appliances related to shower, cleaning, cooling and preparation of food and audio-, video- and telecommunication, etc. Lighting is conditioned by the design of the house together with the lifestyle of the residents. Therefore, it is part of both dwelling related energy use and user behavior related energy use. In this research, energy demand is assumed to depend on both the behavior of the residents and the characteristics of the dwelling. Consequently, energy-saving are related to both the curtailment and investment behavior of the residents. Since the energy use for heating covers more than 50% of the total energy uses of a household (Itard et al. 2009), technical characteristics of dwellings – energy label – are important factors when determining energy demand and consequent energy-saving potentials. Dwelling technical characteristics such as constructional measures, insulation measures and method of heating and lighting are important factors. Ownership of the housing, duration of the residence may influence the maintenance of the dwelling and indirectly impact the energy efficiency. Recent studies have shown that residents’ behavior has a significant impact on the energy demand of households (Guerra

707

Santin et al., 2009). Such behavior has a strong association with the characteristics of the user (Guerra Santin and Itard, 2010a). The study conducted by Leidelmeijer and Cozijnsen (2010) shows that age is an important factor in energy use. The age of residents influences thermostat settings, frequency and length of shower, and the number of used appliances (Groot et al., 2008). Moreover, the age of residents may imply the strength of the habitual behavior, since the behavior is likely to be repeated when outcomes are satisfactory, and such habitual behaviors are commonly observable in elderly people (Ariely, 2009). The household-size and composition, which represents the total number of people living in the same dwelling, determines the frequency of activities over the week, such as washing, dishwashing, tumble drying and refrigeration (Groot et al., 2008), therefore directly related to the total energy demand (Abrahamse and Steg, 2009). Furthermore, other socio-demographic factors, such as income, education level, and work status, may serve as barriers or opportunities for energy usage and saving. There are two types of energy-saving behaviors: investment behavior and curtailment behavior. Investment behavior is about investment in the measures to increase the quality of dwellings in terms of energy efficiency (e.g., change the old single glass window to the double HR glass), or the purchase of energy-efficient appliances to reduce energy usage (e.g., LED light). Behaviorally efficiency improvements usually involve one-time purchase decisions: there is a financial expense and the potential of future monetary savings; it is energy smart technology choice without loss of the amenities. Curtailment behavior is about the decrease in the usage of existing equipments or appliances by behavioral changes, such as shortening shower duration, lowering thermostat setting, etc. Behaviorally these responses usually must be repeated or continual to achieve maximum energy-savings: they rarely cost money, but they do ask change in habit and lifestyle adjustment; it is energy smart lifestyle choice with the possibility of loss of amenities. However, with the energy-saving behavior there is also a risk for rebound behavior (Berkhout et al., 2000). Contextual factors, knowledge, motivations, abilities and socio-demographic variables are the important factors that could influence residents’ energy usage and saving (Steg, 2008). Because investment and curtailment behavior involve different sorts of behavior, they may be influenced by different factors (Lutzenliser, 1993) and consequently promoted by different interventions. For example, investment measures are more available to higher income residents and to homeowners, whereas curtailment measures may be the only option for renters and for those who cannot afford new equipment. There are multiple studies about behavior models in general with causal relations between influence factors and behavior. The MOA-model is often used and it visualizes the theory of reasoned action (Olander and Thøgersen, 1995). According to this model, behavior is caused by three main influence factors (motivation, ability and opportunity). Motivation includes beliefs, attitudes, intention and social norms. Habits and knowledge are part of the ability factor. The behavior model addressing the energy aware behavior and the energy use (Van de Maele-Vaernewijck et al., 1980) concentrates on demographic factors and housing characteristics as influence factors. Certain aspects of this model overlap with the value-belief-norm (VBN) theory model (Stern, 2000). The VBN theory is a causal chain leading to different types of environmental behavior. The model consists the variables such as: personal values (altruistic, egoistic and traditional), belief, and personal norms for pro-environmental action. Considering our specific topic about interventions and energy-saving behavior, an integrated energy-saving behavior model is required. Although people often seem to be aware of the environmental and energy problems, they often do not act in line with their

708

Q. Han et al. / Energy Policy 52 (2013) 706–715

concerns, because they rarely make a conscious decision to use energy (Stern and Aronson, 1984), and the total household energy demand is still rising. This seems to be partly caused by a lack of insight in the relation between user behavior and energy usage (Laitner et al., 2009), and partly caused by the perceived obstruction of regulation and public opinion (Cialdini, 2007). Government has few (financial and legal) means to push the energy neutral target forward in the public sphere. The current governmental financial incentives in the Netherlands appear to be inadequate because household energy demand keeps rising (Abrahamse, 2007; Guerra Santin and Itard, 2010b). The local government is dependent on the voluntary participation of their residents to save energy. Therefore, the interventions that could stimulate voluntary participation are essential.

