1
Accelerating the energy transition:
The incorporation of social influence in governmental information campaigns
Submitted by:
Naomi Joanne Koers
S2680386
MSc BA Strategic Innovation Management
Submitted to:
Dr. K. R. E. Huizingh
Co-assessor:
Prof. dr. M. Mulder
Date of submission: 14
thof February 2019
Word count:
14859
Faculty of Economics and Business University of Groningen
2
ABSTRACT
As the current policy mix is unable to move a large part of the population toward the adoption of energy retrofit measures it is important to consider alternative approaches. This study examines whether policies based on social influence are able to contribute to bridging the value-action gap with regard to the adoption of energy retrofit measures by households. Whether this potential effect is stronger when energy retrofit measures are observable is also considered. In order to test the hypotheses of this study data from a lab experiment is used in binary logistic regression models, a cross tabulation and a Chi-square test. The results do not indicate that social influence positively affects investment in energy retrofit measures. Support for a stronger effect for observable measures is also not found. Hence, from the current study cannot be concluded that policies based on social influence are able to contribute to bridging the value-action gap with regard to the adoption of energy retrofit measures. However, as the inclusion of social influence and the experimental design were potentially flawed, further research has to confirm whether the findings of this study are valid. This study makes a contribution to literature on how to bridge the value-action gap in an energy retrofit measure adoption context.
1. INTRODUCTION
It is important that pro-environmental measures are taken and policies are developed because most of the EU countries, including the Netherlands, have signed a universal, legally binding global climate deal at the Paris climate conference in December 2015. The 195 countries that signed the agreement have the shared ambition to avoid dangerous climate change by limiting global warming to well below 2°C (European Commission). The implementation of successful governmental policies to counteract global warming is thus vital to adhering to the legal obligations while it simultaneously contributes to the societal goal of sustaining life on Earth for future generations.
3
(Grösche & Vance, 2009; Murphy et al., 2010). Another example is the provision of general information about energy conservation, which is an ineffective, yet often employed policy instrument according to Abrahamse et al. (2005).
In addition to this, it is nowadays more difficult to persuade people to act pro-environmentally than it used to be in the past, and therefore governments have to think of new ways to engage citizens and other actors (Ligteringen, 1998). Research on the effectiveness of environmental policies is relevant because insights from these studies can support governmental institutions in the allocation of their budgets towards measures that are known to have an impact. This study attempts to shed light on one aspect of the effectiveness of governmental policies through examining the effect of the incorporation of social influence in governmental information campaigns on the adoption of energy retrofit measures. The aim of the study is to investigate whether policies based on social influence could be an effective addition to existing energy policies. Besides the societal relevance this study also attempts to make a theoretical contribution. Several studies indicate that most people are aware of the fact that the energy consumption has to go down and a transition towards the usage of more renewable energy sources has to be made in order to sustain life on earth in the long run (e.g., Worcester, 1994; Blake & Carter, 1997; Steg, 2008). However, relatively few people take matters into their own hands through adjusting their consumption behaviour or investing in energy retrofit measures. This phenomenon is sometimes referred to as the value-action gap (Blake, 1999). To foster behavioural change within households Palm and Ellegård (2011) call for policies that are adjusted to the specific characteristics of the household sector. In response to this call this study tries to overcome the value-action gap with regard to environmental policies for households by considering social influence as a solution to bridge the value-action gap. Hence, the main research question in this study is the following: RQ1: Are policies based on social influence able to contribute to bridging the value-action gap with
regard to the adoption of energy retrofit measures by households?
4
RQ1a: Are policies based on social influence able to contribute to bridging the value-action gap with regard to the adoption of unobservable energy retrofit measures by households?
The focus of this research will be on the energy requirements of households. It is relevant to consider households in this study because this category continues to face challenges with regard to energy conservation. The implementation of energy efficiency measures in households, such as using domestic heating systems, switching to electrical appliances and insulating houses, has for instance been partly offset by the increase in the total number of households and the decreasing household size (e.g., van der Wal & Noorman, 1998; Ellegård, 2010; Trotta, 2018). Moreover, despite the fact that most household appliances have become more environmentally friendly over the years, households nowadays also own more appliances resulting in an increase in the energy requirement per household (e.g., Ellegård, 2010; Planbureau voor de Leefomgeving, 2012; Trotta, 2018).
This study focusses on the adoption of large energy retrofits. More specifically, the adoption of insulation and solar panels will be examined. Insulating is an invisible energy efficiency measure that results in energy savings whereas solar panels are visible and part of the transition towards more renewable energy sources. Both require a substantial investment and can be classified as energy retrofit measures. While measures that stimulate energy-saving behaviours that do not rely on a monetary investment are also a key component in the development of governmental climate conservation policies (Trotta, 2018), those are left out of scope in this study. The reasoning behind this scoping decision is that investments in energy efficiency measures have a higher energy reduction potential compared to energy-saving behaviours (Stern and Gardner, 1981). In addition to this, according to Murphy et al. (2012) policy instruments that focus on the energy performance of existing dwellings have been under-researched. To date, the focus of environmental policy has mainly been on public authorities and private industries (Palm & Ellegård, 2011).
5
2. LITERATURE REVIEW
In her review on how governments can influence households to invest in energy retrofit measures Mulder (2018) identified four categories of importance to energy retrofit measure investment: (1) household characteristics; (2) home characteristics; (3) retrofit measure characteristics; and (4) social influence. Because it is difficult to directly influence the first three factors via governmental policies she suggests to focus policy development on the latter category: social influence. In this section the theoretical foundation of policies based on social influence is discussed. In addition to this, other factors that influence energy retrofit investment decisions are touched upon.
Figure 1: Mulder’s theoretical framework
Source: Mulder (2018)
2.1 Social influence
This sub-section starts with a discussion of the “information deficit model”, a model on which many energy related information policies are currently based. Then an alternative model that is less often employed is introduced. It is explained how this model can influence the behaviour of people. Emphasis is placed on the social influence aspect of the model as this study examines whether policies based on social influence are able to contribute to bridging the value-action gap with regard to the adoption of energy retrofit measures by households.
2.1.1 Behavioural models
6
“information deficit model” (Owens & Driffill, 2008), which assumes that information campaigns can increase one’s knowledge and awareness. Increased knowledge and awareness can then change one’s attitude toward certain behaviour, which will in turn lead to behavioural change.
Figure 2: Owens and Driffill’s information deficit model
However, Schultz (2002) found that merely providing information might not be an incentive for people to move towards the adoption of energy conservation measures. He reviewed several studies that made use of the distribution of information to educate people and found that increased awareness often only leads to a small, short-lived change in behaviour. Nolan et al. (2008, p. 921) share this point of view, they state that “appealing to people to do the right thing, or to protect the environment, rarely succeeds in increasing levels of pro-environmental behaviour.”
