Electricity saving in households—A social cognitive approach

Energy Policy 38 (2010) 7732–7743

Contents lists available at ScienceDirect

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

Electricity saving in households—A social cognitive approach$ John Thøgersen n, Alice Grønhøj 1 Aarhus University, School of Business and Social Sciences, Department of Marketing, Haslegaardsvej 10, DK-8210 Aarhus, Denmark

a r t i c l e in f o

a b s t r a c t

Article history: Received 13 January 2010 Accepted 17 August 2010 Available online 17 September 2010

We propose a conceptual framework for understanding the (lack of) energy saving efforts of private households based on Bandura’s (1986) social cognitive theory. Results from applying this framework on a sample of Danish private electricity consumers are presented and it is concluded (a) that households’ electricity consumption depends on both structural and motivational factors, (b) that their electricity saving effort depends on the strength of their internalized norms or self-expectations and on selfefficacy related factors, and (c) that there are predictable patterns of interaction among household members that influence their electricity consumption. The results suggest two approaches to promote electricity saving in households: (1) to change the socio-structural environment to be more facilitating for energy saving and empower householders to be more effective in their striving towards this goal through improved feedback about their household’s electricity consumption and (2) social norms marketing, communicating social expectations and others’ successful electricity saving achievements. & 2010 Elsevier Ltd. All rights reserved.

Keywords: Electricity saving Social cognitive theory Dyadic data analysis

1. Introduction In the quest for lowering greenhouse gas emissions and our reliance on fossil fuels, increasing attention is paid to the waste of energy in buildings, including private homes (Dietz et al., 2009; Ekins and Skea, 2009). It is well documented that the electricity consumption in private households could be lowered substantially if they paid more attention to unnecessary use of electricity and if this attention was followed by a change in everyday behaviors (e.g., Gram-Hanssen et al., 2004). Households can reduce their electricity consumption immediately, for example, by being more careful to switch off unnecessary light and electrical devices, by not keeping unused devices on standby, or by following some of the many other pieces of advice that can be found on dedicated websites as well as in many printed publications (e.g., Amann et al., 2007; Clift and Cuthbert, 2007). Private households that pay attention to their everyday electricity consumption might also attend more carefully and in a more timely fashion to the maintenance of their electrical appliances (e.g., defrosting the freezer) and replace devices that use too much energy earlier and more consistently (e.g., switch to energy saving light bulbs and replace fridges, freezers, etc., in a more timely manner).

As input to the development of effective policy to reduce the squandering of electricity in private households, it is the objective of this study to identify the most prominent drivers and impediments for saving electricity in this setting. Identifying the important drivers and impediments is the first step in designing effective intervention programs for reducing home energy consumption (Uitdenbogerd et al., 2007). As we will explain in the following, this implies identifying psychological, social, and structural antecedents of taking action to save electricity. We also address the effectiveness of private consumers’ efforts to save electricity in terms of its impact on the household’s electricity consumption. Based on a review of previous research, a coherent theoretical framework is developed that specifies how psychological, social and socio-structural factors are linked to individual electricity saving behavior and to electricity consumption, directly or indirectly, as drivers or impediments. In the empirical part of the paper, the fit of the proposed model to a combination of individual level survey data and meter reading data is investigated. Finally, implications for the promotion of electricity saving in households are discussed.

2. Previous research $ The research was supported by a grant from ELFOR, project number (Dansk Energi—Net): 338-020. We are grateful for helpful suggestions from three anonymous reviewers. n Corresponding author. Tel.: +45 8948 6440; fax: + 45 8615 3988. E-mail addresses: [email protected] (J. Thøgersen), [email protected] (A. Grønhøj). 1 Tel.: + 45 8948 6471.

0301-4215/$ - see front matter & 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.enpol.2010.08.025

Behavioral research points at a number of reasons why consumers might not adhere carefully to the many useful pieces of electricity saving advice. These include the convenience of doing what one is used to do, lack of motivation, and a whole range of impediments which make behavior change difficult

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

(e.g., Ritchie and McDougall, 1985). By way of organizing the large, diverse, and heterogeneous set of antecedents, a basic distinction is often made between psychological and contextual variables (Wilson and Dowlatabadi, 2007). A wide range of psychological antecedents of home energy use and conservation behavior has been identified, including value priorities, outcome expectations, attitudes, personal norms, and self-efficacy (Ketola, 2000). According to recent research, home energy consumption is unrelated to general value priorities, but related (in the expected way) to more specific attitudes towards saving energy (Vringer et al., 2007). Attitudes towards saving energy depend on individuals’ expectations about positive and negative outcomes of doing so (e.g., Black et al., 1985; Olsen, 1981). Some consumers may feel that the effort that is needed in order to save electricity is out of proportion to the potential private benefits. A large share of the population is also concerned about the environmental impacts of the high and increasing energy consumption (e.g., Bang et al., 2000; Poortinga et al., 2004, 2003), but this concern alone often seems insufficient to motivate action (e.g., Jensen, 2005; Pedersen and Broegaard, 1997). Intentions to save energy also depend on the individual’s social expectations (or subjective social norms) and self-expectations (or personal norms) (e.g., Black et al., 1985; Midden and Ritsema, 1983) as well as self-efficacy or perceived behavioral control (e.g., Harland et al., 1999). Self-efficacy is an issue because it is often difficult for an individual to perceive any effects of his or her effort to save electricity (Grimmig, 1992) or any relationship between their behavior and the household’s electricity consumption (GramHanssen et al., 2004). Further, in the bigger picture an individual’s contribution may seem like a drop in the sea (cf. Berger and Corbin, 1992; Ellen et al., 1991). The contextual conditions that may influence these behaviors are possibly an even more complex and heterogeneous category than the psychological variables, including as diverse elements as physical–structural conditions (e.g., home size, technologies, standards, and the format and frequency of information about the household’s energy consumption), socio-demographic characteristics (household size and composition) as well as cultural and economic aspects (social norms and economic incentives). Contextual conditions influence home energy use and individual conservation behavior in a multitude of ways, including through the formation of outcome expectations (Black et al., 1985) and selfefficacy (Gist and Mitchell, 1992), but also directly by constraining available choice options and determining their attractiveness. For example, it has been consistently found that the household’s electricity consumption increases with the number of household members, household income, and the size of the home (Petersen and Gram-Hanssen, 2005). It is perhaps less obvious that also the age-composition of the household members contributes to the variation in overall household electricity consumption. In particular, it has been found that households with teenagers use significantly more electricity than other households after controlling for household size (Petersen and Gram-Hanssen, 2005). A less visible, but more pervasive kind of contextual condition is cultural understandings, which have been shown to have a profound influence on our perceptions of what is ’’right or wrong’’ and ’’necessary or unnecessary’’ consumption (e.g., Wilhite et al., 2001). Attitudinal and contextual variables seem to have their main influence in different stages of the decision-making process. A review of US based studies found that attitudes are good predictors of intentions to change residential energy use behavior, but structural characteristics (of the residence) are better predictors of specific actions, such as weatherization (Guerin et al., 2000). In the next section, we synthesize these findings in a social-cognitive model of electricity saving behavior and electricity consumption.