3. Interventions In the literature (Abrahamse, 2007; Abrahamse et al., 2005), several types of interventions for stimulating energy-saving are introduced. These interventions seize on the different factors of the behavior, influencing both investment and curtailment behavior. We can classify them into three main categories: antecedent, consequence and structural interventions. However, a systematic oversight of all possible interventions and its subtypes did not exist in the literature yet. In this section, first a brief description of the potential interventions is given, and then a tree structure is developed to provide a clear overview (shown in Fig. 1). Antecedent interventions increase people’s knowledge and strengthen their concern with energy problems in order to encourage energy-saving behavior (Dietz and Stern, 2002). There are different types of antecedent interventions, such as providing information, giving demonstration, building commitment, setting

a goal, offering free products, etc. Providing information aims to increase people’s awareness about energy problems and enrich their knowledge about the way to solve these problems. Information campaigns are commonly used to promote energy-saving. There are different possibilities to provide people with information, to name a few: workshops, mass media campaigns, and websites. In contradiction to the general belief, mass media campaigns appeared not to be very effective (Abrahamse, 2007). Providing specific information that is tailored to a particular type of household is essential and seems to be effective (Abrahamse and Steg, 2009). Demonstration provides examples of recommended behavior. In general, people will follow those examples when they are understandable, relevant, meaningful and rewarding. Free products offer people knowledge and strengthen their concern for saving energy in a passive manner by giving them the possibility to try out. Commitment strategies contain a promise to change behavior, in this case to save energy. If the promise is pledged to oneself it may only activate a personal norm (i.e., a moral obligation). When the promise is made public (e.g., by leaflets), social norms (i.e., expectation of others) also play a role in influencing the energy-saving behavior (Lucas et al., 2008; Cialdini, 2005). The promise to save energy can be linked to a specific goal such as reduce energy use by 5% within 5 years. The commitment of goal setting is often used in combination with other interventions (e.g., feedback). Consequence interventions are based on the assumption that the presence of positive or negative consequences will influence behavior (Darby, 2006). Feedback and rewards are two incentives that associate with the positive or negative outcomes. The study conducted by Abrahamse (2007) reveals that providing feedback about the reached saving rate is very effective, and appeared to be more effective when such feedback was given more frequent and related to a specific saving goal (e.g., the commitment of goal

Mass media

Website

Brochure

TV/Radio

Newspaper /magazine

Posters

Network sites

Mobile advertisement (auto, bus, etc)

Interpersonal

Briefing meeting

Workshop for stakeholders

Energy market

Education programs

News letters (paper)

Flyers

Education films

Tailored media

Front office (helpdesk)

Helpdesk by phone and email

Energy advices

Energy behavior coach

Demonstration

Neighbors that save energy

Nearby companies

Associations ambassador

Display measures in (scaled) model

Demonstrationby celebrities

Free products

Saving measures (energy boxes)

Commitment /goal setting

With lesser

With home owner association

With own household

With neighborhood

With municipality

Feedback

Energy use appliances

Benchmark comparable householders

Energy labels

Insight in behavior (shower coach)

Rewards

Awards /prizes

Tax deduction from land value tax

Price policies

Product tax

Energy tax with larger contrast

Subsidies

Reduction costs founds

Reduction costs measures

Loans

Low rents

Green loan

Information

Interventions

Financial / legislation

Legislation

Building regulation

Current energy usage

Antecedent

Consequence Stamp tax

Removal tax

Building performance certificate

Fig. 1. Tree structure of all possible interventions.

Structural

Emailing

Q. Han et al. / Energy Policy 52 (2013) 706–715

setting). Rewards seem to have a positive effect on reducing energy usage, but it appeared that effects of rewards are shortlived (Geller, 2002) Structural interventions are aimed at changing the contextual conditions to facilitate behavior changes (Steg, 2008). Changes in the circumstances of energy usage may be used to make proenvironmental behavior more attractive. The costs and benefits of behavioral alternatives may be adjusted in various ways through financial support (Steg and Vlek, 2009) and/or legislation. Changes in physical, technical and organizational systems can alter the availability in products and services. High energy consuming products and services can be made less attractive or even unavailable by policies, while energy-efficient products and services may be promoted by subsides (e.g., LED lights and other energy-efficient appliances). Pricing policies can be used to set higher prices for conventional energy intensive products/options via taxation (e.g., higher tax for using fossil fuel energy). Legal regulations can be implemented (to prevent splitincentive for example). Legislative requirement can be employed to dwelling, such as energy performance certificates (i.e., energy label) or building regulation requirement (i.e., building code) addressing refurbishment, minimum requirements on HR glass and other building products. These legislations are often set by state government and engaged with financial support (e.g., subsides) and economic consequences (e.g., tax or fine if not fulfilled). The tree structure in Fig. 1 shows the different main types and subtypes of interventions identified in the literature. The ownerresidents-panel (Vereniging Eigen Huis, 2009) has been used to validate the tree structure. As it appears, there are three categories with seven main interventions recognized: information; demonstration; free products; commitment with goal; feedback; rewards; financial support and legislation.

4. Latent class model analyses As we all know that people are different, the same intervention might have different impacts on different people (Guns, 2007). A latent class model (Greene and Hensher, 2002; Swait, 1994) was used to segment the respondents regarding their preferences in interventions. Latent class model (LCM) involve characterizing segments from observed measures (i.e., choice of the intervention packages) and permit choice attributes data (i.e., preference of the intervention) and individual characteristics (i.e., social-demographics and current energy behavior) to simultaneously explain choice behavior. The advantage of the latent class segmentation approach over other segmentation approaches such as factor analysis using socio-demographics is that the segments are behavior based and are therefore more actionable and more directly relevant to management and planning decision making (Greene, 2001). It has been widely applied and proves to be useful in the fields of social science (Collins and Lanza, 2010), such as transportation (e.g., travel mode/route choice), marketing (Swait and Adamowicz, 2001; BenAkiva and Boccara, 1995), leisure (Kemperman and Timmermans, 2006; Boxall and Adamowicz, 2002). Regardless of its potential, the latent class model has not been widely used in the field of urban sustainable development research especially not in policy making. In our study, a latent class formulation is proposed that simultaneously groups residents and estimates a separate set of utility parameters for each segment. Formally, the model can be described as follows: an individual resides in a latent class, c, which is not revealed to the researcher. The existence of a fixed number of C classes in a population is assumed. It assumes that individuals compare a predefined set of discrete choice alternatives and choose the alternative that is most attractive. Formally the attractiveness of the alternative is