Because information campaigns based on the information deficit model are often unable to foster behavioural change the usage of another behavioural model in governmental information campaigns is proposed in this study. The Theory of Reasoned Action (TRA) is a behavioural adoption theory that can be employed in attempts to provide reasoning for the decisions made by individuals to perform, or refrain from performing, certain behaviour. The TRA that has been developed by Fishbein and Ajzen (1975) is based on the principle that an individual’s behaviour depends on the strength of one’s intention to perform that behaviour. The intention to behave in a certain manner is in turn influenced by an individual’s personal attitudes towards the behaviour and by their subjective norm. The subjective norm reflects how someone perceives that important others feel about the behaviour in question and can thus be considered a form of social influence (Abrahamse & Steg, 2013).
Figure 3: Fishbein and Ajzen’s Theory of Reasoned Action
Information Attitude toward
behaviour Behaviour Knowledge and awareness Attitude toward behaviour Subjective norm
7
2.1.2 Descriptive and injunctive norm
In order to get an understanding about how subjective norms are able to affect people’s decisions one has to delve a little deeper into the concept. Subjective norms can be divided into two categories: (1) descriptive norms and (2) injunctive norms. Descriptive norms are perceptions about what most other people do, they are comprised of factual information (Cialdini et al., 1991). Because imitating what most other people do usually leads to proper and efficient decisions, descriptive norms can provide a decisional shortcut or information-processing advantage when one decides on how to act in a certain situation (Cialdini, 1988). The effect of descriptive norms on behaviour has been tested and confirmed in numerous domains. Burnkrant and Cousineau (1975) for instance found that people that observed others evaluating the taste of a coffee favourably, were more likely to evaluate the taste of the coffee favourable themselves than they would have been without having observed others’ favourable ratings. Another example can be found in a study from Schultz and Tyra (2000). They that argue that descriptive norms are a strong predictor of recycling behaviour. A more recent example is provided by Allcott (2011) who found that providing information that compared one’s electricity usage with neighbours decreased energy consumption among above average consumers. Curtius et al. (2018) confirm the effect of descriptive norms in an energy retrofit investment context. They found that presence of many solar panels in a neighbourhood provided a positive signal to the community that resulted in reduced uncertainty toward the adoption of solar panels among other community members.
Injunctive norms, on the other hand, refer to perceptions about what most people would approve or disapprove of, they are about what people ought to do. Injunctive norms can motivate decisions because people want to adhere to the moral rules of the group. They believe that doing so will result in social rewards whereas failing to do so will lead to social punishment (Cialdini et al., 1991).
8
three sub-sections about household characteristics, home characteristics and retrofit measure characteristics can shape attitudes that individuals have towards investment in energy retrofit measures and therefore together represent the “attitude toward behaviour” branch of the TRA model in an energy retrofit investment context.
2.2 Household characteristics
In this sub-section household characteristics that may influence whether people invest in energy retrofit measures are discussed. Only factors that are known to have an impact are considered. The following influencing factors are touched upon: (1) environmental norms; (2) personal innovativeness; (3) income and funding; (4) risk aversion; and (5) heterogeneous opinions.
Environmental norms. According to Girod et al. (2017) environmental norms and personal innovativeness
affect whether someone will adopt a novel green technology. Environmental norms refer to someone’s moral obligation to act pro-environmentally (Stern, 2000). Stern (2000) argues based on his value-belief-norm theory that one’s environmental value-belief-norms can drive pro-environmental behaviour. However, there are also studies that contradict this argument. Poortinga et al. (2003) for instance found that people with high environmental awareness were less likely to invest in energy efficiency measures. Girod et al. (2017) state that these contradicting findings may result from other factors of influence that were not considered in the study designs. They found that environmental norms are often studied in isolation while a less narrow focus could provide a more reliable basis to explain one’s findings. Hence, they are in favour of including environmental norms among other potential factors of influence in a study design.
Personal innovativeness. Personal innovativeness can be defined as someone’s willingness to experiment
with new technologies (Agarwal & Prasad, 1998). People that score high on personal innovativeness are more likely to positively evaluate new technologies (Lewis et al., 2003). Rogers’ (2003) diffusion of innovation theory provides reasoning for this finding. People that score high on personal innovativeness are more likely to be aware of energy efficiency measures which may result in the creation of a need for an energy efficiency measure and subsequently the actual adoption of the measure.
Income and funding. Frederiks et al. (2015) state that a higher income increases a household’s capacity to
9
Risk aversion. When investments in energy retrofit measures will pay off depends on several uncertain
factors. It is for example unknown how energy prices will change over time, whether the adopted technology is reliable, and how much energy someone is going to consume in future. According to Qiu et al. (2014) more risk averse people are less likely to invest in energy efficient technologies.
Heterogeneous opinions. Achtnicht and Madlener (2014) point to the fact that people within a household
might not agree on whether to invest in energy retrofit measures.
2.3 Home characteristics
In this sub-section home characteristics that may influence whether people invest in energy retrofit measures are discussed. Only factors that are known to have an impact are considered. The following influencing factors are touched upon: (1) renovation required; (2) renovation not required; and (3) home ownership.
Renovation required. The fact that a dwelling renovation was due anyway provided for 46.1% of the
respondents to the survey from Achtnicht and Madlener (2014) a reason to invest in energy retrofit measures. Similarly, Nair et al. (2010) found that owners of houses that were in a poor physical or aesthetical condition were more likely to invest in energy efficiency measures.
Renovation not required. That a renovation of the building envelope was not required was considered a
barrier to investment for 61.5% of the respondents to the survey form Achtnicht and Madlener (2014). In addition to this, 36.5% of the respondents to the survey of Achtnicht and Madlener (2014) stated that their house was already energy-optimized and that investment in energy retrofit measures was thus not necessary. Nair et al. (2010) provide a similar argument as they state that past investment might lead homeowners to think that their house is energy efficient enough and that further investment is not required.
Home ownership. Frederiks et al. (2015) found that homeowners are more likely to invest in energy
10
2.4 Retrofit measure characteristics
In this sub-section retrofit measure characteristics that may influence whether people invest in energy retrofit measures are discussed. Only factors that are known to have an impact are considered. The following influencing factors are touched upon: (1) energy cost savings potential; (2) uncertain cost savings; (3) initial investment; (4) comfort; and (5) aesthetics.