7733

3. A social cognitive model of electricity saving behavior and electricity consumption The available evidence suggests that impeding contextual conditions and, related, low self-efficacy are some of the important reasons for the squandering of electricity in private households. Self-efficacy and (real and perceived) sociostructural conditions are key constructs in Bandura’s (1986) social cognitive theory, which makes this theory a potentially useful frame of reference for the present study. It is also a comprehensive and therefore complex theory, which makes it difficult to operationalize. Hence, like most other empirical studies based on this theory (e.g., Bandura, 2004), we employ a simplified version, adapted to the specific behavioral problem at hand. The core construct in Bandura’s social cognitive theory is selfefficacy, which he defines as the person’s confidence in performing a particular behavior. Bandura argues that behavior change is made possible by a personal sense of control. If people believe they can take action to solve a problem instrumentally, that is, if they have a sense of self-efficacy, they become more inclined to do so and feel more committed to the action. Although behavioral research is usually most interested in determining how individual behavior is contingent on personal and environmental factors, Bandura (1986) emphasizes the reciprocal determinism between personal factors, environment (defined as factors physically external to the person), and behavior. For example, a person’s sense of self-efficacy (a personal factor) is assumed to be an important antecedent of behavior and at the same time to be partly derived from his or her past behavioral experiences. This reciprocal determinism is illustrated by the dotted feedback arrows in the otherwise recursive representation of the core elements of the theory in Fig. 1. What exactly the individual learns from past behavioral experiences depends on the reinforcements in the situations (defined as the responses to a person’s behavior that increase or decrease the likelihood of reoccurrence). In addition to selfefficacy, the individual’s expectations (the anticipated outcomes of a behavior) and the values that the person places on given outcomes are assumed to influence a person’s goals and behavior. Another key proposition in Bandura’s theory is that individuals do not only learn from their own personal experiences, but also by watching others’ behavior and the outcomes of others’ behavior (i.e., observational learning). Such ‘‘vicarious experiences’’ are an

Fig. 1. A social cognitive model of behavior and learning. . Source: adapted from Bandura, 1986

7734

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

important source of self-efficacy as well as of outcome expectations. People are constantly exposed to other people’s behavior, both in everyday life and in documentaries and fiction in the media. However, with regard to electricity consumption and saving in the household, by far the largest amount of observational learning, and other forms of social influence, both for children and for adults living together, goes on between the members of the household. Hence, in research such as this it is especially important to capture the interactions and interdependencies between household members. As indicated in Fig. 1, Bandura (1986) assumes that most, but not necessarily all, of the influence of psychological factors on behavior are mediated through behavioral goals or intentions. Individual’s behavioral goals are assumed to be primarily determined by his or her outcome expectations and felt selfefficacy, but perceptions regarding others’ behavior and regarding facilitators and impediments in the sociostructural environment may account for additional variance. Whether or not a behavior is actually performed depends primarily on the person’s behavioral goals and on possible facilitators and impediments, but one or more of the included psychological antecedents may account for additional variance in behavior. To the extent that the individual’s goal striving is successful, which partly depends on sociostructural factors, the behavior will lead to anticipated outcomes. As already mentioned, the relationship between the mentioned antecedents and behavior is not a one-way street. Over time, individuals learn from the produced behavioral outcomes, which may then lead to adjustments of outcome expectations, perceptions regarding facilitators and impediments, and/or perceived self-efficacy. It follows from this theoretical model that behavior change can be achieved through making the environment more facilitating for the desired behavior (improved facilitators and/or reduced impediments, cf. Thøgersen, 2002), through increasing individuals’ selfefficacy (’’empowerment,’’ cf. Thøgersen, 2005), or through changing expectations related to the desired behavior in a more favorable direction (e.g., Ajzen and Fishbein, 1980). In countries, such as Denmark, where most people are already convinced about the benefits of energy conservation (e.g., Genius Access A/S, 2008), there should be more to gain from improving self-efficacy and (real and perceived) sociostructural factors than from improving (physical) outcome expectations. In the following, we apply this social cognitive model to electricity saving in private households in Denmark. Based on a combination of survey data and meter readings, we will investigate the veracity of the model’s assumptions regarding the relationships between behavioral goals, behavior, and behavioral outcomes, as well as its assumptions regarding these variables’ relationship to self-efficacy, outcome expectations, and perceived facilitators and impediments.

4. Method In April 2007, residents of four villages in the vicinity of Holstebro in Western Jutland, Denmark, in the geographical region traditionally serviced by the Danish electricity supplier NOE, were contacted by telephone and asked to participate in a web-survey pending on the following screening criteria: That they lived in a single-family house, that the household consisted of at least two residents, the oldest of which was less than 70 years old, and that they were willing and able to participate in a follow-up experiment after the survey (not reported in this article).2 NOE supplied a database with households 2 We focused on single-family houses primarily because 80% of the population in Holstebro municipality lives in single-family houses and the percentage is substantially higher in rural villages, such as the ones we sample from. (The national average is 69%, both numbers according to Danmarks Statistik, see http://

in the four villages, containing a total of 1525 names and addresses. Some of these were doublets or no telephone number could be identified. Contact was attempted with 1158 households. After at least ten attempts, 353 had to be discarded because no contact was established and in 13 cases the person did no longer live on the registered address. Of the contacted and eligible households, 188 refused to participate and 292 did not fulfill the screening criteria, which left 312 households who fulfilled the criteria and agreed to participate in the study. Since the household’s electricity consumption is the outcome of all household members’ activities and interactions, we attempted to obtain data from all adult members of participating households. In about half of the consenting households one and in the other half two adults agreed to fill out the web-survey and they received a link to the survey by e-mail. The link was mailed to 462 individuals and after reminders by e-mail and phone 320 individuals from 237 households had completed the survey for a response rate of 70%. A professional market research company, Jysk Analyseinstitut, took care of participant selection and data collection (except for electricity consumption data). 4.1. Measures Electricity consumption: The household’s electricity consumption in the year 2007 was read from the meter by NOE, the households’ electricity supplier, and reported to us in May 2008. Exploratory curve-fitting analyses suggested a non-linear relationship between electricity consumption and behavior and that a better approximation of linearity could be achieved through a logarithmic transformation. Hence, in the following statistical analyses electricity consumption is represented by its natural logarithm. Behavior: There are many ways to save electricity in the home and they differ both in terms of frequency and impact. This heterogeneity is captured by means of a battery of 17 behavior items (see Table 1), drawn from available guides regarding how to save electricity in the home. The battery includes purchasing behaviors (energy saving light bulbs), control behaviors (controlling the temperature in fridges and freezers), and the careful use of equipment (switch off light, lower the temperature when laundering, etc.). The behavior items were questions of the type ‘‘How often do you X,’’ where X refers to each of the 17 behaviors, and using a 5-point scale. For 15 behavior items, we used the labels ‘‘never,’’ ‘‘rarely,’’ ‘‘half the time,’’ ‘‘often,’’ and ‘‘always/every time.’’ For the last two items – controlling the temperature in the fridge and the freezer – we used the labels ‘‘never,’’ ‘‘rarely,’’ ‘‘regularly,’’ ‘‘often,’’ and ‘‘very often (i.e., every week).’’ Respondents were also offered a ‘‘not relevant’’ response option for each of the behavior items. In terms of measurement theory, electricity saving behavior is conceived as a formative construct (Jarvis et al., 2003) and operationalized as a behavior index, calculated as the mean of answered (‘‘relevant’’) behavior items. Due to the way it is constructed, this measure of electricity saving in the household has high face validity. The significant relationship with the household’s electricity consumption, to be reported later, further supports its construct validity. This operationalization of electricity saving in the household is similar to Kaiser’s (Kaiser, 1998; Kaiser and Wilson, 2000) Rasch scaling of environmentally responsible behavior in that it is based on a (relatively) large number of behavior items that vary in their (footnote continued) www.statistikbanken.dk.) Hence, in a random sample of the size employed here there would be too few apartments to allow meaningful analyses of differences between apartments and single-family homes. Another reason for concentrating on single-family houses is that these households use on average nearly twice as much electricity as those living in apartments (Petersen and Gram-Hanssen, 2005), meaning that the saving potential is bigger in the former than the latter.

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

7735

Table 1 Electricity consumption and behavior item means for male and female participants and within-couples correlations. Men

Ln (Elec. cons.) Behavior (index) How often do you y Switch off all light when leaving a room as last person? Buy energy saving light bulbs? Wash at 60 1C instead of 90 1C? Start the washing machine when it is not completely full? (rev) Switch off the TV when nobody watch it? Switch off the computer when it is not used? Switch off standby when electric devices are not used? Defreeze frozen food in the fridge? Use only minimal water in the pot when boiling egg, vegetables and the like? Put a lid on the pot when boiling egg, vegetables and the like? Put oven dishes in a cold rather than a preheated oven? Start the dishwasher when it is not completely full? (rev) Use the energy saving program on the dishwasher? Use the tumble dryer in the summer period? (rev) Fill coffee from the coffee machine on a thermos? Control the temperature in the fridge? Control the temperature in the freezer?

