Master Thesis: Energy Transition
The impact of social influence on retrofit
adoption and household acceptance
Name: Ajjoub Rifi
Student number: S2541750
Master: MSc Business Administration
Track: Strategic Innovation Management (SIM)
Supervisor: Dr. Eelko Huizingh
Co-assessor: Prof. dr. Machiel Mulder
2
Abstract
Household acceptance of energy technologies plays an important role in adopting retrofit measures. This study investigates the relations of social influence, perceived usefulness and perceived ease of use with the investment in retrofit to guide future policies. Results from a randomized controlled lab experiment show that written normative messages do not influence individuals in the adoption of retrofit, and that perceived usefulness does influence individuals’ adoption. Therefore, policies should focus on increasing the usefulness of retrofit for households by information campaigns. Future research should study interactional social influence with regard to investments in retrofit, and developing a tailored technology acceptance model for retrofit measures.
1. Introduction
The first steps towards the energy transition started in the 1960s, it received since a lot of societal, political and scientific interest (Verbong & Geels, 2007). The current energy transition is defined as: the shift from high carbon emitting, non-renewable energy systems (i.e. dominated by fossil fuels) to low carbon-emitting (or carbon neutral), renewable energy systems (i.e. dominated by non-fossil fuels) (Meadowcroft, 2009). Due to the Paris agreement, the energy transition has gained more momentum and became of great importance for policymakers and academics. Based on the Paris agreement of 2015, the Dutch government has set goals to limit the increase of the global temperature below 2 degrees Celsius this century, by reducing 20% of CO2 emissions in 2020, and 49% CO2 reduction in 2030 compared to the 1990s (Rijksoverheid, n.d.; United Nations, n.d.).
3
Besides the value-action gap, governmental policies in the Netherlands also show insufficient results in order to persuade households into action (e.g. Beerepoot & Beerepoot, 2007; Murphy, Meijer & Visscher, 2012; Vringer, van Middelkoop & Hoogervorst, 2016). Murphy et al. (2012) even state that households who make use of these policies are still poorly understood. Mulder (2018), shows that several factors are of importance for the willingness of households to invest in building retrofits: (1) household characteristics, (2) retrofit measure characteristics, (3) home characteristics, and (4) social influence. Retrofit in this context is defined as “the upgrading of the building fabric, systems or controls to improve the energy performance of the property” (Brown, Swan & Chahal, 2014 p. 461).
4
do not accept retrofit, and that therefore the actual adoption of retrofit is relatively low. For policymakers, it is also important to understand whether or not households accept retrofit in the first place. Insights can be useful to shape future policies to effectively influence household adoption. The Technology Acceptance Model (TAM) suggests that Perceived ease of use (hereafter called ‘PEOU’) and Perceived usefulness (hereafter called ‘PU’) can positively influence actual acceptance of a technology (Davis, 1989). To the best of the researcher’s knowledge, those factors are still insufficiently represented in the context of retrofit adoption. The TAM is widely known for its use to study acceptance of technologies in the Information System literature (Venkatesh & Davis, 2000). Although, in the energy transition literature several studies exist which explore the TAM for technology acceptance, but not in the case of retrofit measures (e.g. Huijts, et al., 2012; Broman Toft, Schuitema, Thogersen, 2014; Heesen & Madlener, 2014). Furthermore, technology acceptance is also studied in combination with social influence (see Venkatesh & Davis, 2000; Venkatesh & Morris, 2000). This makes studying social influence, PU and PEOU combined, a valuable contribution to the existing energy technology acceptance and adoption literature.
The goal of this study is to empirically test the effect of social influence on the actual investment in retrofit, for both insulation and solar panels. Besides exploring social influence, the goal is also to discover whether PEOU and PU affect the adoption and acceptance of retrofit. Determining whether or not households accept retrofit in the first place, will give valuable insights in order to guide future policies and future research in the area of retrofit adoption. By addressing the literature gap, I aim to contribute to the field of energy transition and adoption, as well as to the field of the technology acceptance theory within the energy transition. Based on the gap and the goal of this research, the following research question will be answered in this study:
RQ: ‘How can household adoption of retrofit measures guide new policies in order to influence
households in retrofit investments?’