709

called ‘‘utility’’, which is a relative measure. The utility underlying individual i’s choice among J intervention packages at choice situation t, given that he/she belongs to latent class c, can be expressed as 0

U jit ¼ bc Xjit þ ejit

ð1Þ

where Xjit is a union of all attributes that appear in all utility functions, and bc’ is a class specific parameter vector. The ejit indicates the unobserved heterogeneity for individual i and package j in choice situation t. The number of choice situations and the size of the choice set may vary by individual with the assumption that the same individual is observed in several choice situations (i.e., makes one or more choices). Within the class the choice probabilities are assumed to be generated by the multinomial logit (MNL) model that is the alternative which percepts with higher utility will have a higher chance to be chosen  0    exp bc Xjit P yit ¼ j9class ¼ c9 ¼ PJ ð2Þ   0 i j ¼ 1 exp bc Xjit As noted, the class is not observed. Class probabilities are specified by the MNL form  0  exp yc Zi Pðclass ¼ cÞ ¼ Q ic ¼ PC ð3Þ  0  , yc ¼ 0 c ¼ 1 exp yc Zi where Zi is an optional set of observable individual, choice situation invariant characteristics. If no such characteristics are observed the class specific probabilities are a set of fixed constants that sum to one. In other words, the individual has an equal probability belong to one of the classes. For individual i, the model’s estimate of the probability of a specific choice is the expected value (over classes) of the class specific probabilities 2  0  3   exp bc X jti 4 P yit ¼ j ¼ Ec PJ  0 5 i exp bc X jti j¼1 2  0  3 C X exp b X jti c ð4Þ Prðclass ¼ cÞ4PJ ¼  0 5 i exp bc X jti c¼1 j¼1 The model is estimated by direct maximization of the log likelihood (Greene, 2001).

5. Data collection Data for this study were obtained using an online questionnaire. With the help from the municipality of Eindhoven, the survey has been distributed to a sample of 1500 households in Eindhoven in 2010. Each head of a household (i.e., the family member with the highest income) in the sample received a letter with a request to participate in the survey. In total 309 respondents filled in the questionnaire on behalf of their household and 265 were valid to be used in the analysis. The lower respondents’ rate might be caused by that it was not possible to send the respondents in the sample an email with a direct web link to the survey on the Internet. The choice experiment is designed with the identified seven main interventions, which are explained briefly to respondents as follows: (1) information about energy efficiency and opportunities for households (e.g., brochure, information evening or an energysaving help desk, etc.); (2) demonstration of how other energy-saving measures apply (e.g., by neighbors, acquaintances, association or a model home, etc.); (3) free products of small energy-saving measures at home to test (e.g., energy box with LED bulbs, weather stripping, water-saving shower head, etc.); (4) commitment/ goal setting on energy use (e.g., agreement with own household or district neighbors: we will reduce energy consumption by 10% within 5 years);

710

Q. Han et al. / Energy Policy 52 (2013) 706–715

(5) feedback of insight into energy use and behavior in the house (e.g., display of consumption per device, using smart meter, or a comparison with other households, etc.); (6) rewards received when actually saved energy (e.g., a prize or a deduction for tax, etc.); (7) financial/ legislation support from the government (e.g., via a loan with a low interest rate, green mortgage, subsidies or adjusting price, etc.). Each intervention has two levels about whether or not it is present in the intervention package. As it would be troublesome to test a full factorial design of 27 ¼ 128 different packages, in this experiment an orthogonal fraction of the full factorial design was selected resulting in a fractional design of 16 different packages. Each respondent was asked to complete eight choice tasks. One choice task consists of two selected intervention packages from these 16 packages and a baseline option of neither of the two. One example of the choice task is presented in Fig. 2. In addition, there are many other questions in the survey about the current energy-saving behavior such as past and intended investment in HR glass, knowledge of energy problems, motivation for energy-saving, experienced obstruction by regulation and public opinions, etc. Table 1 provides a brief list of concerned aspects and related questions. Respondents need to report self evaluation regarding these aspects on a 0 (worst) to 5 (best) scales. Finally, personal information such as age, household composition, education level, employment status, duration of current residence and income level was collected. See the top of the web page for an explanation of the different interventions. Which of the following packages convinces you to save energy? If neither of the 2 packages convinces you to save energy, choose “neither”. Package 1 Information Free products Feedback Commitment Financial /legislation

Package 2

Neither

Rewards Financial /legislation

To test the questionnaire for its clarity and ambiguity, it was first presented to 20 respondents for a pilot test. Taking into account their comments and suggestion, unclear questions such as lifestyle were excluded, some open questions were changed to semistructured and other too ambiguous questions were adjusted. Also the presentation of the choice sets of the experiment is improved in an attempt to avoid flat lining answer as much as possible.

6. Results The results of the analysis included a description of the distribution of the respondents in the sample on a series of socio-demographics, an integrated energy-saving behavior model of casual relations, the results from the estimation of the LCM model including the relation between preference for interventions and the characteristics of the residents. 6.1. Sample descriptions The profile of the respondents is presented in Table 2. The respondents in the sample were compared with census data of the residents of the region according to some selected characteristics including age, household type, income, education level housing type and the number of working hours. Several observations were made. First, slightly more high educated people were willing to participate in our study. Further, and not surprisingly, the younger than 27 years of age category participated less in our study. For household type and income, no significant differences between the sample and the residents of the region were found. 6.2. The energy-saving behavior model

Fig. 2. An example of choice task for intervention packages.