Women

N

Mean

237 162

8.50 3.41

160 162 135 145 162 162 160 150 151 155 155 153 148 141 148 158 160

N

Mean

.34 .43

237 158

8.50 3.52

.34 .39

83

4.15 2.54 3.82 3.83 3.91 3.83 2.36 3.21 2.95

.75 1.13 1.26 .76 .94 1.15 1.34 1.08 1.29

158 156 157 156 158 153 158 158 158

4.05 2.63 4.02 3.97 3.89 3.92 2.42 3.22 3.11

.85 1.08 1.28 .65 1.01 1.15 1.25 1.09 1.26

4.39 2.57 4.04 3.49 3.13 3.84 2.44 2.27

.86 1.11 .83 1.44 1.06 1.36 1.19 1.09

158 158 154 148 154 140 155 157

4.43 2.82 4.35 3.53 3.31 4.07 2.59 2.32

.88 1.26 .55 1.36 1.16 1.28 1.18 1.14

demand on individual abilities and resources. The similarity of the behavior index to a Rasch scale is reflected in a marginally acceptable separation reliability coefficient of .59. According to the Rasch scaling philosophy, the higher a person’s ability and effort relative to the difficulty of a behavior item, the higher the probability that he or she performs that behavior. Very easy behaviors are likely to be performed by many people. As they become more demanding, fewer individuals are likely to perform the behaviors, reflecting the variation in individual motivation and resources.3 In Rasch scaling terms, the greater the total score on the behavior index, the greater the individual’s attainment level with regard to electricity saving behavior. Behavioral goals/intentions: The individual’s goal or intention regarding electricity saving was measured with two items: (1) ‘‘In the next month, I intend to do everything I can to keep the electricity consumption as low as possible,’’ registered on a 7-point scale from 1¼‘‘is not true at all’’ to 7¼‘‘is absolutely true.’’ (2) ‘‘Describe your personal goal regarding the household’s electricity consumption for the next month,’’ on a 7-point scale from 1¼‘‘I will do nothing at all to keep the electricity consumption as low as possible’’ to 7¼‘‘I will do everything I can to keep the electricity consumption as low as possible.’’ The construct has an acceptable composite reliability4 (Goalm ¼.73, Goalw ¼.76; subscripts refer to separate calculations for men (m) and women (w) in the sample). Outcome expectations: We distinguish empirically between four types of outcome expectations: expected positive and negative (physical) outcomes of efforts to save electricity, expected social outcomes (operationalized as perceived social pressure or subjective norms), and expected self-evaluative outcomes (or personal 3 Since the relationship between electricity saving behavior and electricity consumption is negative, the finding that their relationship seems to be logarithmic suggests that there is a decreasing marginal effect of increasing the number and intensity of electricity saving behaviors. The possible implications of this finding will be discussed in the concluding section of the article. 4 Composite reliability or Raykov’s reliability rho (r), tests if it may be assumed that a single common factor underlies a set of variables. This is one among a number of measures of reliability as internal consistency commonly used in psychological research. For an overview of various ways of measuring construct reliability, see http://faculty.chass.ncsu.edu/garson/PA765/reliab.htm.

Std. deviation

Couples Std. deviation

N

r

Sig.

t

Sig.

.57

.000

 3.58

.001

83 82 64 71 83 78 81 74 76

 .01 .64 .55 .09 .09 .64 .74 .41 .45

.910 .000 .000 .449 .411 .000 .000 .000 .000

.10  1.02  .80  1.26 .84  2.54  .38  .83  1.13

.919 .313 .428 .214 .401 .013 .708 .408 .262

77 77 78 76 68 67 81 82

.32 .37  .03 .40 .60 .46 .27 .24

.005 .001 .821 .000 .000 .000 .014 .028

 .81  3.30  2.37  .73  .91  .90  1.97  .17

.418 .001 .020 .469 .366 .369 .052 .866

norms). With a single exception, mentioned below, all answers were given on a 7-point Likert scale from ‘‘completely disagree¼1’’ to ‘‘completely agree¼7.’’ Expected positive outcomes were measured with the items: (1) ‘‘By saving electricity I contribute to avoiding global warming’’ and (2) ‘‘By avoiding unnecessary electricity consumption I save quite a bit of money on the electricity bill.’’ Expected negative outcomes were measured with the items: (1) ‘‘It takes a lot of effort to all the time carefully switch off light and equipment that are not used’’ and (2) ‘‘If I all the time switch off light and equipment that are not used they do not last so long.’’ Expected social outcomes were measured with the items: (1) ‘‘Most of my acquaintances expect from me that I save electricity in my home’’ and (2) ‘‘Most people who are important to me think that I y (‘‘should not¼ 1’’ to ‘‘should¼7’’) y make an effort to save electricity.’’ Expected self-evaluative outcomes were measured with the items: (1) ‘‘It makes me feel like a better person to save electricity in my home,’’ (2) ‘‘I feel bad about using more electricity than necessary in my home,’’ and (3) ‘‘I get a bad conscience if I waste electricity in my home.’’ Most of the constructs have acceptable composite reliabilities, although especially those for social outcome expectations leave something to be desired (Positivem ¼.61, Negativem ¼.59, Socialm ¼.56, Selfevaluativem ¼.78, Positivew ¼.71, Negativew ¼.55, Socialw ¼.45, Self-evaluativew ¼.80). Self-efficacy: We used the format and procedure recommended by Ajzen (2002) to measure self-efficacy regarding electricity saving. From the outset, we formulated nine self-efficacy items, which were included in the questionnaire. Factor and item analysis were used to derive a unidimensional instrument with acceptable composite reliability from this item pool. The final instrument contained four items: (1) ‘‘I believe that I’m able to avoid all unnecessary electricity consumption in my home,’’ registered on a 7-point scale from 1¼‘‘most certainly not’’ to 7¼‘‘most certainly.’’ (2) ‘‘To which extent do you believe that you are able to limit electricity consumption in your household to the absolute necessary?’’ registered on a 7-point scale from 1 ¼‘‘is not at all able to do it’’ to 7¼‘‘is able to do it to a high extent.’’ (3) ‘‘How certain are you that you are able to avoid all unnecessary electricity consumption in your home,’’ registered on a 7-point scale from 1¼‘‘very uncertain’’ to 7 ¼‘‘very certain.’’ (4) ‘‘How

7736

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

much control do you feel that you personally have over how much electricity is consumed in your household?’’ registered on a 7-point scale from 1 ¼‘‘very little control’’ to 7¼‘‘complete control.’’ The construct has an acceptable composite reliability (Self-efficacym ¼.73, Self-efficacyw ¼.76). Perceptions of other household members’ behavior: The individual’s perception of other household members’ behavior was measured with two items: (1) ‘‘The other members of my household do what they can to save electricity.’’ (2) ‘‘The other members of my household are careful saving electricity.’’ Answers were given on a 7-point Likert scale from ‘‘totally disagree ¼1’’ to ‘‘totally agree¼7.’’ The construct has an acceptable composite reliability (Others’ behaviorm ¼ .82, Others’ behaviorw ¼.88). Perceived impediments/facilitators: Because of the heterogeneity of relevant sociostructural conditions, we decided to operationalize these in two ways: (a) as participants’ self-reports of the most obviously relevant structural characteristics of the family and the home (see next paragraph), and (b) as their overall assessment of how impeding or facilitating these conditions are for avoiding unnecessary electricity consumption in their home. We measured the latter with two items: (1) ‘‘As the conditions are today, it is difficult to avoid unnecessary electricity consumption in the home,’’ and (2) ‘‘It will demand a big extra effort of me to avoid unnecessary electricity consumption in my home.’’ Answers were given on a 7-point Likert scale from ‘‘completely disagree¼1’’ to ‘‘completely agree¼7.’’ The construct has an acceptable composite reliability (Impedimentsm ¼.60, Impedimentsw ¼ .70). Structural conditions: Four structural characteristics that have been found to be related to the household’s electricity consumption were included in the analyses: home size, household size, household income, and the number of teenagers in the household.