Energy costs savings potential. For 65.0% of the respondents to the survey by Achtnicht and Madlener
(2014) the possibility that investment in retrofit measures could lower their energy bill proved a motivation to invest in energy retrofit measures. Qiu et al. (2014) found similar results as 48% of the respondents to their survey identified expected energy bill savings as a motivation to conduct home energy efficiency improvements.
Uncertain cost savings. That it is uncertain whether energy retrofit measures will pay off proved to be a
barrier to investment in energy retrofit measures for 50.5% of the respondents to the survey of Achtnicht and Madlener (2014). In addition to this, 42% of the respondents to the survey from Qiu et al. (2014) stated that the uncertain energy savings posed a barrier for them to investment in energy retrofit measures.
Initial investment. That it is expensive to invest in energy retrofit measures provided a barrier to investment
for 48% of the respondents to the survey by Qiu et al. (2014). As money can only be spend once there are also opportunity costs associated with investments in energy retrofit measures.
Comfort. That energy retrofit measures can increase living comfort was for 37.3% of the respondents to the
survey from Achtnicht and Madlener (2014) a motivation to invest in energy retrofit measures. In line with this finding, several authors reported that a desire for increased thermal comfort can drive investments in insulation (e.g., Berry et al., 1997; Fuchs et al., 2004; and Herring et al., 2007).
Aesthetics. Faiers and Neame (2006) argue that some people do not want to invest in solar panels because
they dislike the aesthetics of solar panels.
3. Conceptual model and hypotheses
11
line with the TRA it is expected that social influence increases the likelihood that energy retrofit measures are carried out.
H1: Social influence positively affects the investment in energy retrofit measures by households.
3.1 Visibility and reputational effects
The visibility of the energy retrofit measure may affect the relationship between social norms and the adoption of energy retrofit measures because social norms tend to have a more evident effect when the desired behaviour is observable by others (Schultz, 2002). This is the case because the implementation of visible energy retrofit measures can result in a “green” reputation. People value a “green” reputation because they experience effect dependence, they are concerned about how others evaluate their behaviour because they rely on these individuals to meet their own needs (Jones & Gerard, 1967). Unlike a pro-self reputation, a pro-social reputation could for instance lead to relationships, leadership opportunities and friends. Hence, visualizing significant others’ “green” reputation compared to one’s own “grey” reputation might trigger people that are not receptive to existing policies to invest in energy retrofit measures after all (Delmas & Lessem, 2014).
Meyer and Vlieg (1979) found support for this visibility argument. Their results indicate that wall-cavity insulation was less popular than having double glass windows installed due to the visibility of double glass. Moreover, Bollinger and Gillingham (2012) suggest that peer effects are stronger when relatively large solar panel installations are present in one’s neighbourhood in comparison to the presence of smaller, less visible installations. However, it has not been tested whether a positive effect between social influence and adoption can also be found for unobservable energy retrofit measures. Hence, the question remains whether visibility of an energy retrofit measure is a necessity for the relationship between social influence and investment in energy retrofit measures to hold or if it merely strengthens the relationship.
In line with Schultz’s (2002) observability argument it is expected that the relationship between social influence and the adoption of energy retrofit measures is stronger for visible energy retrofit measures. H2: Visibility of energy retrofit measures positively moderates the relationship between social
influence and the adoption of energy retrofit measures.
3.2 Conceptual model
12
shape attitudes toward the adoption of energy retrofit measures and thus jointly represent the “attitude toward behaviour” branch of the TRA model.
Figure 4: Conceptual model
* Influencing factors that are not measured in the current study but taken into consideration in the experimental design.
** Not applicable to all energy retrofit measures.
In response to the call from Steg (2008) actual investment behaviour is studied instead of behavioural intention because behavioural intention not necessarily results in adoption. In an energy retrofit investment decision and subjective norm context Curtius et al. (2018) also call for research that focuses on actual adoption. They found that the amount of solar panel systems installed in the neighbourhood, the descriptive norm, and the perceived social pressure to act according to the norm, the injunctive norm, increased the intention of participants to have solar panels installed. However, they could not examine to what extent people responded in a socially desirable manner because only the intention to invest was recorded and not whether participants actually invested in a solar panel system. As energy retrofits require a relatively high
Social influence Energy retrofit investments by households H1 + Visibility of energy retrofit measure H2 + Influencing factors: Household characteristics o Environmental norms o Personal innovativeness o Income and funding* o Risk aversion
o Heterogeneous opinions* Home characteristics
o Renovation required* o Renovation not required* o Home ownership*
13
initial investment and are accompanied with uncertain energy cost savings, it is important that monetary consequences are taken into account in experimental designs.
4. RESEARCH DESIGN AND DATA
4.1 Materials
This study made use of a lab experiment with a social influence (yes versus no) between subjects design. In this sub-section the experimental design is introduced. The entire experiment can be reviewed in Appendix A. The online software tool Qualtrics was used to run the experiment.
4.1.1 Treatment
In this sub-section the treatment design is explained. First of all, the investment decisions are discussed. In the experiment participants had to decide if they were going to invest in insulation and in solar panels or not. They were told to assess the decisions independently from each other. One hundred and thirty-seven participants (49.6%) received the insulation decision first whereas the remainder of the participants (50.4%) decided upon the installation of solar panels first to rule out any potential ordering effects.
14
“You decided to invest in insulating your home. The average temperature during the winters has been quite high. This means that it took you between 9 and 11 years to recoup your investment. As mentioned in the payoff matrix that was part of the information about the investment decision, investing in insulation
while the winters are warm is unfavourable from a strictly financial point of view. Hence, your experiment participation fee will be decreased by €1.”
Secondly, the inclusion of social influence is discussed. Social influence was incorporated in the experiment via two different approaches. The first approach was a between subjects design. Half of the participants were assigned to the social influence treatment whereas the other half functioned as the control group. Participants were assigned to either one of the categories in a random fashion. The only condition entered in Qualtrics was that both groups should be evenly represented in the sample. This resulted in 138 (50%) out of the 276 participants receiving the social influence treatment.
The social influence treatment incorporated information about what the hypothetical neighbours thought that was appropriate and information about what they actually decided to do. Hence, both the injunctive and the descriptive social norm were included in the treatment. Descriptive as well as injunctive social norms were included because a combination of norms can lead to larger behavioural changes than using either one of the norms on its own (Cialdini, 2003). In the social influence treatment group the information about the investment decisions included the following sentences: “Your neighbours approve of investments in insulation/solar panels and consider the promotion of the municipality a good deal. The majority of them is enthusiastic about the promotion and have already signed up.”. To stress the social influence aspect the participants were reminded that the majority of their neighbours invested when they were about to make their own investment decisions. In the control condition exactly the same information about the energy retrofit measures was provided, the only difference being that information about the decisions and thoughts of neighbours was absent. Participants received both decisions either with or without social influence. The reasoning behind this was to avoid confusion among the participants and to make sure that the social influence effect did not unintentionally influence subsequent decisions in which social influence was not applied.