With the survey data, an integrated energy-saving behavior model was developed and represented in Fig. 3. The social-demographic factors and the dwelling characteristics are included as influential factors in the model. Since social-demographic characteristics

Table 1 List of self evaluation aspects regarding current energy-saving behavior. Category

Aspects

Context Energy-saving opportunity possibilities Public opinion, regulation Motivation To invest (or not) To change (or not) Knowledge

Curtailment behavior

Energy problems Measures, advantages Obvious saving Efficient electricity

Investment behavior

Shorter use Relocate appliances Less standby Commonly known investments Special heating Shutters Double HR glass Heat retention

Related questions Money budget, technological possibilities, etc. Opinion of acquaintances, governmental regulation, etc. House quality, save energy bill, environmental concern, etc. Effort, comfort, importance, use appliance efficiently, experience in energy-saving behavior, etc. Climate change, environmental problems, depletion of fossil fuels, future uncertainties, etc. Measures to save energy, advantages for society and/or for individual household, etc. Lower thermostat setting, lights off when leaving room, cook with lid on the pan, dry laundry outside, low temperature washing and full drum, etc. Defrost Freezer/refrigerator regularly, thaw frozen products in refrigerator, set refrigerator/ freezer one degree less cold, replace broken appliances for energy-efficient variants, etc. Shorter showers, water-saving shower head, etc. Refrigerator under shade and/or away from heating system, etc. Less devices on standby, use sockets with power switch, etc. Insulation: Wall, roof, floor, pipe, etc. Boiler with high efficiency, HR boiler, etc. Solar boiler, micro combined heat and power generator, heat pump, low temperature heating, heat recycle system, floor heating, etc. Blinds Double glass, double HR glass, etc. Reflective material behind radiators, crack sealing, cowl flap, storm windows, etc.

Q. Han et al. / Energy Policy 52 (2013) 706–715

Variables

Levels

%

Age

o 27 27—34 35—45 46—59 459 Primary level Medium level Professional education University level Below h2,500 Between h2500 and h3750 Between h3750 and h5000 More than h5.000 No answer No information Less than 12 h More than 12 h Single Couple Family with children Private owned house Private rental house Rental house from housing corporation

15 21 23 20 21 3 15 13 69 31 23 18 18 10 38 10 52 39 28 33 59 10 31

Housing type

Socio-demographic characteristics C: Energy saving possibilities C: Public opinion & regulation

- Financialsupport /Legislation Antecedent strategies - Commitment /Goal setting - Demonstration - Information - Free products

The latent class models were estimated using maximum likelihood estimation. As the right number of segments in the data could not be observed and predefined, latent class models from one up to 4-segment solutions are estimated. The model became unstable when it is estimated for more than 4-segments. Table 3 presents the number of segments, the corresponding log likelihood values at convergence (LLB), rho-squares, Akaike Information Criteria (AIC) and Bayesian Information Criteria (BIC). The log likelihood values at convergence and rho-squares showed improvement in model fit as the number of segments increased. This finding confirmed heterogeneity in preference and suggested the existence of latent segments. To determine the optimal number of segments, the AIC and BIC statistics were inspected together with the corresponding estimates. The AIC kept decreasing with an increasing number of segments that suggests a 4-class model, while the BIC values deviate very slightly Table 3 Statistics for the latent class models with 265 respondents. Number of classes

Number of parameters (p)

Log likelihood at convergence (LLB)

Log likelihood evaluated at 0 (LL0)

r2

1 2 3 4

8 17 26 35

 1492.42  1263.47  1241.89  1197.29

 1616.94  1616.94  1616.94  1616.94

0.1605 0.2227 0.2336 0.2588

Housing characteristics

Financial ability

Motivation

Household type

Knowledge

Work status

Context

Income level

Structural Strategies

Education level

6.3. Segments in preference for interventions

M: Willingness to invest M:Effort / willingness to change

Obvious saving e.g., heating

Curtailmentbehavior

Table 2 Characteristics of respondents.

pricing policies, subsidies, etc) influence the context conditions not only in terms of energy-saving possibilities but also public opinion and regulation. Consecutively it modifies both financial ability and motivation of the residents, and in sequence influences their energysaving behavior. In Fig. 3, these causal relations are depicted by the arrows in the energy-saving behavior model. However, the model should not be visualized as a strict flowchart since the relations are sometimes not strictly only one direction.

Energy saving behavior Rewards

K: Energy problems K: Measures & advantages

Fig. 3. The energy-saving behavior model.