In order to answer the research question, the main question is divided into two sub-questions based on the three main concepts of this research, namely social influence, PU and PEOU:
RQ1a: How does social influence affect the adoption of retrofit measures?
RQ1b: What are the relationships of PU and PEOU with the actual investment in retrofit measures?
5
The gathered data is analysed to explore whether social influence affects the adoption (i.e. investment) of retrofit, and whether PU and PEOU are related to retrofit adoption, in the context of the TAM of Davis (1989). The remainder of this paper is divided into four chapters. First, the literature is analysed in the theoretical background section. Second, the methodology and the data gathering of the lab experiment are explained in the methods section. Furthermore, the results of the lab experiment are presented and discussed. The paper ends with a discussion and a conclusion, including implications and suggestions for future research.
2. Theoretical background
In this chapter a review of the Technology Acceptance literature is presented. First, the general TAM is evaluated and brought into context of the energy transition. Second, the main concepts that are used in this study are explained and theoretically substantiated, including a brief overview of the social influence literature. Lastly, the hypotheses for this study are formulated, and the conceptual model is presented.
2.1. Technology acceptance and general behavioural models
The TAM is a combined model which is derived from behavioural and social psychology models such as the Theory of Reasoned Action (TRA), and the Theory of Planned behaviour (TPB) (Malhotra & Galletta, 1999). The TAM, among others, is very popular to study adoption and acceptance of a technology on the individual level (Broman Toft et al., 2014). The technology acceptance theory addresses the questions of how a technology is perceived by the end user, and how they accept and use this technology (Momani & Jamous, 2017). The TAM can be seen as an extension on the TRA model, specified to the context of technology acceptance, with two additional technology acceptance constructs: PEOU and PU of a technology (Momani & Jamous, 2017).
6
Attitude to the behaviour
Subjective norms
Perceived behavioural control
Behavioural intention Behaviour
Figure 1: The TPB model (Adapted from Ajzen, 1991)
In the context of behaviour in the energy transition, and due to the value-action gap, the general behavioural models are limited to study why individuals accept energy technologies (Davis, Bagozzi & Warshaw, 1989; Momani & Jamous, 2017). The value-action gap suggests that even when people have positive attitudes towards, and even intend to perform a behaviour, that this not always lead to executing that particular behaviour (Frederiks et al., 2015). The TAM offers, respectively to the TRA and TPB, more context by addressing two important additional constructs: PU and PEOU (Momani & Jamous, 2017). PU and PEOU are key constructs to understand why people accept or reject a technology (Marangunić & Granić, 2015; Momani & Jamous, 2017).
2.2. TAM in the energy transition
The TAM is introduced by Davis in 1989, and explains that the intention of an individual to use a system lays in the PU and PEOU (Venkatesh & Davis, 2000). Davis (1989) states that there are many causes for people to accept (or reject) a technology, but the two most important determinants are PU and PEOU. PU can be defined as: “the degree to which a person beliefs that using a particular system would enhance his or her job performance” (Davis, 1989 p. 320). PEOU can be defined as “the degree to which a person beliefs that using a particular system would be free of effort” (Davis, 1989 p. 320). Subsequently, the PU and the PEOU of a technology lead to a certain attitude towards the technology, which leads to an intention and finally to the actual use (and thus acceptance) of a certain technology (Davis, 1989).
7
for the use of heating systems and thermal comfort, while Broman Toft et al. (2014) test the TAM empirically for Smart Grids. Broman Toft et al. (2014) define the elements of the TAM in their research as follows: PU is “the degree to which the use of that particular technology is believed to enhance the achievement of valued goals” (p. 393). PEOU is defined as: “the degree to which use of that particular technology is believed to be easy and effortless” (Broman Toft et al., 2014 p. 393). The PU and the PEOU explain the attitude of an individual towards the technology and the actual use (adoption) of the technology (Broman Toft et al., 2014). Figure 2 shows a schematic representation of the TAM.