5.1. Descriptive analyses Item means and dyadic correlations are shown in Tables 1 and 2. Because of our screening criteria, the average household is rather large (3.5 members) and, among others, contains more than one ‘‘teenager.’’ Reflecting this, the average home size is over 150 m2. According to Table 1, the participants make quite an effort to save electricity, the average score for 17 activities being about 3.5 on a 5-point scale. Consistent with this, average electricity saving intentions are above the scales midpoint (i.e., around 5 on a 7-point scale) and so are self-efficacy, positive outcome expectations and self-evaluative outcome expectations, and negative outcome expectations are mostly below the scale’s midpoint. Less favorable for electricity saving are perceived impediments around the scale’s midpoint and social outcome expectations as well as perceptions regarding other household members’ electricity saving below the scale’s midpoint. The very high positive outcome expectations and the more moderate scores for self-efficacy and related constructs confirm the assumption that there is more to gain from improving the latter than the former in the analyzed context. The correlations and (paired-samples) t-tests reported in Tables 1 and 2 are limited to the couples in the sample. Few of the item means differ significantly between partners, but women report stronger conservation goals (only one of the two items) than men and conservation behavior seems to be more important for women’s than for men’s self-evaluation (two out of three items). Women also report more conservation behaviors than men, mostly with regard to traditional women’s tasks, but they also more often switch off the computer when not in use. Inter-couple correlations are strong for the behavior index, medium for perceptions of other household members’ behavior, and weak for self-efficacy and other self-related evaluations, but also for some behavior items (switching off the light and starting the dishwasher when not full).

5. Analyses and results The data analysis and hypothesis test were done by means of structural equation modeling (SEM) using AMOS 16 (Arbuckle and Wothke, 1999) and for the most parts as dyadic data analysis (Kenny et al., 2006). Reflecting that private electricity consumption is registered at the household level, the unit of analysis is the household (although we limited the data collection to adults in the household) and, because the traditional division of labor by gender in the household makes this a relevant distinguishing characteristic, individuals within a household (i.e., members of dyads/couples) were identified by sex. In SEM, the measurement model is a confirmatory factor analysis (CFA) model and the theoretical constructs are latent factors extracted from the manifest variables (Bagozzi, 1994; Bollen, 1989). In the analyses reported below, the usual assumptions about a simple structure factor pattern in the measurement model and uncorrelated item error terms were applied, except that errors of the same item reported by different persons from the same household were allowed to correlate. As is commonly the case in behavioral research, there are instances of item non-response in the analyzed data set. Especially, since our study focuses on the household level, households with only one respondent ‘‘miss’’ all the values from the potential other respondent. AMOS is one of the first applications that offered Full Information Maximum Likelihood (FIML) to deal with item nonresponse. Research into ways of dealing with missing values suggests that currently FIML is the most effective method to deal with missingness due to item non-response, not only because it minimizes the loss of information and, hence, statistical power, but also because it leads to the most unbiased parameter estimates (Arbuckle, 1996). Furthermore, FIML is ideal for dyadic analysis where some ‘‘dyads’’ are represented by only one person.

5.2. The structural model The model in Fig. 1 is an individual level model. For the present purpose, a household was represented by up to two individuals so the household level model consisted of two identical gender-specific individual level models, operationalized as described above, but with controls for interdependencies between the members of a dyad. This household model was fitted to the survey and meter data. In order to reduce the complexity of the statistical analyses, we split the model into three parts and, hence, fitted three partial models to the data: (1) One for predicting electricity consumption, (2) one for predicting conservation behavior, and (3) one for predicting conservation goals or intentions. As the basic rule, for each of the three models, possible antecedents according to Fig. 1 were dropped from the model if their effect was completely mediated through other included constructs. For example, the behavioral impact of self-efficacy is completely mediated through behavioral goals. Hence, self-efficacy is dropped from the model predicting behavior in spite of the possible direct path from selfefficacy to behavior indicated in Fig. 1. Exceptions from this general rule are income and family size, which are included as control variables in the model predicting electricity consumption. In model number 1, household electricity consumption is modeled as a function of household characteristics (home size, household size, number of teenagers in the household, and household income) and household members’ conservation behavior.5 Because of the strong positive correlation between household members’ electricity conservation behavior (cf. Table 1), 5 It was also controlled whether perceived impediments and other possible antecedents further back in the presumed causal chain had a significant direct

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

7737

Table 2 Item means for male and female participants and within-couples correlations. Men

Goal 1 Goal 2 Self-eff. 1 Self-eff. 2 Self-eff. 3 Self-eff. 4 Perc. imp. 1 Perc. imp. 2 Neg OE 1 Neg OE 2 Pos OE 1 Pos OE 2 Soc. OE 1 Soc. OE 2 Self-ev. OE 1 Self-ev. OE 2 Self-ev. OE 3 Househ. mem. beh. 1 Househ. mem. beh. 2 Income Home sizea Househ. size Teenagers a

Women

Within couples

N

Mean

Std. deviation

N

Mean

Std. deviation

N

r

Sig.

t

Sig.

162 162 162 162 162 162 162 162 162 162 162 162 162 162 162 162 162 162 162 232 237 157 237

4.43 4.99 4.67 5.11 4.39 4.45 4.14 4.32 4.12 2.94 5.99 5.99 3.52 3.75 4.48 4.77 4.30 3.44 3.40 4.19 2.57 3.51 1.44

1.50 1.43 1.61 1.28 1.69 1.68 1.77 1.86 1.86 1.72 1.40 1.41 1.48 1.08 1.54 1.64 1.65 1.70 1.52 .83 .71 1.08 .69

158 158 158 158 158 158 158 158 158 158 158 158 158 158 158 158 158 158 158 232 237 157 237

4.53 5.23 4.71 4.99 4.49 4.46 4.00 4.01 3.58 3.21 6.30 5.99 3.52 3.73 4.94 4.97 4.39 3.39 3.47 4.19 2.57 3.51 1.44

1.56 1.21 1.49 1.18 1.38 1.43 1.60 1.69 1.92 1.84 1.21 1.35 1.57 1.16 1.40 1.43 1.56 1.70 1.57 .83 .71 1.08 .69

83 83 83 83 83 83 83 83 83 83 83 83 83 83 83 83 83 83 83

.12 .24 .03 .15 .22 .01 .14 .31 .25 .19 .27 .22 .27 .21 .07 .07 .17 .36 .35

.280 .028 .777 .168 .043 .957 .213 .005 .025 .085 .014 .051 .014 .062 .555 .519 .119 .001 .001

 1.327  2.215  .212 .496  .174 .894  .104 1.398 .940  1.029  1.485 .482  1.486  .869  2.814  1.855  1.215  .363  1.313

.188 .030 .833 .621 .863 .374 .917 .166 .350 .307 .141 .631 .141 .387 .006 .067 .228 .717 .193

Four categories: 1¼ under 100 m2, 2¼ 100–150 m2, 3¼ 150–200 m2, and 4¼ over 200 m2.

behavior is modeled as a latent construct at the household level for this analysis, using the individual-level behavior indices as manifest variables. Hence, for this analysis all constructs are conceptualized as household level, between dyad variables. In accordance with standard practice, all exogenous variables are allowed to correlate (in this and the following analyses). 5.3. Electricity consumption The standardized structural parameters of the model predicting household electricity consumption are reported in Fig. 2. The fit indices, reported in the note of Fig. 2, show an excellent fit of the model to the data. However, the included antecedents only explain a modest 25% of the variance in electricity consumption. Unexplained variance can be attributed to methodological shortcomings, notably electricity consumption being measured for a whole year while antecedents were measured at a specific point in time and inevitable measurement imperfections with regard to both electricity consumption and its antecedents. Another likely source is omitted variables, such as the quality of the house and the quantity, types and quality of electricity consuming equipment in the house. The measurement error for electricity consumption is probably substantially lower than for psychological variables. Also, there is no reason to suspect that electricity saving behavior changed a lot during the year of the study. Hence, it seems likely that omitted external factors account for a larger share of unexplained variance than measurement issues. Fig. 2 shows that a household’s electricity consumption increases with the size of the home. This is as expected and ads face validity to the results. It also illustrates the importance of structural factors for electricity consumption. Home size and other structural factors limit the electricity savings that can be achieved through changes in everyday behavior. Home size is (footnote continued) effect on electricity consumption after controlling for these variables. None of them had and, hence, they were omitted from this analysis.