Investmentbehavior

influence all three psychological factors (knowledge, motivation and ability), it is shown as a layer below these factors. The model consists of both contextual factors such as energy-saving possibilities at macro level and individual factors such as financial ability at micro level. The seven main interventions are also included to show the potential influences. The causal relation between influence factors and behavior are tested with factor analysis, correlation and regression analysis using the self reported evaluation levels. The model revealed that energy-saving behavior has strong associations with some main influence factors such as motivation, knowledge, financial ability and context opportunity. The various interventions could influence different factors directly and/or indirectly, which consequently impact the energy-saving behavior. For example, structural interventions (such as legislations,

711

Efficient electricity Shorteruse behavior Relocate appliances Less standby Commonly known investments (insulation etc) Special heating

Shutters Double HR glass Heat retention

AIC

BIC

2.0167 1.7211 1.7042 1.6563

2.0452 1.7817 1.7969 1.7811

adjusted

712

Q. Han et al. / Energy Policy 52 (2013) 706–715

among 2-segment, 3-segment and 4-segment model. Since our goal of the latent class analysis is managerial relevancy of the segmentation solution, therefore the most important aspects to consider when choosing a solution for segmentation purposes are its interpretability and stability (reproducibility). On the basis of these considerations, the 4-segment model seems the best model in this particular case. The adjusted rho-square values – indicating the proportion of variability in a data set that is accounted for by a statistical model – are equal to 0.26 for 4-segment model. If the adjusted rhosquare value is above 0.20, it indicates a fair fit of the model. We had a fair fit for the 4-segment model, even though the number of respondents is low. A higher number of respondents may improve the model performance, but may not change the underlying division of the segments, since the estimation mechanism simultaneously groups residents and estimates a separate set of utility parameters for each segments. The results of the 4-segment latent class model are presented in Table 4. For reasons of comparison, the results of the onesegment model are also shown in this table. For each segment, the parameter estimates and their significance are presented with significant values indicated in bold with a superscript star. Most parameter values are significant at the 95% confidence level. The segment probabilities are all significant at the 95% confidence level and also presented in Table 4. As it turned out, it is clear that the estimates of 1-segment model could not reveal the detailed preferences of the respondents as shown in 4-segment model. For example, the estimate of the specific constant of intervention packages is significant with the value of 1.17 for the 1-segment model. It means that respondents have negative preference of choosing a package over selecting the option ‘‘neither’’, which reflects the negative initial interests in the

pro-environmental interventions. However, in the 4-segment model, the estimate of the specific constant varies from the significant value of  1.94 for segment 1 to the significant value of 3.03 for segment 4, which reveals that segment 1 dislikes interventions, but segment 4 prefers it. Moreover, the estimates of intervention ‘‘commitment’’ are significant for segment 2 and segment 3 with opposite preferences in 4-segment model. However, in the 1-segment model estimate, these opposite preferences cancelled each other out, and left us with an insignificant outcome. Therefore, it is clear that if we only look at the 1-segment model, we will be misleading. In Fig. 4 the preferences and dislikes of residents in 4-segment model for the seven main interventions are pictured. Further, for each respondent, individual preference parameters and segment probabilities were estimated. Based on these individual parameters respondents were classified into the four segments. Cross-tabulations, chi-square tests and analysis of variances were used to test and explore the relations between the four preference segments and the socio-demographic and current energy-saving behavior. The results are show in Table 5. Segment 1 was labeled cost focused residents (20%, 46 respondents). Respondents in this segment are mainly about 27 till 35 years old and have medium knowledge about energy problems. In this segment, the degree of education is average and the main daytime activity is work for more than 12 h per week. Respondents in this segment have the relative low average income and experience medium obstruction by governmental regulation and public opinion to save energy. Many of them live in their current residences for less than 2 years and a few of them live in their houses for more than 10 years. More of them live in rental houses with a lower energy label. They are highly aware of cost and basically save energy to reduce costs.

Table 4 Results for the latent class model estimation. 1-Segment model

4-Segment model Segment 1, cost focused residents

Constant Information Demonstration Free products Commitment Feedback Rewards Financial Segment probabilities

Segment 3, ease driven residents

Segment 4, environment minded residents

Para-meter t-value Para-meter

t-value

Para-meter

t-value

Para-meter

t-value

Para-meter

t-value

 1.17n 0.29n 0.27n 0.40n  0.11 0.45n 0.82n 0.83n

 3.90  1.15  0.72  0.48 1.88 1.96 4.48 6.51 7.98

 0.64n 1.03n 0.41n 1.63n 0.80n 0.99n 2.11n 1.42n 0.43n

 2.10 7.13 3.82 11.10 6.24 9.72 11.57 10.88 12.65

 0.16 0.85n 0.49n 0.38n  1.69n  0.004 0.40n 0.12 0.18n

 0.68 5.29 3.23 2.38  9.16  0.03 2.19 0.79 6.85

3.03n  0.51n  0.1 0.27  0.42 0.49n 0.61n 0.67n 0.19n

 8.99  2.06  0.42 1.13  1.78 2.09 2.66 2.88 7.13

 8.34 3.48 3.52 4.88  1.31 5.88 8.63 10.33

 1.94n  0.38  0.22  0.18 0.70 0.54n 2.16n 3.14n 0.20n

Significant at po 0.05 level.

4 Segment 1

Segment 2

Segment 3

Segment 4

3 2 1

Fig. 4. Preference in intervention strategies per segment.

Financial

Rewards

Feedback

Commitment

Free products

-2

Modelling

-1

Information

0 Constant

n

Segment 2, conscious residents

Q. Han et al. / Energy Policy 52 (2013) 706–715

713

Table 5 Main characteristics of four segments of respondents.

Age Income monthly Education level Housing ownership type Housing energy label Duration of current residence Current curtailment energy-saving behavior Current investment behavior Experienced obstruction by regulation and public opinion Knowledge of energy problem Motivation

Segment 1, cost focused residents

Segment 2, conscious residents

Segment 3, ease driven residents

Segment 4, environment minded residents

27–35 o 2500 Medium More public rental More E/F Many o 2yrs High Low Medium

4 35 o 2500 Medium More rental — Many 4 10yrs Medium Medium Medium

35–59 — 2500–3750 43750 Medium High More private More private More C/D More A/B 2–5 yrs and Many 410yrs Many 410yrs Low High Low High Low High

Medium Low

Medium Medium

Low Low

High High

Note: As respondents self-evaluated on a 0 (worst) to 5 (best) scales, here the level of low, medium and high indicates that the majority reported below 3, around 3 and above 3 respectively.