Perceived usefulness (PU)
Perceived ease of use (PEOU)
Attitude Intention to use Actual use
Figure 2: TAM (adapted from Davis, 1989)
2.3. Social influence and the TAM
Social influence comes from the fact that under uncertainty, individuals tend to look to social norms, in order to behave accordingly to social situations (Cialdini, 2001). These social norms can be distinguished in two types: injunctive norms and descriptive norms (Cialdini & Goldstein, 2004). Injunctive norms “inform us about what typically is approved/disapproved” (Cialdini & Goldstein, 2004 p. 597) and descriptive norms “inform us about what typically is done” (Cialdini & Goldstein, 2004 p. 597). Subjective norm, as proposed in the behavioural literature (Fishbein & Ajzen 1975; Ajzen, 1991), is often related to the injunctive norms, rather than the descriptive norms (Branscum, Rivera, Fairchild & Fay, 2017). Social influence can be applied in the form of descriptive norms by changing the social norms. These social norms that can be changed are: (1) conformity: “the act of changing one’s behaviour to match the response to others” (Cialdini & Goldstein, 2004 p. 606) and (2) compliance: “a particular kind of response- acquiescence- to a particular kind of communication – a request” (Cialdini & Goldstein, 2004 p. 592). Furthermore, social influence in the form of injunctive norms can be measured, as what behaviour in the context of retrofit adoption is approved or disapproved.
8
influence in the TAM2. They empirically show that subjective norm is positively related to intention, use and image, and that image is positively related to PU. Voluntariness significantly moderates the relation between subjective norm and the intention to use (Venkatesh & Davis, 2000). The most important construct is subjective norm, as it influences the intention, and thus the actual use of a technology. The subjective norm is adopted from the TRA and TPB theories, which already incorporated that subjective norm influences behavioural intention (Venkatesh & Morris, 2000; Momani & Jamous, 2017). Subjective norm is defined as: the perception of what people, who are of importance to the individual, think of certain (in)action or behaviour (Fishbein & Ajzen, 1975; Venkatesh & Davis, 2000). The reason for the inclusion of subjective norm in the TAM2, is that the actual behaviour of an individual can be affected by social influence, even if they are not in favour of that behaviour (Venkatesh & Davis, 2000; Venkatesh & Morris, 2000).
Perceived usefulness (PU)
Perceived ease of use (PEOU)
Attitude Intention to use Actual use Job Relevance Image Output Quality Result demonstrability Subjective norm Experience Voluntariness
Figure 3: TAM 2 (adapted from Venkatesh & Davis, 2000)
Various studies found proof that social influence affects behavioural intention and behaviour significantly in the context of technology acceptance (e.g. Venkatesh & Morris, 2000; Venkatesh & Davis, 2000). The study of Malhotra & Galletta (1999) even suggests that besides the relation between social influence and behavioural intention, that social influence also influences the attitude (and therefore the behavioural intention indirectly). Their empirical results show that social influence is positively related to attitude, but that social influence does not influence behavioural intention directly (Malhotra & Galletta, 1999). The main theoretical contribution of all those studies is that social influence has a positive significant indirect effect on the actual behaviour.
9
attitude and intention do not always lead to actual behaviour (Frederiks et al., 2015), investigating the relation between social influence and attitude and intention will lead to a flawed perception. Therefore, this study will only focus on the relation between social influence and the actual behaviour. Taylor & Todd (1995) suggest that intention is a direct determinant of behaviour and that attitude has an indirect effect on behaviour. Furthermore, they also suggest that intention mediates the relation between attitude and actual behaviour (Taylor & Todd, 1995). Thus, the main assumption for this study is that when the actual investment in retrofit is positive, the attitude and intention are also both positive and preliminary to the actual behaviour of the individual (i.e. actual investment in retrofit). Therefore, I suggest the following hypothesis:
H1: Social influence is positively related to investing in retrofit.