Fig. 2. From conservation behavior to electricity consumption. Note: Only the structural model. The measurement model for this and the following analyses can be acquired from the authors. w2 (4) ¼ 2.174, p ¼.704, root mean square error of approximation (RMSEA)¼ .000 (90% confidence interval: .000–.074). A dotted line means that the relationship is not significant at the 5% level. CR (behavior) ¼.75.

measured in square meters, but respondents did not report the exact size of the home; only to which of four broad categories their home belongs. For this reason, the estimated regression weight most likely underestimates the true relationship between home size and electricity consumption. Household size and income are positively correlated with home size and they have no direct impact on electricity consumption when home size is controlled. The negative correlation between income and electricity saving behavior shows that low-income households tend to do more to save electricity than high-income households. Not all household members are equal in terms of electricity

7738

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

consumption, which is reflected in the number of ‘‘teenagers’’ (actually, individuals in the age group 14–20 years old) in the household further boosting the electricity consumption when the household size is controlled. That teenagers’ electricity consumption is above average has also been found in other studies (Petersen and Gram-Hanssen, 2005). Also as expected, the household’s electricity consumption is negatively related to household members’ electricity saving behavior. This confirms that households’ electricity saving efforts matter. The regression weights suggest that household members’ effort to save electricity is as important for household electricity consumption as the structural characteristics reflected in the size of the home. The significant effect of electricity saving behavior also validates that the behavior index measures what it is supposed to measure (i.e., individual household members’ electricity saving effort). The household behavior construct has an acceptable composite reliability of .75 and the two factor loadings are about equal, which suggests that men and women’s influence on the household’s electricity consumption is more or less equal.

5.4. Electricity conservation behavior For the remaining analyses, all constructs are conceptualized as mixed within and between dyads variables. Here, individual level behavioral goals and conservation behavior, respectively, are modeled as functions of psychological variables while also taking possible mutual influences between household members into account. In general, equation errors were assumed to be uncorrelated within individuals, but exceptions were made for the errors of equations having outcome expectation constructs as the dependent variable (because significant partial correlations between these variables, after controlling for self-efficacy and perceived impediments, cannot be ruled out). Errors of the same equation for men and women were allowed to correlate, but their covariances were fixed to zero if possible without significantly worsening the model fit. In order to reduce unnecessary complexity, all parameters that could be fixed to be equal between men and women without a statistically significant loss of fit were fixed to be equal. The results of the hierarchical analyses used to determine these constraints are reported with the individual analyses. The standardized structural parameters of the model predicting electricity conservation behavior by the adult household members are reported in Fig. 3.6 The fit indices, reported in the note of Fig. 3, show an acceptable fit of the model to the data. The final model explains 26%/19% of the variance in male/female electricity saving behavior. As with electricity consumption, the unexplained variance can be attributed to neither behavior nor its antecedents being perfectly measured and possibly to omitted 6 It was possible to impose the following equality constraints between the sub-models for men and women without a statistically significant loss of fit: (a) equal factor loadings (w2 change (3) ¼ 2.088, p ¼.554), (b) equal intercepts except for behavior (w2 change (6) ¼8.838, p ¼.183), (c) equal covariances that involved one and the same exogeneous variable(s) from each dyad member (w2 change (1) ¼.000, p ¼ .983), (e) equal variances for exogeneous variables and equal equation residual variances (w2 change (4) ¼ 2.628, p ¼ .622). In addition, it was possible to restrict the following possible paths to zero without a statistically significant loss of fit: (d) covariances between error terms for the same item for men and women (w2 change (6) ¼ 4.235, p¼ .645), (e) cross effects between dyad members (w2 change (8) ¼ 10.094, p ¼.258), (f) the covariance between goal residual variances (w2 change (1) ¼ 2.355, p ¼.125). As in the previous model, it was controlled whether possible antecedents further back in the presumed causal chain had a significant direct effect on electricity consumption after controlling for the included variables. None of them had and, hence, they were omitted from this analysis.

Fig. 3. From intentions to behavior. Note: w2 (78)¼113.239, p o .01,w2/df ¼1.452, RMSEA ¼.044 (90% confidence interval: .024–.061). Only the structural model. A dotted line means that the relationship is not significant at the 5% level.

antecedents (objective sociostructural factors constraining the individual’s opportunities to save electricity, task specific knowl¨ lander and Thøgersen, 1995). edge, habits, cf. O Unexpectedly, the antecedents of electricity conservation behavior differ between men and women in the sample (Fig. 3). The strongest predictor of behavior for both men and women, and the only predictor for women, is how impeding the sociostructural conditions are perceived to be. Hence, how much an individual does to save electricity is to a high extent determined by how easy or difficult this is felt to be. In addition, men’s, but not women’s, electricity saving behavior is positively related to their goals or intentions to save electricity. A third identified determinant of men’s electricity saving behavior is their perception of other household members’ behavior. Men are more likely do something to save electricity the more they perceive that other members of their household do the same. The correlation between perceptions about other household members’ behavior and perceived impediments is substantially stronger for women than for men. This may reflect that also women’s electricity conservation behavior is influenced by their perceptions of other household members’ behavior, but in that case the influence is mediated through perceived impediments. There are relatively strong positive correlations between spouses’ perceptions about other household members’ electricity conservation effort and about how impeding or facilitating conditions are for electricity saving and the residual variances of their electricity saving behavior are positively correlated as well. These correlations are as one should expect for dyad members (i.e., spouses) sharing the same structural conditions for electricity saving. Hence, they add face validity to the results. 5.5. Electricity saving intentions The standardized structural parameters of the model predicting electricity saving goals or intentions are reported in Fig. 4.7 7 In order not to clutter the figure, some correlations were omitted: Within: Impediments—Other’s behavior: .31 (both). Within, residual variances: Self-evaluativem–Positivem: .60, Self-evaluativef-Positivef: .49, Self-evaluativef–Negativef: .28. Between, same construct: Self-efficacym–Self-efficacyf: .27, Impedimentsm–Impedimentsf: .48, Other’s behaviorm–Other’s behaviorf: .57. Between, different constructs: Other’s behaviorf–Self-efficacym: .33, Other’s behaviorm-Self-efficacyf: .33, Other’s behaviorf–Impedimentsm:  .42, Other’s behaviorm–Impedimentsf:  .42, Self-efficacyf–Impedimentsm:  .26, Self-efficacym–Impedimentsf:  .26. Between, residual variances: Socialm–socialf: .72, Positivem–Positivef: .37.

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

7739

Fig. 4. From self-efficacy to electricity conservation goals. Note: w2 (665)¼ 906.763, p o .001, w2/df ¼ 1.364, RMSEA ¼ .039 (90% confidence interval: .033–.045). Only the structural model. A dotted line means that the relationship is not significant at the 5% level.

The fit indices, reported in the note of Fig. 4, suggest an acceptable fit of the model to the data.8 The model explains 61% of the variance in electricity saving intentions, which is unusually high for this kind of study. As is usually the case, the applied social cognitive model is better at predicting individual’s goals or intentions than their behavior and behavioral outcomes (cf. Abrahamse and Steg, 2009). This can partly be explained by measurement issues, especially that the social cognitive variables, including behavioral goals, are measured in a similar way, which differs from how the behavior and especially electricity consumption variables are measured. However, a likely additional reason is the fact that an individual’s goals and intentions are less constrained by external factors than is behavior and outcomes of the behavior. Neither the amount of explained variance in electricity saving intentions nor the pattern of contributing antecedents differs statistically between men and women. As predicted by theory, the individual’s electricity saving intentions depend on his or her outcome expectations as well as perceived self-efficacy. Among the outcome expectations, selfevaluative outcome expectations are most important for intentions to save electricity. The direct effects of social and physical outcome expectations are much less important. There are at least two possible reasons for this. One is the strong agreement with the

8 It was possible to impose the following equality constraints between the sub-models for men and women without a statistically significant loss of fit: (a) equal factor loadings (w2 change (11)¼19.562, p ¼.052), (b) equal intercepts (w2 change (12)¼ 11.355, p ¼.499), (c) equal covariances that involved one and the same exogeneous variable(s) from each dyad member (w2 change (6) ¼ 9.733, p ¼.136), (e) equal variances for exogeneous variables and equal equation residual variances (w2 change (8) ¼7.566, p ¼.477). In addition, it was possible to restrict the following paths to zero without a statistically significant loss of fit: (d) covariances between error terms for the same item within the dyad (w2 change (18)¼ 17.186, p ¼ .510), (e) cross effects between dyad members except for the path from Other’s behaviorm to social outcome expectationsf (w2 change (32)¼ 38.288, p ¼.206), (f) covariances of residual variances for the same endogeneous variable across dyad members except for social outcome expectations and positive outcome expectations (w2 change (3) ¼ 5.686, p ¼ .128).