Especially, respondents in segment 1 score high on the current curtailment behavior comparing with other segments. This probably has to do with the fact that curtailment behavior of energy-saving involves the lowest money cost. They also scores high on the investment in HR glass, and it is most likely because there are a lot of financial supports currently implemented like the subsidy on investment in double glass (see: www.subsidie-dubbel-glas.nl). This segment shows the least initial interests in the pro-environmental interventions (reflected in the estimates of the ‘‘constant’’). They generally are not interested in any antecedent interventions, but highly sensitive to financial structural interventions and consequence interventions such as rewards and feedback. The main reasons of not save energy for this segment are the perceived expensive investment and the long payback time. Segment 2 was described as Conscious residents (43%, 129 respondents). These householders prefer comfort, but also take into account cost and environment. Respondents in segment 2 have a medium level of knowledge about the energy problems and do not experience high obstruction by governmental regulation and public opinion. They like all types of interventions. Furthermore, most respondents in this group are above 35 years old and live in a rental house. The main duration of current residence is long (i.e., most are more than 10 years). Respondents in segment 2 save energy not only for the purpose of saving money, but also for environmental concerns. They prefer to get well informed of the new technologies and new products. They are the group of residents that like to try out free products most. Financial interventions are highly preferred by this segment. Respondents belonging to segment 3 were ease driven residents (18%, 43 respondents). Residents in this segment act to enjoy comfort and amenities, and have less sense or interest in energy problem or environment. Most of respondents in this segment are between 35 and 59 years old with just above average monthly income. They have a low level knowledge of energy problems and experience a low degree of obstruction by governmental regulation and public opinion. There are many respondents in this segment that live in their current residents for about 2–5 years. The score for investment in HR glass in this segment is very low. Respondents in this segment show least motivation and involvement in energy problem. They like to know a bit about pro-environmental behavior, but they definitely dislike commitment. It is probably because they believe that making commitments will decrease their freedom in comfort seeking. They are the only group of residents that are not sensitive to financial interventions. It is likely that due to the relative high average age in this segment, these respondents show more of a habitual behavior and less willing to modify their behavior to save energy.

Segment 4 was described as environment minded residents (19%, 47 respondents). These residents act mainly from the viewpoint of environment. Respondents in segment 4 have a high level of knowledge about the energy problems and already invest sufficient (assessed by themselves) in energy-efficient products. They have on average the highest education level and monthly income among the four segments. About 70% of them live in private houses with a high energy label and the main reason for them to save energy is environmental concern. They show the high initial interests in the involvement of energy-saving behavior and moderate sensitivity regarding interventions. They are the only group that dislikes information interventions. This probably has to do with the fact that they consider themselves already well informed and expert in this topic.

7. Policy making recommendations Based on the survey results, it appeared that respondents currently invested only a small amount of money in the improvement for their dwellings in terms of energy efficiency. The amounts they do spend are mostly in HR glass. Moreover, the results of the respondents’ self evaluation also show that they do even worse in energy-savings by curtailment behavior. Therefore, it is interesting for the local government to address the focus on stimuli of curtailment behavior changes for saving energy, besides the current focus on the investment behavior of energy-saving. Our results indicated that it is best to distinguish four different segments of residents in the study area and use different combinations of interventions accordingly. Therefore, the local government, in this case study the municipality of Eindhoven, should apply the strategy that target these different segments with the interventions they prefer in order to achieve the target effect in stimulating them to save energy. We now interpret our empirical results to provide policy making recommendations. About 20% residents in the study area were in segment 1 that is cost focused. The respondents belonging to segment 1 are those who are young, earning lower income, living in rental house with energy label E/F. They are highly sensitive for rewards and financial (structural) interventions. Subsidies are a very good financial incentive to encourage residents in segment 1 to invest in the improvement of their houses in terms of energy efficiency. Providing feedback is another good intervention for this group, since it provides the consequence of their behavior (in relation to cost). A relatively large segment of the residents (43%) in the study area was in segment 2 that is conscious residents. The respondents

714

Q. Han et al. / Energy Policy 52 (2013) 706–715

belonging to segment 2 are those living in a rental house for a longer period. They are highly sensitive for all interventions especially regarding rewards. However, the short-lived effect of rewards should be taken into account when using this method. Providing free products is another good intervention for this group, since it influences both knowledge and motivation. In practice, these products are presented in a gift box with small test examples of various energy-saving behavior categories, and it services as an incentive for energy-saving behavior in general. These residents are aware of their own energy use but still welcome more knowledge about their opportunities to save more energy. These residents are highly sensitive for information. The best approaches to provide them information are through websites, TV and radio, and workshops. Information should be provided very clearly, unambiguously and tailored. Residents in segment 2 are also appealing for demonstration of energy-saving behavior. They like to visit a model house with all types of new measures and get a real life experience with these measures. Demonstration by neighbors and acquaintances also prompt them to follow up. Regarding the financial intervention, residents in segment 2 highly prefer the adjustment of removal tax on appliances. Apparently they consider this incentive as a good opportunity to invest in energy-efficient appliances. Financial supports, such as subsidy, loan with low interests, pricing policy are all useful interventions for this group. About 18% residents, who are in segment 3, are a bit aged, have an above average income and living in a private house for a longer period. They can be labeled ease residents as they value thermal comfort and resent deprivation. They like enjoying comfort and amenities. They cannot be urged with feedback and financial interventions, although they have a preference for information, demonstration and free products interventions. Residents in this segment generally do not bothered by financial interventions and have less interest in energy-saving behavior change, probably because they believe that it is too much effort with loss of amenities to save energy, especially they dislike make any commitments. Therefore, the best intervention strategy for this segment is to show them that the energy-saving can also be reached without loss of comfort or restricting freedom, especially emphasizing on how various actions can improve well-being, health and convenience. On average residents in segment 4 are those who have high income and high education level, and already live in private house with label A/B. They are more interested in environment problem. It seems as if they consider themselves expert in this filed as they dislike information interventions. They know a lot of information and examples about how they can save more energy, and already executed in their behavior. Contrast with the majorities that is more cost focused as in segment 2, this segment is a relative small group and likely to be the innovators and the early adopters in a S-shaped growth path associated with innovation diffusion. They mainly react to their own principles and the technological or other contextual innovations.