2.4. Perceived usefulness and Perceived ease of use
In their empirical research, Broman Toft et al. (2014) found that PU and PEOU are significantly and positively related to the attitude towards smart-grids in the model. The study of Davis (1989) states that PU and PEOU are both correlated with the use of a technology. Furthermore, the study of Scherer, Siddiq & Tondeur (2019) suggests that PU and PEOU both significantly predict behavioural intentions via attitudes, and that the effect on intention was more extensive for PU than for PEOU. As proposed earlier, for this study, the attitude and intention are out of scope. Hence, this study will only investigate whether PU and PEOU have an effect on the behaviour itself. Moreover, it is assumed that if the behaviour is positive, that the attitude and intention towards the behaviour are both positive and preliminary to the actual behaviour. Therefore, based on the aim of this research to study only investments in retrofit, I suggest the following hypotheses:
H2: The PU of retrofit is positively related to investing in retrofit.
H3: The PEOU of retrofit is positively related to investing in retrofit.
2.5. Factors influencing the adoption of retrofit
10
2.5.1. Household characteristics
Environmental norms are personality specific factors which are the “moral obligation to act” (Girod,
Mayer & Nägele, 2017, p. 417). It suggests that people that highly care for the environment, tend to adopt ‘green’ technologies more easily (Jansson, Marell & Nordlund, 2011; Girod et al., 2017).
Personal innovativeness is also a personality specific factor, but a more general one. It is related to the
fact that people who are more interested in new technologies, will be more eager to use such technologies (Girod et al., 2017).
Risk attitude is also a very important personality specific factor. Individuals that tend to be more risk
averse, are less likely to invest in energy technologies (Fischbacher, Schudy & Teyssier, 2015).
Income is a relevant socio-demographic characteristic that influences someone’s decision to invest in
retrofit (Mulder, 2018). People who have a higher income, tend to invest more in energy efficient measures (Frederiks et al., 2015).
Heterogeneous opinions also influence adoption of retrofit. A household can consist of multiple
individuals, where different individuals can have different opinions about retrofits, as Achtnicht & Madlener (2014) suggest.
2.5.2. Home characteristics: Home ownership
As proposed by Koers (2019), home ownership plays a role in one’s decision to invest in retrofit measures. The study of Frederiks et al. (2015) suggests that people who own a house, are more likely to invest in energy saving measures, compared to people who rent.
2.5.3. Retrofit measures characteristics
11
2.6. Conceptual model
The hypotheses, as proposed in the above sections and their relations, are shown in the conceptual model of this research, which is based on concepts of the TAM2 (see figure 4). This model shows that the independent variables, social influence, PU and PEOU are all positively related to the dependent variable, actual investment in retrofit. Based on the above literature review, four important control variables are used in this research in order to control for the actual investment in retrofit. The control variables environmental norms, personal innovativeness, risk attitude and perceived price value influence individual’s decision to adopt retrofit directly, and are therefore controlled for in this study (see figure 4). All the other mentioned influencing factors in section 2.4 are used in the research design of this study and are explained in the methods chapter.
Social influence
Perceived usefulness (PU)
of retrofit H2
H1
Perceived ease of use (PEOU) of retrofit
Control variables
Environmental norms Personal innovativeness Risk attitude
Perceived price value
+ + Actual investment in retrofit H3 +
Figure 4: Conceptual model
3. Methods
12
3.1. Procedure and participants
For this study, a randomized controlled social experiment is used. The reason to use such a design is that a lot of factors can influence an individual to invest in retrofit. By conducting a randomized lab experiment, the direction of causality can be assured (Burtless, 1995), because all the characteristics of both groups are ought to be identical (Barnow, 2010). In the case of investing in retrofit, several factors can influence this decision, as proposed earlier. By conducting a lab experiment with two groups that are identical (one treatment group, and one control group), the only difference between these groups is the inclusion of social influence. To create exact the same groups, all the influencing factors need to be controlled for, and consequently the difference can be eliminated (Archibald & Newhouse, 1980). Lab experiments offer, compared to field studies, a convenient way to study a specific part of a population, within a specified context, and with factors that are fixed in isolation (Levitt & List, 2007).