statements about positive outcomes (Table 1, first column), which may have created a ‘‘ceiling effect’’ for this variable (Cohen et al., 2003). The other possible reason is that the effects of outcome expectations, and partly also social expectations, are mediated through self-expectations (Thøgersen, 2006, 2009). The latter explanation is supported by the significant (partial) correlations between outcome expectations and self-expectations (see footnote 6) and supplementary calculations finding significant bivariate correlations between positive outcome expectations and electricity saving goals for both men (.45) and women (.38) (Baron and Kenny, 1986). Be that as it may, electricity saving intentions seem to be more dependent on energy saving being perceived as a normative issue and, especially, an internalized, personal norm than as an instrumental activity directed at achieving tangible outcomes (such as saving money). If behavioral costs are expected (i.e., negative outcome expectations), this has a negative, but relatively weak direct impact on electricity saving intentions. Also as expected, electricity saving intentions are strongly related to the individual’s felt self-efficacy. The more individuals feel able to do something to avoid unnecessary electricity consumption, the more strongly they intend to do so. Perceptions regarding how much other members of the household do to save electricity influence the individual’s electricity saving goals both directly and indirectly, especially via social outcome expectations. The latter effect suggests that people to a high extent infer social expectations about electricity saving from their spouse’s and other household members’ behavior. The former effect suggests that observing others in the household saving electricity is motivating in itself. Interestingly, the only significant ‘‘cross-effect’’ from one dyad member’s antecedents to the other’s dependent variables is a negative path from men’s perceptions of other household member’s’ behavior to women’s social outcome expectations. Hence, women, but not men, feel a stronger social pressure to save electricity when their spouse perceives that they (‘‘other household members’’) do little to save electricity. This suggests that men are more inclined than women to put pressure on their spouse when they perceive that she does too little to save electricity.

7740

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

Apart from this effect, the interdependencies between dyad members (i.e., partners) at this level are mostly reflected in positive correlations between exogeneous constructs in the model. As already mentioned in connection with the previous analysis, both spouses’ perceptions of other household members’ electricity saving effort and their perceptions about the sociostructural conditions (impediments and facilitators) for electricity saving are positively and relatively strongly correlated. Spouses’ perceived self-efficacies with regard to electricity saving are also positively correlated, but less strongly than the previously mentioned constructs. Self-efficacy is partly a function of knowledge and skills, which are more individual than the contextual factors mentioned before. Also spouses’ social and positive outcome expectations, but neither their negative nor self-evaluative outcome expectations are strongly and positively correlated, after controlling for self-efficacy. Importantly, spouses’ behavioral intentions are not correlated after controlling for the included antecedents. Hence, it seems that the unexplained variance in this case is due to individual rather than to shared contextual factors.

6. Discussion 6.1. Summary of results It is an important conclusion from this study that, within the limits set by structural conditions, such as the size of the house, household members’ electricity saving effort makes a difference for a private household’s electricity consumption. Hence, it makes sense to target individual and household behavior when the aim is to reduce electricity consumption, and especially the large amount of electricity that seems to be wasted in private households. Regarding electricity saving in the household, the study identifies motivational drivers, captured by and mediated through the individual’s electricity saving goals or intentions, and structural barriers—sociostructural conditions that currently limit households’ opportunities to save electricity, or make it more difficult. From these observations follow two main approaches for making households save electricity: a motivational and a structural. The two influence paths are not completely independent, however. Perceptions about how facilitating or impeding the sociostructural conditions are both affect and are affected by motivational factors, including the person’s self-efficacy and outcome expectations. Electricity consumption is purely ‘‘derived demand,’’ integrated in practically all activities in a modern household (cooking, listening to music, personal hygiene, cleaning, working, etc.). This makes it difficult to delimit the relevant set of sociostructural conditions, and to completely account for it in a study such as this. Also, the relatively homogeneous sample means that, for example, cultural factors that determine the ‘‘normal’’ level of household electricity consumption are shared among all respondents and therefore cannot account for variation in the sample. Still, we did find evidence suggesting a strong effect of sociostructural factors – home size and family composition – on a household’s electricity consumption, consistent with other studies in the field. Sociostructural factors such as these probably especially determine the quantity of electric devices in the household. As we argued earlier, the large amount of unexplained variance is at least partly due to omitted variables, some of which are sociostructural factors such as the quality of devices owned by the household (e.g., types of TV and computer monitors, age of freezer, pumps, etc.), the amount of time family members spend in the home versus other places (at work, traveling, second home, etc.). In the studied context, variations in the motivation to save electricity are related to self-efficacy and social and self-evaluative outcome expectations (or social and personal norms). Because

most people in the studied population are already convinced about the important benefits of energy saving, changing these expectations is not the way to promote energy saving intentions or behavior. Rather, social normative and self-efficacy enhancement ¨ and Johnsson, (often referred to as ‘‘empowerment,’’ e.g., Lindstrom 2003; Thøgersen, 2005) approaches are needed to increase the motivation (i.e., goals or intentions) to save electricity even further. Because we succeeded in getting responses from two adults in many of the participating households, we have information both about gender differences and about social dynamics within the family, directly or indirectly influencing electricity consumption. Women report to do (slightly) more to save electricity in the home than men, mostly, but not only, with regard to household chores that have traditionally been women’s tasks. However, the differences in behavioral antecedents and the motivational interdependencies between partners are perhaps even more interesting. Whereas both men’s and women’s electricity saving behavior is related to their experience of facilitating or impeding sociostructural conditions for electricity saving, for men, but not women, additional variance in electricity saving behavior is accounted for by including their own goals or intentions and their perceptions about other household members’ behavior. This suggests two things: (1) that men have a more agentic approach to electricity saving than women (cf. Bandura, 2001) and (2) that women’s electricity saving effort sets a positive example that their spouse tends to follow. There are also findings suggesting that men influence their spouse’s behavior, but apparently at least partly through a different route. There is a relatively strong negative relationship between how much men do to save electricity and their spouse’s perception of how easy or difficult it is to avoid unnecessary electricity consumption. Based on the current study, one can only speculate why this is so. For example, this finding might reflect that, according to the traditional division of labor in households, it is often a man’s task to implement the electricity saving ideas generated by his wife. In that case, if a man is dragging his feet, his wife may find it (even more) difficult to do anything to save electricity. Reflecting the communication within families, the study reveals quite high levels of agreement between spouses about potential outcomes of electricity saving efforts and social expectations. In terms of non-verbal communication, the example set by one’s spouse seems to influence perceptions about social expectations, the self-evaluative relevance, and the individual goals about electricity conservation of both men and women. Not all interactions between spouses are necessarily symmetric, however. We have presented evidence suggesting that men who feel they do more than their wife to save electricity, to a higher extent than vice versa, use explicit pressure to influence their partner to save more. Programs to promote electricity saving in the household should take into account the different needs, abilities and other prerequisites of household members as well as insights such as these about the interactions between household members. Social expectations (or subjective social norms) are a weak direct antecedent of participants’ electricity saving goals. However, they are quite strongly related to self-evaluative expectations (or personal norms), a relationship that has been attributed to personal norms to some extent being internalized social norms (Thøgersen, 2006, 2009). Hence, the total effect of influencing electricity consumers’ social expectations may be stronger than what appears from their direct effect on electricity saving goals. Self-evaluative outcome expectations are also related to the person’s beliefs about positive outcomes of saving electricity. Previous research suggests that by combining social normative and informational influences, where the latter increases the salience of positive outcome expectations, it is possible to obtain behavioral effects that are larger than their additive effect (LaTour and Manrai, 1989; Thøgersen, 2009).