provided some evidence for the argumentation that antecedent interventions has less impact compare to consequence interventions, while structural (financial/legislation) interventions have the most effects in changing residents’ behavior (Steg and Vlek, 2009). This is an important finding from a policy perspective in the sense that policies aiming to reduce energy use may especially want to target high user of energy through these interventions, because of a higher energy-saving potential. Remarkable is that four different segments of residents were found in this research that is in demand for different interventions: cost focused residents, conscious residents, ease driven residents and environment minded residents. This finding is in line with early studies on energy use behavior of the residents (Groot et al., 2008). This contributes to the current literature that municipal intervention strategy should focus on specific target groups, and utilize the interventions accordingly that are appreciated and effective for each target group. The findings of studies such as this can inform local governments with information about the impact prospects of interventions for various segments of the local population, and provide the support of energy neutral development. The preferences of interventions can be useful in developing optimal and effective policy strategies. Adopting energy-saving behavior often simply requires that people change their minds (Doppelt, 2008). A right mixture of interventions could increase the level of engagement, empowerment and sense of efficacy of involved people. The level of success is foreseeable for such a strategy concerning behavior change interventions. As such, our study provide an case that is successful at mapping the diversity that exists across the local population and providing targeted and appreciated interventions that effectively build on those differences. Our study had some limitations that are worth noting. First, several aspects such as detailed technical characteristics of dwelling features and lifestyle of the residents which might affect energy-saving behavior were not included in the study. Second, the respondents’ rate is rather low and from only one local municipality. It would be interesting to investigate whether or not differences can be found between different municipalities. Municipalities in urban areas may need to focus their strategies differently comparing with municipalities in rural areas. Third, the actual effectiveness of different interventions has not been tested. It is better in the next step to conduct some cognitive interviews with experts to validate the findings. It is also advised to do the reweight sampling to map it with the local census data before deploy them in reality. In addition, future research could focus on develop new evaluation methods to measure behaviorrelated energy-savings as well as evaluate the contexts and circumstances in which these energy-saving behavior are most likely to persist over time.

References 8. Conclusions and discussions The aim of this study was to investigate how the local government could use interventions effectively to stimulate local residents to save energy. Specifically, the relationship between the segments and preferred interventions and between current energy behavior of residents and their characteristics are explored. Our results revealed that the estimates of feedback, rewards and financial interventions are most significant and positive for almost all four segments, while information, demonstration, free products and commitment are not. In particular, commitment intervention is valued by one segment with positive preference and by the other segment with absolute negative effect. It also

Ariely, D., 2009. Predictably Irrational: The Hidden Force That Shape Our Decisions. HarperCollins, New York. Abrahamse, W., 2007. Energy Saving Through Behavioral Change: Examining the Effectiveness of a Tailor-Made Approach. Thesis of State University Groningen, the Netherlands. Abrahamse, W., Steg, L., 2009. How do socio-demographic and psychological factors relate to households’ direct and indirect energy use and savings? Journal of Economic Psychology 30, 711–720. Abrahamse, W., Steg, L., Vlek, C., Rothengatter, T., 2005. A review of intervention studies aimed at household energy saving. Journal of Environmental Psychology 25, 273–291. Arkel, W. van, Jeeninga H., Menkveld, M., Ruig, G., 1999. Energieverbruik van gebouwgebonden energiefuncties in woningen en utiliteitsgebouwen. ECN-C99-084. Ben-Akiva, M., Boccara, B., 1995. Discrete choice models with latent choice sets. International Journal of Research in Marketing 12 (1), 9–24.