3.1.1. Pilot experiment
The study of Koers (2019) showed unrealistic percentages of retrofit adoption, compared to practice. Therefore, the lab experiment as carried out by Koers (2019) is analysed and adjusted for this study. A short analysis of all adjustments can be found in Appendix B. In order to test if the adjustments were effective, a pilot is carried out on the 24th and 25th of April 2019 to test if more realistic adoption percentages could be found. Also, the pilot helped to determine whether social influence was perceived by the participants. The pilot sample size was 22, and 50% received the social influence treatment. A lower adoption rate was found in this pilot, compared to the study of Koers (2019). Insulation had an adoption rate of 55%, and under social influence 64%; solar panels had an adoption rate of 45%, and under social influence 64%. Nevertheless, the perceived social influence of the hypothetical neighbours in the pilot was still marginal (insulation: M = 3.18, SD = 1.720, N = 11; solar panels: M = 3.55, SD = 2.018, N = 11), measured on a seven-point Likert scale. The findings of the pilot resulted in valuable insights for improvements, see Appendix B.
3.1.2. Final experiment
13
possible (see scenario matrix in Appendix A). The consequences of the decisions were randomly assigned by the Qualtrics, based on a probability of 13. From the 319 participants, 224 (70.2%) participated for money and 95 (29.8%) participated for research credits.
The lab experiment of this study was combined with another experiment from the Master HRM. The participants first got a study about online consulting, and directly after finishing, this experiment was presented. The research objectives and contexts of the studies were unrelated, and therefore independent. Upon registering, the student is informed that the study is about online consulting and household adoption of retrofit. After registration online, the participant came to the FEB Research Lab and needed to read and sign the informed consent before participation. This informed consent contains all the important information about the experiment (see Appendix C).
One of the lab assistants explained the details of the experiment to the participant (i.e. that the experiment contained two different parts, that at the end the extra payment is displayed on the screen, and that the participant needs to contact the lab assistant so the extra payment is noted), and administrated the process. Qualtrics randomly assigned the participants into one of the following four groups, see table 1:
Table 1: Assignment of the conditions
Condition Decision order Number of participants Percentage
Control group 1. Insulation 2. Solar panels
79 24.76%
Control group 1. Solar panels 2. Insulation
81 25.37%
Social influence treatment 1. Insulation 2. Solar panels
79 24.76%
Social influence treatment 1. Solar panels 2. Insulation
80 25.08%
As presented in table 1, approximately 25% of the total participants were assigned to one of the four groups. In total, 50.2% of the participants are part of the control group without social influence, and 49.8% of the participants are part of the treatment group.
3.2. Experimental design
In this section the experimental design is explained. The experiment contains three parts: the decision-making part, including the questions about what influenced or motivated the participant’s decision, the (control) variables part, and the demographics part (which includes treatment checks).
14
text is used to frame all participants into the same starting point, and to exclude the possible effect of factors that could influence a participant’s decision. After the participant was framed, he or she needed to make the investment decision, and to take into account all the factors mentioned in the situation text. This set up is chosen (first situation text, then decision making) because the participant needs to be framed in the controlled situation, which is equal to all participants. Starting with the control variables or the other independent variables would have influenced the decision negatively, and could create a biased and therefore an uncontrolled experiment. First, the general assumption is presented that the participant has graduated from university, has a job with a stable income, has enough savings to afford retrofit, and lives by themselves in a townhouse (that they possess).