In order to check whether participants actually perceived social pressure while making the investment decisions a control was added at the end of the experiment in which participants had to rate the following statement on a 7-point Likert scale ranging from strongly disagree to strongly agree: “I experienced social pressure from my hypothetical neighbours during the experiment.”.
15
to strongly agree how much the following statements would apply to them if they assumed that they had graduated from university, found a job with a stable income, owned a house and had a decent amount of savings: (1) “People who are important to me think that I should insulate/install solar panels”; (2) “People who influence my behaviour think I should insulate/install solar panels”; and (3) “People whose opinions I value prefer that I insulate/install solar panels”. These questions showed good internal consistency and could therefore be grouped into an injunctive norm for insulation (α = .855) and an injunctive norm for solar panels (α = .888).
A One-Way ANOVA was carried out to check for differences between the reported injunctive norms of people in the social influence and the no social influence treatment groups. The reported injunctive norm for insulation by participants in the social influence condition (M = 4.50, SD = 1.26) was not significantly different from the one reported by participants in the control condition (M = 4.78, SD = 1.20), F(1,274) = 3.73, p = .054. Similarly, the reported injunctive norm for solar panels by participants in the social influence condition (M = 4.46, SD = 1.26) was not significantly different from the one reported by participants in the control condition (M = 4.40, SD = 1.35), F(1,274) = .17, p = .678. As a consequence, it can be inferred that the treatment did not influence the self-reported injunctive norms of participants. This implies that the measured injunctive norms can be used as a predictor of investments in energy retrofit measures separately from the treatment induced social influence.
Finally, the inclusion of the visibility aspect is discussed. The moderator in this study was the visibility of the energy retrofit measure. Solar panels represented the observable energy retrofit measure category whereas insulation represented the unobservable energy retrofit measure category. While solar panels on a roof can be easily observed by neighbours and one’s community, it is more difficult to identify whether parts of a house have been insulated. The visibility versus the invisibility of the measures was stressed in the measure specific texts presented to the participants. For insulation the text stated: “Insulation is not visible in or outside your house” while the descriptive text for solar panels included: “Solar panels are visible on the outside your house”.
In addition to the difference in visibility between the measures, there were several other reasons to opt for the usage of insulation and solar panels as energy retrofit measures in this study. Both measures for instance tend to be known among the general public which made them relatively easy to grasp for participants. Hence, it was unlikely that decisions not to invest were motivated by unfamiliarity with the energy retrofit measure.
16
required an investment of €4990. The investment rates could have been influenced if participants perceived differences in the value offered by the measures at a price of €4990. The perceived price values of the energy retrofits were therefore measured. The measurement scales were taken from Girod et al. (2017) and adjusted to the presented energy retrofit measures. To derive the perceived price value for insulation the following questions were posed: (1) “Insulation is reasonably priced.”; (2) “Insulation is good value for money.”; and (3) “At the current price, insulation offers good value.”. Participants had to indicate to what extent they agreed to each of the statements on a 7-point Likert scale ranging from strongly disagree to strongly agree. Similar questions were included for solar panels. Both the questions for the perceived price value of insulation as well as the questions for the perceived price value of solar panels showed good internal consistency and could therefore be grouped into one price value for insulation measure (α = .831) and one price value for solar panels measure (α = .842). To test whether the perceived price value from insulation (M = 4.94, SD = .98) differed from the perceived price value of solar panels (M = 5.00, SD = 1.07), a dependent samples t-test was performed. The results indicate that the means were not significantly different from one another, t(275) = -.94, p = .350. This implies that it is unlikely that a difference in perceived value for money affected the investment decisions.
4.1.2 Influencing factors
The influencing factors included in this study that represent the “attitude toward behaviour” branch of the TRA model were taken from Faiers and Neame (2006), Achtnicht and Madlener (2014), Qiu et al. (2014), Frederiks et al. (2015), and Girod et al. (2017). How these household characteristics, home characteristics and retrofit measure characteristics were included in the experimental design is discussed in this sub-section. First, the inclusion of the household characteristics is discussed. Secondly, home characteristics are elaborated upon. Finally, the incorporation of retrofit measure characteristics in the present study is reviewed.
Environmental norms and personal innovativeness. In response to the call from Girod et al. (2017) to
17
extent they agreed to each of the statements on a 7-point Likert scale ranging from strongly disagree to strongly agree. All statements can be reviewed in Appendix A.
Income and funding. The current study controlled for the potential effect of income and funding through
the assumptions that participants had to make. Participants were told that they were graduated from university, had a job with a stable income and that they possessed a decent amount of savings.
Risk aversion. As Qiu et al. (2014) stated that more risk averse people are less likely to invest in energy
efficient technologies, this variable has been taken into consideration in the present study. The measure used in the experiment to test the risk aversion of participants has been adjusted from Holt & Laury (2002). They developed a lottery-choice experiment that provides the opportunity to measure participants’ degree of risk aversion. The measure has been applied to the context of home energy retrofits as several scholars indicated that one’s degree of risk aversion may differ between situations (e.g., Barseghyan et al., 2011; Einav et al., 2012; Reynaud & Couture, 2012). Participants were presented with 10 investment decisions. They were told that they had to purchase new window frames. In each of the 10 decision rows they had to choose between option A and option B, where option A had the more certain maintenance cost savings whereas option B had greater potential maintenance cost savings. They were also told that they should assume that the window frames were identical in every way, except for the potential maintenance cost savings. The purchase of window frames was chosen because this fitted the context of investing in home improvement while it was also clearly distinguishable from the energy retrofit decision making tasks that the participants had already performed by the time that they arrived at the risk aversion test. The risk aversion task can be reviewed in Appendix A. Risk neutral people tend to switch from option A to option B at the 5th decision row whereas risk seeking people tend to switch sooner and risk averse people later. Based on when participants switched rows they were classified on a scale ranging from 1: “highly risk loving” to 9: “stay in bed” (Holt & Laury, 2002).
Heterogeneous opinions. To mitigate the effect of heterogeneous opinions within a household, participants
had to assume that they lived by themselves.
Renovation required. In the current study, nothing about the condition of the house was mentioned to
18
Renovation not required. The fact that participants could invest in energy retrofit measures in the current
study signaled to participants that the house in the presented scenarios was not already energy-optimized. Moreover, as explained above, details about the condition of the house were left out of scope in the current study.