Q. Han et al. / Energy Policy 52 (2013) 706–715

Berkhout, P., Muskens, J., Velthuijsen, j., 2000. Defining the rebound effect. Energy Policy 28 (6-7), 425–432. Boxall, P.C., Adamowicz, W.L., 2002. Understanding heterogeneous preferences in random utility models: a latent class approach. Environmental and Resource Economics 23, 421–446. CBS. (2011). Last Retrieved on April 19, 2012 from Compendiumleefomgeving: /http://www.compendiumvoordeleefomgeving.nl/indicatoren/nl0036-Huishoude lijk-energieverbruik-per-inwoner.html?i=6-40S. Cialdini, R.B., 2005. Basic social influence is underestimated. Psychological Inquiry 16 (4), 158–161. Cialdini, R.B., 2007. The secret to using social norms to reduce household energy consumption. In: Behavior, Energy and Climate Change Conference, Topic Session 1B, November 8. Collins, L.M., Lanza, S.T., 2010. Latent Class and Latent Transition Analysis for the Social Behavioral and Health Sciences. Wiley, New York. Darby, S., 2006. The Effectiveness of Feedback on Energy Consumption: A Review for DEFRA of the Literature on Metering, Billing and Direct Displays. Environmental Change Institute, University of Oxford, Oxford, UK. Dietz, T., Stern, P., 2002. New Tools for Environmental Protection: Education, Information, and Voluntary Measures. National Academy Press, Washington, DC. Doppelt, B., 2008. The Power of Sustainable Thinking: How to Create a Positive Future for the Climate, the Planet, Your Organization and Your Life. Earthscan, London. Gardner, G.T., Stern, P., 2008. The short list: the most effective actions U.S. householders can take to curb climate change. Environment 50, 5. Geller, E., 2002. The challenge of increasing proenvironmental behavior. In: Bechtel, R., Churchman, A. (Eds.), Handbook of Environmental Psychology. Wiley, New York, pp. 525–540. Greene,W., 2001. Fixed and Random Effects in Nonlinear Models. Working Paper EC-01–01, Stern School of Business, Department of Economics, New York University. Greene, W., Hensher, D., 2002. A Latent Class Model for Discrete Choice Analysis: Contrast with Mixed Logit. Working Paper ITS-WP-02–08, Institute of Transport Studies, the University of Sydney, Australia. Groot, E.de, Speikman, M., Opstelten, I., 2008. Dutch Research into User Behavior in Relation to Energy Use of Residences. In: PLEA 2008–25th Conference on Passive and Low Energy Architecture, Dublin, 22–24 October 2008. Guerra Santin, O., 2011. Behavioural patterns and user profiles related to energy consumption for heating. Energy and Buildings 43 (10), 2662–2672. Guerra Santin, O., Itard, L., 2010a. Occupants’ behaviour: determinants and effects on residential heating consumption. Building Research & Information 38 (3), 318–338. Guerra Santin, O., Itard, L., 2010b. Effects of energy performance regulations on the heating energy use in Dutch dwellings. Rehva Journal 47 (6), 32–38. Guerra Santin, O., Itard, L., Visscher, H., 2009. The effect of occupancy and building characteristics on energy use for space and water heating in Dutch residential stock. Energy and Buildings 41, 1223–1232.

715

Guns, B., 2007. People really are different: leveraging segmentation to accelerate climate action. In: Behavior, Energy and Climate Change Conference, Topic Session 2B, November 8. IEA, 2008. World Energy Outlook. International Energy Agency (IEA), France. Itard, L., Meijer, A., Guerra Santin, O., 2009. Consumenten onderzoek lenteakkoord. Rapport in opdracht van NVB, Research Institute OTB, Delft, the Netherlands. Kemperman, A., Timmermans, H., 2006. Preferences, benefits, and park visits: a latent class segmentation analysis. Tourism Analysis 11 (4), 221–230. Laitner, J.A., Ehrhardt-Martinez, K., McKinney, V., 2009. Examining the scale of the behavior energy efficiency continuum. In: Proceedings of the 2009 ECEE Summer Study, La colle sur Loup, Cote d’Azur, France. Leidelmeijer, K., Cozijnsen, E., 2010. Energiegedrag in de woning: aanknopingspunten voor de vermindering van het energiegebruik in de woningvoorraad. RIGO Research, Amsterdam, the Netherlands. Lucas, K., Brooks, M., Darnton, A., Jones, J., 2008. Promoting pro-environmental behavior: existing evidence and policy implications. Environmental Science & Policy 11 (5), 456–466. Lutzenliser, L., 1993. Social and behavioral aspects of energy use. Annual Review of Energy and the Environment 18, 247–289. Van de Maele-Vaernewijck, M.C.L., van Raaij, W.F., Verhallen, Th.M.M., 1980. Energiegrdrag in de woning: literatuuroverzicht en gedragsmodel, i.o.v. Ministerie van VROM. Olander, F., Thøgersen, J., 1995. Understanding of consumer behavior as a prerequisite for environmental protection. Journal of Consumer Policy 18, 345–385. Poortinga, W., 2003. Household preferences for energy-saving measures: a conjoint analysis. Journal of Economic Psychology 24 (1), 49–64. Steg, L., 2008. Promoting household energy saving. Energy Policy 36, 4449–4453. Steg, L., Vlek, C., 2009. Encouraging pro-environmental behavior: an integrative review and research agenda. Journal of Environmental Psychology 29, 309–317. Stern, P.C., 2000. Toward a coherent theory of environmentally significant behavior. Journal of Cunsumer Policy 22, 461–478. Stern, P.C., Aronson, E., 1984. Energy Use: The Human Dimension. W.H. Freeman and Company, New York, NY. Swait, J., 1994. A structural equation model of latent segmentation and product choice for cross-sectional revealed preference data. Journal of Retailing and Consumer Services 1 (2), 77–89. Swait, J., Adamowicz, W., 2001. The influence of task complexity on consumer choice: a latent class model of decision strategy switching. Journal of Consumer Research 28, 135–148. Tiedemann, K., (2009) Residential behavioral energy savings: a meta-regression analysis. In: Processdings of the International Energy Program Evaluation Conference, Portland, OR, USA. Vereniging Eigen Huis, 2009. Eigen Huis Panel. VROM—Energie. Vringer, K., Blok, K., 1995. The direct and indirect energy requirement of householders in the Netherlands. Energy Policy 23, 10.