Next, the price of the retrofit measure and a brief explanation about the retrofit measure, the advantages, the requirements for investing in retrofit, and other factors of importance are presented. The information is presented in text form with bullet points in order to clarify the situation (see Appendix A). The text contains several factors of influence (personal factors, retrofit measure factors and other factors) which are fixed for all participants, and are adopted from the experiment of Koers (2019). For a complete list of all the fixed factors accounted for in the experiment, see Appendix D. Next, the information is partly summarized in a table, which also contains the possible gains or losses if invested (or not), and the pay outs for the experiment. In order to make the decision uncertain, there was a chance of one of three possible outcomes, which were related to the weather in the next years. These scenarios are related to the outdoor temperature in the winters (cold winter, normal winter, warm winter) for insulation, and to the amount of sun during the day (sunny days, normal days, cloudy days) for solar panels. The decision to invest was, just as in the real world, more uncertain than the decision to not invest. Participants could get +€0.50, €0 or -€0.50 if not invested, and +€0.75, €0 or -€1.00 if invested. The possible pay out if invested is made more uncertain by the possibility of losing more money than you can gain on your investment. These possible pay outs per scenario were presented in the scenario matrix, including the fictional possible return on investment. The not investing decision was presented first in the matrix, for the reason that it is the status quo (i.e. a house without retrofit measures). Qualtrics drew one of three possibilities (probability 1
3), based on the participant’s decision (investing or not investing).
The matrix and the information as presented to the participant can be found in Appendix A. The participant could make a decision by selection one of the following answers:
o Yes, I would like to invest in insulation for my house for €4990. o No, I do not want to invest in insulation for my house for €4990.
15
Table 2: Displayed answer order
Condition Decision order Percentage ‘Yes’ displayed first
Control group Insulation 47%
Solar panels 59%
Control group Solar panels 48%
Insulation 43%
Social influence treatment Insulation 47%
Solar panels 59%
Social influence treatment Solar panels 49%
Insulation 44%
In the treatment group, social influence was added to the situation text, while in the control group this information was absent. The social influence was integrated in the text in the form of storytelling. Studies have shown that storytelling has a positive influence on memory and individual judgement (e.g. McGregor & Holmes, 1999) and that stories stimulate one’s feelings and emotions and can even create a distance from reality, while reading the story (e.g. Green & Brock, 2000; Omeragic, 2016). A detailed explanation of the incorporation of social influence in the situation text can be found in Appendix B. The social influence text for solar panels differs from the text for insulation, due to the fact that solar panels are visible (see Appendix A).
After both decisions are made by the participant, the questions related to what influenced one’s decision were presented to the participant. These questions are adopted from the experiment of Koers (2019). For the control groups, this part contained thirteen questions (insulation), or twelve questions (solar panels). For the treatment groups, this part consisted of fifteen questions (insulation), or fourteen questions (solar panels). The two additional questions in the treatment group are related to the perceived social pressure and the perceived influence of the hypothetical neighbours. All the questions are measured on a seven-point Likert scale, ranging from strongly disagree to strongly agree.
The second part consists of questions related to the control variables and independent variables. Respectively, the questions related to risk attitude, environmental norms, personal innovativeness, perceived price value, injunctive norms, PU and PEOU were presented to the participant.
Risk attitude: for the control variable risk attitude, the participant is asked to answer questions about
16
financial decisions, leisure and sport, one’s professional career, and in general. The questions are measured on a seven-point Likert scale, ranging from very risk avoiding to very risk-taking, and are adopted from the studies of Fischbacher et al. (2015) and Dohmen et al. (2011). The main reason to include a second measure for risk attitude is due to inconsistent responses in the study of Koers (2019).
Environmental norms: is used as a control variable for the experiment, suggesting that people who
score high on environmental norms, will more likely invest in retrofit. The questions are adopted from the study of Koers (2019). Environmental norms are measured with three items each (see Appendix A) on a seven-point Likert scale, ranging from strongly disagree to strongly agree.