Home ownership. Home ownership has been controlled for through the assumptions that participants had
to make. Participants were told that they had to assume that they owned their own house. It was important to control for home ownership because the study was conducted among a student population. Students mostly live in rental housing and they often pay rent including expenses for energy consumption, which makes them a category that generally does not tend to invest in energy retrofit measures.
Energy costs savings potential, uncertain cost savings, initial investment, comfort, and aesthetics. Energy
cost savings potential, uncertain cost savings, initial investment, comfort, and aesthetics have been included in the study design via statements. Participants were asked to indicate to what degree these statements motivated their investment decisions. The statements were presented to the participants in a random order. An example of such a statements is: “That insulation can reduce my gas bill influenced my insulation decision.”. Participants had to indicate on a 7-point Likert scale ranging from strongly disagree to strongly agree how much each statement influenced their decision to invest or not to invest. All statements can be reviewed in Appendix A.
4.2 Participants
The number of participants in the experiment was 276 (140 male and 136 female). The sample comprised of students from the University of Groningen who completed an experiment in exchange for credit toward a course requirement or a monetary reward. In addition to this, participants were incentivised through the possibility of earning an extra monetary reward of 0 to 4 euros. This additional payment was based on the decisions that participants made during the experiment and the consequences that followed these decisions. Those consequences were randomly assigned by Qualtrics, the online software tool used to operationalise this experiment. One hundred and thirty-two participants in the sample participated for a monetary reward (47.8%), while 144 participants participated for research credits (52.2%). Participation in this study was on a voluntary basis. Participants could sign up via SONA, an online platform. The sample consisted of 181 bachelor students (65.6%), 51 pre-master students (18.5%), 43 master students (15.6%), and 1 PhD student (0.4%).
19
generalizability of results based on student samples is thus questionable. However, in the current study the usage of a student sample could potentially be an advantage. As students are generally not very embedded and active in their neighbourhoods they do not tend to have many positive or negative associations in mind about their neighbours. Hence, they might be more open toward the concept of hypothetical neighbours. This is favourable, as the present study attempts to study the effect of social influence in general and not the effect that neighbour X or neighbour Y might have on a participant’s behaviour.
4.3 Procedure
The experiment ran in the Faculty of Economics and Business research lab from the 3th of December 2018 until the 14th of December 2018. When students registered online for the experiment they were provided with the following information:
“Household and business decision simulation
This study consists of two parts, in the first you will have to make investment decisions, in the second part your task will be to manage a virtual small business and answer some questions about how you did.”
The study was combined with another study that had an independent research objective. As little information as possible was provided in the study description to avoid a self-selection bias. Hence, no information about energy retrofit measures was included in the description to make sure that interest in such measures, or climate related issues in general, could not have biased the sample. Once participants were signed up for a specific time slot and came to the research lab they were informed that the investment decisions were about the adoption of home energy retrofits. It was also mentioned that questions about what motivated or influenced their decisions were going to be posed. Moreover, participants were told that they could earn an additional payment between €0 and €4 based on the decisions that they made during the experiment and the randomly generated consequences of these decisions. In addition to this, they were notified that their personal details would be processed anonymously, handled confidentially, and solely used for scientific purposes. When participants had read this information, they were asked to sign to register their consent.
20
conduct this experiment with a student population. The assumptions were repeated at every investment decision that participants had to make.
Participants were randomly assigned to one of the following four treatment groups: (1) control + insulation decision first (24.6%); (2) control + solar panel decision first (25.4%); (3) social influence + insulation decision first (25.0%); or (4) social influence + solar panel decision first (25.0%). The only condition entered into Qualtrics was that the groups should be evenly represented in the final sample. After making the investment decisions participants were asked to rate how much potential motivations and barriers influenced their decisions. Then they continued with the risk aversion test. Subsequently, the consequences from their investment decisions were presented. These were presented after the risk aversion test because a positive or negative financial outcome could otherwise have influenced the risk perception of the respondents. After the risk aversion test, information on the remaining influencing factors and demographics was gathered. Participants completed the entire experiment in about 30 minutes. As the experiment was matched with another experiment, participants then moved on to a second, unrelated, experiment of about 30 minutes. Before the start of the second experiment participants were provided with a written debriefing and thanked for their participation.
4.4 Data analysis
21
5. RESULTS
5.1 Descriptive statistics
The descriptive statistics for the variables included in the analyses can be reviewed in the following tables. Table 1 presents the descriptive statistics and correlations for the insulation investment decision. In Table 2 the descriptive statistics and correlations for the solar panel investment decision can be reviewed. Moderately high correlations can be observed between the variables in the retrofit measure characteristics category. As the variables “gas/electricity cost savings potential”, “initial investment”, and “uncertain gas/electricity cost savings” are all associated with monetary aspects of investing in energy retrofit measures, it is not extraordinary that they are slightly correlated with one another. In addition to this, most variables gathered information on what motivated participants’ investment decisions and the majority of the variables were recorded on a 7-point Likert scale. This may explain the moderately high correlation between “increased thermal comfort” and the other retrofit measure characteristic variables. Both the opportunity to save on one’s gas bill as well as the possibility to increase one’s thermal comfort may for instance motivate people to invest in insulation.
Table 1: Descriptive statistics insulation investment decision
Mean SD Min Max
S o cial in flu en ce trea tme n t In ju n cti v e n o rm En v ir o n m en tal n o rm s P erso n al in n o v ati v en ess Ga s co st sa v in g s p o ten ti al Un ce rtain g as co st sa v in g s In it ial in v estm en t Social influence Social influence treatment .50 .50 0 1 Injunctive norm 4.64 1.24 1 7 -.116 Household characteristics Environmental norms 5.24 1.07 1 7 .009 .260** Personal innovativeness 4.57 1.26 1.33 7 -.081 .023 .031 Retrofit measure characteristics
Gas cost savings
potential 5.61 1.30 1 7 -.048 .205** .141* -.054 Uncertain gas cost
savings 3.92 1.67 1 7 .076 -.208** -.107 -.120* -.283**
Initial investment 4.47 1.55 1 7 .068 -.133* -.045 -.042 -.078 .439** Increased thermal
comfort 5.56 1.45 1 7 -.050 .237** .123* .033 .519** -.484** -.233** * Correlation is significant at the 0.05 level (2-tailed)
22 ** Correlation is significant at the 0.01 level (2-tailed)
Table 2: Descriptive statistics solar panel investment decision
Mean SD Min Max
S o cial in flu en ce trea tme n t In ju n cti v e n o rm En v ir o n m en tal n o rm s P erso n al in n o v ati v en ess El ec tri cit y c o st sa v in g s p o te n ti al Un ce rtain elec tri cit y c o st sa v in g s In it ial in v estm en t Social influence Social influence treatment .50 .50 0 1 Injunctive norm 4.43 1.30 1 7 .025 Household characteristics Environmental norms 5.24 1.07 1 7 .009 .174** Personal innovativeness 4.57 1.26 1.33 7 -.081 .106 .0.31 Retrofit measure characteristics
Electricity cost savings
potential 5.45 1.54 1 7 .061 .316** .074 .051 Uncertain electricity cost
savings 4.28 1.79 1 7 .041 -.385** -.188** -.166** -.331**
Initial investment 4.61 1.58 1 7 .048 -.170** -.091 .010 -.115 .461**
Aesthetics 3.27 1.72 1 7 -.028 .113 -.047 .068 .082 .017 -.010
Table 3 presented below indicates the percentage of participants that decided to invest in insulation and in solar panels. A distinction is made between participants that received the social influence treatment and participants that were part of the control condition.