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

7741

Electricity saving intentions are strongly related to self-efficacy and theory suggests that efficacy-related factors may also influence intentions indirectly, via outcome expectations. Further, self-efficacy is related to how facilitating or impeding the sociostructural conditions for electricity saving are perceived to be, which is directly related to behavior. Hence, the parameter estimates suggest that strengthening self-efficacy could have substantial effects on electricity saving intentions and behavior. This speaks in favor of using empowerment approaches to promote electricity saving in the household. Electricity saving intentions are even more strongly related to social and self-evaluative outcome expectations, combined, than to efficacy. This suggests that social normative approaches are also potentially useful means to increase electricity conservation. However, campaigners using an empowerment approach can avoid some difficult issues in relation to social normative approaches, such as the ethicality of evoking guilt feelings in people for performing common, everyday activities and the risk of provoking psychological reactance (e.g., Tertoolen et al., 1998).

model, has been shown to produce unbiased estimates in cases such as this. Hence, the biggest problem caused by the missing values is that they are likely to reduce the statistical power of estimates, especially estimates of between person paths in this case. The consequence is an increased risk of Type 1 error for the affected paths, which means that our estimates of between person influences are likely to be conservative. It is also a serious limitation that important antecedents seem to be omitted, leading to a large amount of unexplained variance in both behavior and behavioral outcomes. For example, it is likely that variation in the quality of electrical devices in the household can account for some of the unexplained variance in behavioral outcomes in electricity consumption. It is equally likely that many electricity-consuming behaviors are performed in a habitual way and that habit can account for some of the unexplained variance in behavior (Barr et al., 2005). To the extent that omitted variables overlap included ones, parameter estimates may be biased. Hence, future research should attempt to obtain a more complete set of predictor variables.

6.2. Limitations

6.3. Implications

A number of limitations need to be kept in mind when interpreting these results, including the use of self-reports, crosssectional data, omitted variables, missing values, and the specific items used to represent electricity saving behavior. Psychological data can only be obtained from self-reports and in practice even information about (electricity saving) behavior is difficult to obtain in any other way. We have minimized the problem that self-reported data are fallible by using multiple items to measure constructs, which makes it possible to estimate measurement error when using structural equation modeling. In SEM, composite reliabilities of .60 and above are usually considered acceptable. Some of the psychological constructs (i.e., negative and social outcome expectations) have lower reliabilities than this, which could be part of the reason why they are weakly related to electricity saving intentions. By using 17 specific items that were easy to answer, the validity of the measure of electricity saving behavior was optimized. Its significant correlation with electricity consumption, read from the meter, attests to the validity of this measure. The set of behavior items is subject to critique on one account, however. It appears that relatively few items account for the difference in electricity saving behavior between men and women. Three out of four of these items refer to activities that are traditionally mostly carried out by women. Hence, it is possible that the finding of an overall gender difference in electricity saving behavior is sensitive to the specific items included in the instrument and that a more gender balanced set of items would have led to a different result. Our data being cross-sectional means that they are mute about the direction of causality (as is true for most survey based studies). Causal relationships imply a time sequence (the cause must come before the effect), which cannot be captured in crosssectional research. Hence, inferences about the direction of relationships are based on the theoretical model and previous research and cannot be validated by the type of empirical study reported here. The data set suffers from missing values, especially with regard to households where only one of the adults filled out a questionnaire. Since the missing responses are reasonably equally divided between the sexes, about 160 respondents are available to estimate each within-person path. However, the between person paths are estimated with fewer observations (83 couples). Full information maximum likelihood, which was used to estimate the

Households can change some of the mentioned sociostructural conditions determining their electricity consumption (e.g., the quality of electrical devices), which means that they can be influenced through the motivational means discussed in the following. Various sociostructural conditions can also be influenced directly through standards (e.g., a legal standard for standby consumption) or pricing (e.g., a CO2 tax increasing the price of electricity or a subsidy for replacing white goods with a new, energy-saving model). The available evidence, in this and other studies, suggest that authorities should focus most of their attention at adapting important sociostructural conditions to become more facilitating for electricity conservation. Electricity suppliers are probably in the best position to reduce impediments with regard to electricity saving. Especially, research suggests that electricity suppliers could be much better at servicing households with relevant electricity consumption information at the time and place where they need it (Fischer, 2008). Effective, timely and convenient feedback about home electricity consumption would directly reduce some of the impediments that makes saving electricity difficult (Abrahamse et al., 2007; Fischer, 2008). It ¨ and also has the potential to ‘‘empower’’ consumers, (cf. Lindstrom Johnsson, 2003; Thøgersen, 2005), that is, improve their selfefficacy regarding electricity saving through (a) increasing and improving individual consumers’ knowledge about their electricity consumption and especially about how their electricity consumption is related to their individual and collective behavior, (b) creating mastery experiences when individuals are able to see the outcomes of their efforts, and (c) providing additional social encouragement and support in this endeavor (cf. Bandura, 1986). However, feedback does not always result in the intended effects. If the feedback reveals poor performance, this may result in decreased motivation, which eventually may result in the reverse effect, that is, an increase in energy consumption (Kluger and DeNisi, 1996). Also, if people are already doing well (saving as much energy as they possibly can) feedback will hardly improve their results. It has been reported that feedback suggesting that the household is doing better than average may actually lead to an increase in its energy consumption (Schultz et al., 2008). Schultz et al. (2008) found that the latter effect can be neutralized by accompanying the feedback with a happy smiley, communicating approval of the household’s excellent performance. However, all in all, research shows that considerable care is needed when designing the feedback.

7742

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

The finding that the relationship between electricity saving behavior and electricity consumption seems to be logarithmic also may have interesting implications. Since the relationship is negative, the finding that it seems to be logarithmic suggests that there is a decreasing marginal return from increasing the number and intensity of electricity saving behaviors. This is in accordance with the notion that there are ‘‘low hanging fruits’’ to be reaped in this area and suggests that households that are already doing an average effort to save electricity get less out of increasing their effort than households that are only doing little. The logarithmic relationship also suggests that the relatively infrequently performed electricity saving behaviors have a smaller impact than the more frequently performed ones. Perhaps this means that it is especially when going beyond the ordinary or obvious that empowerment is needed? These inferences are speculative, but they are important enough to merit attention in future research. In sum, it is an important message from this study that the amount of squandering of electricity in private households can be reduced by reducing the impediments for saving electricity in the household and increasing their effectiveness in this endeavor, especially by electricity suppliers. Recent studies show that effective feedback about the household’s electricity consumption that is timely, convenient and has an appropriate level of detail has the potential to make electricity saving easier and more effective, in addition to focusing the attention of private consumers towards unwanted high electricity consumption in the home. According to the present study, such a change in the sociostructural conditions for electricity saving has the additional benefit of helping consumers perceive the effects of their effort, which will strengthen their self-efficacy with regard to this task and thereby motivate them to increase their effort.

References 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., 2007. The effect of tailored information, goal setting, and tailored feedback on household energy use, energy-related behaviors, and behavioral antecedents. Journal of Environmental Psychology 27, 265–276. Ajzen, I., 2002. Perceived behavioral control, self-efficacy, locus of control, and the theory of planned behavior. Journal of Applied Social Psychology 32, 1–20. Ajzen, I., Fishbein, M., 1980. Understanding Attitudes and Predicting Social Behavior. Prentice-Hall, Englewood Cliffs. Amann, J.T., Wilson, A., Ackerly, K., 2007. The Consumer Guide to Home Energy Savings: Save Money, Save the Earth ninth ed. New Society Publishers. Arbuckle, J.L., 1996. Full information estimation in the presence of incomplete data. In: Marcoulides, G.A., Schumacker, R.E. (Eds.), Advanced Structural Equation Modeling: Issues and Techniques. Lawrence Erlbaum Associates, Mahwah, NJ, pp. 243–277. Arbuckle, J.L., Wothke, W., 1999. Amos 4.0 Users’ Guide. Marketing Division SPSS Inc.: SmallWaters Corporation, Chicago, IL. Bagozzi, R.P., 1994. Structural equation models in marketing research: basic principles, Principles of Marketing Research. Blackwell Publishers, Cambridge, pp. 317-385. Bandura, A., 1986. Social Foundations of Thought and Action: A Social Cognitive Theory. Prentice-Hall, Englewood Cliffs, NJ. Bandura, A., 2001. Social cognitive theory: an agentic perspective. Annual Review of Psychology 52, 1–26. Bandura, A., 2004. Health promotion by social cognitive means. Health Education and Behavior 31, 143–164. Bang, H.-K., Ellinger, A.E., Hadjimarcou, J., Traichal, P.A., 2000. Consumer concern, knowledge, belief, and attitude toward renewable energy: an application of the reasoned action theory. Psychology and Marketing 17, 449–468. Baron, R.M., Kenny, D.A., 1986. The moderator–mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology 51, 1173–1182. Barr, S., Gilg, A.W., Ford, N., 2005. The household energy gap: examining the divide between habitual- and purchase-related conservation behaviours. Energy Policy 33, 1425–1444. Berger, I.E., Corbin, R.M., 1992. Perceived consumer effectiveness and faith in others as moderators of environmentally responsible behaviors. Journal of Public Policy and Marketing 11, 79–89.