Table 3: Treatment * Investment decisions cross tabulation
Decision
Insulation Solar panels Treatment % that invested within control condition 84.1% 71.7%
% that invested within social influence condition 83.3% 71.7%
Total % of total 83.7% 71.7%
5.2 Regression outcomes
Before the results are examined it has to be checked whether the treatment resulted in statistically significant differences in perceived social influence between the treatment group (M = 2.76, SD = 1.76) and the control group (M = 2.18, SD = 1.27). A One-Way ANOVA determined that there was a statistically significant difference between the groups, F(1,273) = 9.76, p = .002.
23
from four binary logistic regressions have to be combined as the experiment included social influence via two approaches and gathered data on two energy retrofit measures.
First of all, the insulation decision is examined. The models that resulted from the binary logistic regressions with the insulation investment decision are presented in Table 4. Model 1 of Table 4, which includes control variables based on household characteristics and energy retrofit measure characteristics, is statistically significant, χ2(6, N = 276) = 161.38, p < .001. The model correctly classifies 94.6% of cases. In Model 2a of Table 4 the social influence treatment is added to Model 1. This step proved to be insignificant, χ2(1, N
= 276) = .46, p = .497. Hence, from the current dataset cannot be inferred that the social influence treatment is a predictor of the insulation decision, B = .41, Wald χ2(1) = .46, p = .499. Its inclusion in the binary
logistic regression also does not improve the classification accuracy of Model 1. In Model 2b of Table 4 the injunctive norm is added to Model 1. This step proved to be insignificant as well, χ2(1, N = 276) = .39,
p = .533. Hence, from the current dataset cannot be inferred that the injunctive norm is a predictor of the insulation decision, B = .17, Wald χ2(1) = .39, p = .530. Its inclusion in the binary logistic regression also
does not improve the classification accuracy of Model 1.
Secondly, the solar panel decision is examined. The models that resulted from the binary logistic regressions with the solar panel decision as dependent variable are presented in Table 5. Model 1 of Table 5, which includes control variables based on household characteristics and energy retrofit measure characteristics, is statistically significant, χ2(6, N = 276) = 201.72, p < .001. The model correctly classifies 90.6% of cases. In Model 2a of Table 5 the social influence treatment is added to Model 1. This step proved to be insignificant, χ2(1, N = 276) = .91, p = .341. Hence, from the current dataset cannot be inferred that the social influence treatment is a predictor of the solar panel decision, B = -.45, Wald χ2(1) = .89, p = .345. Its inclusion in the binary logistic regression improves the classification accuracy with .3% in comparison to Model 1. In Model 2b of Table 5 the injunctive norm is added to Model 1. This step proved to be significant, χ2(1, N = 276) = 4.00, p = .046. Hence, from the current dataset can be inferred that the injunctive norm is
a predictor of the solar panel decision, B = .40, Wald χ2(1) = 3.88, p = .049. However, its inclusion in the binary logistic regression does not improve the classification accuracy of Model 1.
24
Table 4: Binary logistic regression models insulation investment decision
MODEL 0 MODEL 1 MODEL 2a MODEL 2b
B SE Sig Exp(B) B SE Sig Exp(B) B SE Sig Exp(B) B SE Sig Exp(B)
Constant 1.636 .163 .000 5.133 1.449 2.708 .592 4.261 .948 2.815 .736 2.580 1.164 2.732 .670 3.202
Social influence
Social influence treatment .409 .605 .499 1.506
Injunctive norm .170 .271 .530 1.185 Household characteristics Environmental norms .123 .274 .653 1.131 .113 .277 .684 1.119 .082 .279 .769 1.085 Personal innovativeness -.187 .248 .452 .830 -.142 .256 .580 .868 -.234 .261 .371 .792 Retrofit measure characteristics
Gas cost savings potential 1.095 .256 .000 2.990 1.100 .254 .000 3.003 1.070 .257 .000 2.915 Uncertain gas cost savings -1.439 .327 .000 .237 -1.430 .322 .000 .239 -1.445 .328 .000 .236
Initial investment -.038 .220 .863 .963 -.035 .216 .870 .965 -.016 .223 .943 .984
Increased thermal comfort .457 .192 .017 1.579 .467 .195 .016 1.596 .464 .191 .015 1.590
-2 Log likelihood 84.086 83.625 83.698
χ2
(6,N=276)=161.378, p<0.001 χ2(7,N=276)=161.839, p<0.001 χ2(7,N=276)=161.767, p<0.001
Cox & Snell R2 .443 .444 .444
Nagelkerke R2 .752 .753 .753
25
Table 5: Binary logistic regression models solar panel investment decision
MODEL 0 MODEL 1 MODEL 2a MODEL 2b
B SE Sig Exp(B) B SE Sig Exp(B) B SE Sig Exp(B) B SE Sig Exp(B)
Constant .932 .134 .000 2.538 -.053 1.867 .977 .948 .125 1.877 .947 1.133 -1.329 1.991 .504 .265
Social influence
Social influence treatment -.447 .473 .345 .640
Injunctive norm .398 .202 .049 1.489 Household characteristics Environmental norms .046 .219 .835 1.047 .054 .220 .804 1.056 .004 .222 .984 1.004 Personal innovativeness .021 .190 .913 1.021 -.005 .192 .981 .995 -.037 .200 .853 .964 Retrofit measure characteristics
Energy cost savings potential 1.438 .227 .000 4.212 1.468 .233 .000 4.339 1.458 .237 .000 4.297 Uncertain energy cost savings -1.226 .215 .000 .293 -1.233 .216 .000 .291 -1.168 .218 .000 .311
Initial investment -.188 .170 .269 .829 -.197 .169 .245 .821 -.229 .170 .180 .796
Aesthetics .042 .132 .749 1.043 .060 .133 .654 1.061 .007 .135 .961 1.007
-2 Log likelihood 126.939 126.032 122.941
χ2
(6,N=276)=201.722, p<0.001 χ2(7,N=276)=202.629, p<0.001 χ2(7,N=276)=205.720, p<0.001
Cox & Snell R2 .519 .520 .525
Nagelkerke R2 .745 .747 .755
26
5.3 Cross tabulation and Chi-square test
Hypothesis 2 predicts that the visibility of energy retrofit measures positively moderates the relationship between social influence and the adoption of energy retrofit measures. A cross tabulation and a Chi-square test are used to test hypothesis 2. The results of the cross tabulation are presented in Table 6. From the table can be derived that more participants decided to invest in insulation (83.7%) than in solar panels (71.7%) while hypothesis 2 predicted the opposite. A Chi-square test of independence was performed to examine the relation between the insulation investment decision and the solar panel investment decision. The relation between these variables was insignificant, χ2(1, N = 276) = 1.41, p = .235. As participants were not more likely to invest in solar panels, the visible energy retrofit measure, than in insulation, the invisible energy retrofit measure, hypothesis 2 must be rejected.