Black, J.S., Stern, P.C., Elworth, J.T., 1985. Personal and contextual influences on household energy adaptations. Journal of Applied Psychology 70, 3–21. Bollen, K.A., 1989. Structural Equations with Latent Variables. Wiley-Interscience. Clift, J., Cuthbert, A., 2007. Energy: Use Less-save More: 100 Energy-saving Tips for the Home. Chelsea Green Publishing Company. Cohen, J., Cohen, P., West, S.G., Aiken, L.S., 2003. Applied Multiple Regression/ Correlation Analysis for the Behavioral Sciences third ed. Lawrence Erlbaum, Mahwah, NJ. Dietz, T., Gardner, GT., Gilligan, J., Stern, P.C., Vandenbergh, M.P., 2009. Household actions can provide a behavioral wedge to rapidly reduce US carbon emissions. Proceedings of the National Academy of Sciences (PNAS) 106, 18452–18456. Ekins, P., Skea, J. (Eds.), 2009. UKERC Energy 2050: Making the Transition to a Secure and Low-Carbon Energy System: Synthesis Report. UK Energy Research Centre, London. Ellen, P.S., Wiener, J.L., Cobb-Walgren, C., 1991. The role of perceived consumer effectiveness in motivating environmentally conscious behaviors. Journal of Public Policy and Marketing 10, 102–117. Fischer, C., 2008. Feedback on household electricity consumption: a tool for saving energy? Energy Efficiency 1 (1), 79–104. Genius Access A/S, 2008. Elsparefondens Indsats over for Private Husholdninger. Elsparefonden, København. Gist, M.E., Mitchell, T.R., 1992. Self-efficacy: a theoretical analysis of Its determinants and malleability. Academy of Management Review 17, 183–211. Gram-Hanssen, K., Kofod, C., Petersen, K.N., 2004. Different everyday lives— different patterns of electricity use. In: Paper presented at the Proceedings of the 2004 American Council for Energy Efficient Economy Summerstudy in Buildings, Washington, DC. ˚ Grimmig, M., 1992. Brugsmønstre og energiforbrug i SmørumGardhuse. - et forsøg med energistyring i boligen. Dansk Teknologisk Institut. Guerin, G.A., Yust, B.L., Coopet, J.G., 2000. Occupant predictors of household energy behavior and consumption change as found in energy studies since 1975. Family and Consumer Sciences Research Journal 29, 48–80. Harland, P., Staats, H.J., Wilke, H.A.M., 1999. Explaining pro-environmental intention and behavior by personal norms and the theory of planned behavior. Journal of Applied Social Psychology 29, 2505–2528. Jarvis, C.B., Mackenzie, S.B., Podsakoff, P.M., 2003. A critcal review of construct indcators and measurement model misspecification in marketing and consumer research. Journal of Consumer Research 30, 199–218. Jensen, O.M., 2005. Barrierer for realisering af energibesparelser i bygninger (Notat). Statens Byggeforskningsinstitut, København. Kaiser, F.G., 1998. A general measure of ecological behavior. Journal of Applied Social Psychology 28, 395–422. Kaiser, F.G., Wilson, M., 2000. Assessing people’s general ecological behavior: a cross-cultural measure. Journal of Applied Social Psychology 30, 952–978. Kenny, D.A., Kashy, D.A., Cook, W., 2006. Dyadic Data Analysis. Guilford Publications, New York, NY. ¨ ¨ Ketola, A., 2000. Manskliga dimensioner vid energianvandning—Vetenskaplig ¨ forskningsfront ur nationellt och internationellt perspektiv. Institutionen for ¨ ¨ Varme-och Kraftteknik, Tekniska Hogskolan i Lund, Lund. Kluger, A.N., DeNisi, A., 1996. The effects of feedback interventions on performance: a historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin 119, 254–284. LaTour, S.A., Manrai, A.K., 1989. Interactive impact of informational and normative influence on donations. Journal of Marketing Research 26, 327–335. ¨ Lindstrom, M., Johnsson, P., 2003. Environmental concern, self-concept and defence style: a study of the Agenda 21 process in a Swedish municipality. Environmental Education Research 9, 51–66. Midden, C.J.H., Ritsema, B.S.M., 1983. The meaning of normative processes for energy conservation. Journal of Economic Psychology 4, 37–55. Olsen, M.E., 1981. Consumers’ attitudes toward energy conservation. Journal of Social Issues 37 (2), 108–131. ¨ lander, F., Thøgersen, J., 1995. Understanding of consumer behaviour as a prerequisite O for environmental protection. Journal of Consumer Policy 18, 317–357. Pedersen, L.H., Broegaard, E., 1997. Husholdningernes elforbrug—en analyse af ˚ attituder og adfærd pa˚ energi- og miljøomradet. AFK forlaget, København. Petersen, K.N., Gram-Hanssen, K., 2005. Husholdningers energi- og vandforbrug: Afhængighed af socio-økonomiske baggrundsvariable (No. SBi 2005:09). Statens Byggeforskningsinstitut (Danish Building Research Institute), Hørsholm. Poortinga, W., Steg, L., Vlek, C., 2004. Values, environmental concern, and environmental behavior: a study into household energy use. Environment and Behavior 36, 70–93. Poortinga, W., Steg, L., Vlek, C., Wiersma, G., 2003. Household preferences for energy-saving measures: a conjoint analysis. Journal of Economic Psychology 24, 49–64. Ritchie, J.R.B., McDougall, G.H.G., 1985. Designing and marketing consumer energy conservation policies and programs: implications from a decade of research. Journal of Public Policy and Marketing 4, 14–32. Schultz, P.W., Nolan, J.M., Cialdini, R.B., Goldstein, N.J., Griskevicius, V., 2008. The constructive, destructive, and reconstructive power of social norms. Psychological Science 18, 429–434. Tertoolen, G., Kreveld, D.V., Verstraten, B., 1998. Psychological resistance against attempts to reduce private car use. Transportation Research A 32, 171–181. Thøgersen, J., 2002. Waste removal systems and recycling participation in ¨ residential environments. In: Aragone´s, J.I., Francescato, G., Garling, T. (Eds.), Residential Environments: Choice, Satisfaction, and Behavior. Bergin & Garvey, Westport, CT, pp. 241–256.

J. Thøgersen, A. Grønhøj / Energy Policy 38 (2010) 7732–7743

Thøgersen, J., 2005. How may consumer policy empower consumers for sustainable lifestyles? Journal of Consumer Policy 28, 143–178. Thøgersen, J., 2006. Norms for environmentally responsible behaviour: an extended taxonomy. Journal of Environmental Psychology 26, 247–336. Thøgersen, J., 2009. The motivational roots of norms for environmentally responsible behavior. Basic and Applied Social Psychology 31, 348–362. Uitdenbogerd, D., Egmond, C., Jonkers, R., Kok, G., 2007. Energy-related intervention success factors: a literature review. In: Paper presented at the the 2007 ECEE Summer Study.

7743

Vringer, K., Aalbers, T., Blok, K., 2007. Household energy requirement and value patterns. Energy Policy 35, 553–566. Wilhite, H., Nakagami, H., Masuda, T., Yamaga, Y., Haneda, H., 2001. A crosscultural analysis of household energy-use behavior in Japan and Norway. In: Miller, D. (Ed.), Consumption: Critical Concepts in the Social Sciences, vol. 4. Routledge, London, New York, pp. 159–177. Wilson, C., Dowlatabadi, H., 2007. Models of decision making and residential energy use. Annual Review of Environment and Resources 32, 169–203.