Table 6: Insulation decision * solar panel decision cross tabulation
Solar panel decision
No Yes Total
Insulation decision No Count 16 29 45
% within insulation decision 35.6% 64.4% 100.0% % within solar panel decision 20.5% 14.6% 16.3%
% of total 5.8% 10.5% 16.3%
Yes Count 62 169 231
% within insulation decision 26.8% 73.2% 100.0% % within solar panel decision 79.5% 85.4% 83.7%
% of total 22.5% 61.2% 83.7%
Total Count 78 198 276
% within insulation decision 28.3% 71.7% 100.0% % within solar panel decision 100.0% 100.0% 100.0%
% of total 28.3% 71.7% 100.0%
6. DISCUSSION
27
influencing factors is addressed. The section concludes with a discussion of the policy implications that can be derived from this study. Throughout the discussion limitations and suggestions for future research are brought up.
6.1 Social influence effect
In this sub-section hypothesis 1 is discussed. That hypothesis 1 was rejected and the predicted relationship between social influence and the adoption of energy retrofit measures was not found can imply three things. First of all, it could be the case that social influence simply does not have an effect on the adoption of energy retrofit measures. As indicated by Mulder (2018) there are numerous factors that influence one’s adoption decision. Some of these factors might have a more salient effect on the adoption decision than a potential effect of social influence. However, concluding that there is no effect would contradict findings from serval scholars (e.g., Kallgren et al., 2000; Allcott, 2011; Curtius et al., 2018) that found a relationship between subjective norms and the adoption of pro-environmental behaviour. A limitation of these studies is that they oftentimes study the effect of social influence in isolation, neglecting other potential factors of influence. It is therefore advisable to include more rigorous controls in future studies on the effect of social influence on the adoption of pro-environmental behaviour to investigate if the relationships also holds after the inclusion of controls.
A second explanation to why the expected relationship was not found could be that the social influence was not perceived as social influence by participants and that the inclusion of social influence in the treatment was flawed. Although the self-reported perceived social influence by the treatment group was relatively low (M = 2.76, SD = 1.76), one cannot automatically conclude that the social influence treatment failed. Nolan et al. (2008) for instance argue that the normative messages remain unnoticed while they do affect behaviour. They found that in comparison to emphasising other reasons to conserve energy, normative social influence resulted in the largest behavioural change, even though respondents reported that the normative information motivated their behaviour the least.
28
Another potential flaw in the treatment could stem from the fact that the nature of the relationships among the hypothetical neighbours was not specified in the treatment. Warren and Clifford (1975) for instance state that in integrated neighbourhoods energy saving measures are spread faster in comparison to neighbourhoods where people interact less with one another. Moreover, Schultz and Tyra (2000) find that people are more likely to adopt the behaviour of people that are close to themselves. Participants in their study that received information about the recycling behaviour of their neighbours were more likely to increase their own recycling efforts than participants that received information about the recycling behaviour of people in their neighbourhood or information on city wide recycling rates. As no picture was painted about the relation to one’s hypothetical neighbours in the current study, participants might have had difficulties to identify with their neighbours. As a consequence, they may not have perceived this information as a source of social influence. In future research the social cohesion of (hypothetical) neighbourhoods and communities should be taken into account and network theory could be applied. The way the social influence was induced and framed could also provide reasoning for the low perceived social influence. In the current study design social influence has been induced via information provision on social norms while Abrahamse and Steg (2013) argue that public commitment or using block leaders are more effective ways to induce social influence. In addition to this, the experiment talked about hypothetical neighbours instead of actual neighbours and incorporated written text instead of spoken text while Abrahamse and Steg (2013) argue that face-to-face interactions intensify the effect of social influence. Moreover, participants would never meet their hypothetical neighbours or interact with them again after the experiment while this is not a proper reflection of reality. Furthermore, the text about one’s hypothetical neighbours was incorporated in a relatively long description that included information on all kinds of factors of influence, which may not have resulted in the aspired prominent role of social influence in the minds and decision considerations of the participants.
In future research the different approaches to induce social influence should be compared with one another to determine which are most suitable for, and applicable to, the energy retrofit investment context. While the inclusion of block leaders in policies and study designs is for instance difficult, public commitment could be suitable for group laboratory studies. Group studies also provide the opportunity to include face-to-face interactions. Examples of policies based on public commitment in a field setting could be to register and visualize residents’ commitment in community centres or to attach a certain distinguishable feature to the homes of committed residents.
29
colleagues. Bale et al. (2013) found that 40-50% of the people in their sample discussed energy usage issues with family members, friends or colleagues, which increases the likelihood that social pressure will be perceived. In addition to this, more elaborate pre-tests should be conducted in order to avoid running a study with a potentially flawed treatment. This study has been pre-tested by a limited amount of people that were already familiar with the research objectives of the study. As a result, potential flaws in the treatment did not surface in the pre-testing stage of the experimental design.
A field experiment with the same research objective is currently running to overcome the limitation of not including actual neighbours in the study. In this field experiment a semi-governmental organization that installs energy retrofit measures induces the social influence. When they install solar panels, carry out insulation measures or place a heat pump at an address in the treatment area, ten neighbours on each side of the dwelling and on the opposite side of the street receive a note that their neighbours are having energy retrofit measures installed. In addition to this, a board is temporarily attached to a window of the dwelling where the measure is carried out. Findings from the lab experiment could also be validated through this field experiment. Moreover, the generalizability of the results could be extended beyond the student population on which the current study is based. A drawback of this field study is that numerous factors cannot be controlled. Hence, it would be advisable to combine the results of this field study with a properly designed lab experiment.
30
6.2 High adoption rates
In this sub-section the relatively high adoption rates of energy retrofit measures in the experiment are discussed. Most participants decided to invest in either one (33.0%) or both (61.2%) of the energy retrofit measures presented to them. These results are not a proper reflection of investments in energy retrofit measures in reality. An explanation for this finding could be that people responded to the investment decisions in a socially desirable manner while they might not invest in reality. Moreover, it could be that the risk embedded via the payment structure was not severe enough. The possibility of earning or losing €0.50 or €1 per investment decision may not have mattered that much to the participants as they had the certainty that they would receive at least the standard monetary payment or research points for their participation. Hence, the incorporation of monetary consequences for participants might not have been representative of the actual costs of investment.
This would also explain the contradiction with the risk aversion literature as the participants tended to be relatively risk averse (M = 5.57, SD = 1.57) while most of them decided to invest in the energy retrofit measures as well. Both the possibility that participants responded in a socially desirable manner and the relatively small monetary consequences stress the importance of studying actual adoption. It is expected that much lower adoption rates will result from the field experiment. Although the current study included monetary consequences for participants and thus simulated investment, this field experiment is an even more appropriate response to the call of Curtius et al. (2018) for research in the social influence domain that focuses on actual adoption.
The favourable investment conditions could also provide an explanation for the high adoption rates. Participants assumed that they had a job with a stable income and a decent amount of savings, all of which are not a given for many people. In addition to this, the investment decisions were presented in isolation of other ways to spend one’s savings while in reality there are often opportunity costs associated with investments in retrofit measures. In future research one could include energy retrofit measures among a list of other investment options to determine whether participants would stick to their green investments or prefer to spend their (hypothetical) savings otherwise when granted the opportunity. In a lab experiment setting this could for instance be operationalised by offering participants the opportunity to donate €X to a subsidy fund for investments in energy retrofit measures or to have the opportunity to win a travel voucher of €Y. Besides the financial barrier that was lifted in this experiment, most of the other influencing factors were also simplified while all of these might pose barriers to investment in reality.
31
invest in energy efficiency measures than older people (>55) that have solely received primary education. However, as findings on the effect of age and education on the adoption of energy retrofit measures show inconsistent results, these factors were not considered in the theoretical discussion on household characteristics and thus not included in the conceptual model. Contrary to Nair et al. (2010), Barr et al. (2005) for instance argue that older people are more likely to be energy savers and Frederiks et al. (2015) state that education level is not a predictor of pro-environmental behaviour.
6.3 Absence of visibility effect
In this sub-section is discussed why support for hypothesis 2 was not found. Contrary to what was predicted by hypothesis 2 more participants invested in insulation, the invisible energy retrofit measure, than in solar panels, the visible energy retrofit measure. This could stem from an aversion towards the usage of gas because gas extraction caused earthquakes in the province of Groningen. Whether reduced dependence on one’s gas provider motivated people to invest in insulation has been added to the controls. As this control does not have a strong theoretical foundation it has not been included in the conceptual model. To test whether one’s dependence on a gas provider (M = 4.66, SD = 1.66) influenced the insulation decision more than one’s dependence on a electricity provider (M = 4.62, SD = 1.71) influenced the solar panel decision, a dependent samples t-test was performed. The results indicate that the means were not significantly different from one another, t(275) = .42, p = .674. This implies that an aversion towards the usage of gas is probably not the cause of the observed difference in energy retrofit measure investment.
32
“Personal preferences about the fact that I am often feeling very cold.”
“Personally, I do feel a lot of cold in winter times, so for me to have a warm home it is essential
for my well-being.”
“I assumed the house is located in the Groningen area. For what I have seen in the past months,
it is cloudy/rainy very often.”
“That the weather in the Netherlands is bad and not many sun hours are available during the
year influenced my solar panel decision”
Faiers and Neame (2006) argue that some people do not want to invest in solar panels because they dislike the aesthetics of solar panels. However, from the results of this study cannot be derived that this motivated participant’s decisions not to invest in solar panels as the aesthetics control variable is insignificant in all solar panel models. Participants also tended to disagree to the statement that the aesthetics of solar panels influenced their solar panel decision (M = 3.27, SD = 1.72).
While hypothesis 2 cannot be confirmed with the results of this study, it cannot be inferred that the effect is generally non-existent either. As mentioned before, the results of the binary logistic regressions with the injunctive norm even suggest that there might be an effect. This finding is in line with the results from Curtius et al. (2018) that found that injunctive norms increase people’s intention to adopt solar panels. In future studies the visibility hypothesis should be tested with multiple visible and invisible energy retrofit measures to verify that other attributes of the measures are not causing potential differences in investment. In addition to this, the investment decision descriptions should be pre-tested more thoroughly to make sure that the differences between the measures are minimized and the focus remains on the predictor variables. Other assumptions that people have about the measures and the descriptions should be addressed in the descriptive text or taken into account via the inclusion of control variables.
33
there are no measure specific differences between the visible and invisible energy retrofit measures. This leaves little room for alternative explanations such as the potential weather effect discussed above.
6.4 Influencing factors
In this sub-section the effect of the influencing factors on the adoption of the presented energy retrofit measures is discussed. Several household characteristics and retrofit measure characteristics were included as controls in this study because literature suggested that they could influence energy retrofit measure adoption decisions. However, from the binary logistic regression models can be inferred that economic motives seem to predominate. The energy cost savings potential and the uncertain energy cost savings were for insulation as well as for solar panels strong predictors of investment. The only other statistically significant predictor for insulation was the possibility to improve the thermal comfort of the dwelling, which may be due to a weather effect as is explained above. The importance of economic motives could stem from several sources.
First of all, it could be the case that not all of the included variables influence investment in capital intensive energy retrofits. Girod et al. (2017) for instance show that environmental norms and personal innovativeness impacted the intention to adopt an intelligent thermostat. However, intelligent thermostats require significantly less investment than insulation and solar panels. In addition to this, the purchase of an intelligent thermostat is a one-off investment that cannot be recouped via direct energy cost savings, while the energy cost savings potential is a factor of importance for investments in insulation and in solar panels. Hence, future research should examine whether economic motives become stronger predictors in comparison to other potential motives if the required initial investment or the energy cost savings potential becomes larger. Findings from Gamtessa (2013) suggest that this might be the case. Similar to this study, he found that financial incentives are important predictors of investments in energy efficiency measures. In addition to this, he found that households are more likely to adopt energy retrofit measures when the expected financial gains are high.