Personal innovativeness: suggest that people who score high on personal innovativeness, will more
likely invest in retrofit. This control variable is measured with three items each on a seven-point Likert scale, ranging from strongly disagree to strongly agree. The items are adopted from the study of Koers (2019) and can be found in Appendix A.
Perceived price value: is used as a control variable for the experiment, suggesting that the higher the
price value is perceived, the more likely an individual will invest in retrofit. The items are adopted from the study of Koers (2019), and can be found in Appendix A. Perceived price value is measured on a seven-point Likert scale, ranging from strongly disagree to strongly agree, with three questions.
Injunctive norms: are used as an independent variable related to social influence, which is a
self-reported norm. The questions are adopted from the study of Koers (2019), suggesting that someone who scores high on injunctive norms, will more likely invest in retrofit (Curtius et al., 2018; Koers, 2019). This independent variable is measured with three items each, on a seven-point Likert scale, ranging from strongly disagree to strongly agree.
PU and PEOU: are the last independent variables measured in this experiment. The items related to PU
and PEOU in the context of investing in retrofit are non-existent in the current literature. In order to operationalize the correct items for this context, the items as proposed in the general TAM studies (e.g. Davis, 1989; Malhotra & Galletta, 1999; Venkatesh & Davis, 2000) are reviewed. Second, the more specific items of PU and PEOU for the context of the energy transition (i.e. Broman Toft, et al., 2014; Heesen & Madlener, 2014; Kardooni, Yusoff, Kari, 2016; Ahmad, Mat Tahar, Cheng & Yao, 2017) are reviewed. These items were found useful and are adjusted to the specific case of investing in retrofit. The constructs are measured with three items each on a seven-point Likert scale, ranging from strongly disagree to strongly agree. The used items can be found in Appendix A.
17
.937), PEOU of insulation (α = .766), PEOU of solar panels (α = .817), are also above the threshold level of acceptable internal consistency (α = .700), and useful. PU of insulation (α = .662) and PU of solar panels (α = .667), were lower than the minimal acceptable threshold level of acceptable internal consistency (α = .700). Although, the constructs are below the threshold level, the alphas are still closer to 0.7, than to the unacceptable alpha of 0.6. Also, the rounded alphas are both 0.7, and therefore for this study all constructs have an acceptable internal consistency and are useful.
Finally, the last part of the experiment contained manipulation checks and two question to measure social influence (conformity) in order to measure differences between the treatment group and the control group, the questions “I would feel embarrassed if all my neighbours had invested in solar panels, except me” and “I would feel embarrassed if all my neighbours had invested in insulation, except me” are used (both measured on seven-point Likert scale). Lastly, the demographics (gender, age, nationality, education programme, and faculty) of the participants are collected, and the final extra payment screen is displayed to the participant. See Appendix A, for the experiment as presented to the participants.
3.3. Dealing with biases and student sample
In this study, several measures are taken to avoid biases and unreliable experiment results. First, the self-selection bias is reduced by presenting that the experiment is only about household adoption of retrofit. No information about solar panels, insulation, energy efficiency, climate change or the energy transition is presented to the potential participants, to reduce the possibility of selection bias. Second, as proposed in section 3.2., the acquiescence bias is reduced by randomizing the answer order of the investment decisions. Thirdly, the confirmation bias (confirming to the social desirable answer to invest in insulation and solar panels), is reduced by increasing the uncertainty for investing (chance of -€0.25 extra payment), regarding not investing (chance of €0 extra payment). Furthermore, the confirmation bias is reduced by presenting that a participant will incur transactions cost (both monetary and non-monetary costs), if they pursue to invest in retrofit. Lastly, the question order bias is reduced by randomizing the order of the questions, and the order of the presented retrofit measure.
18
3.4. Data analysis
The statistical programme SPSS 24 is used to test the hypotheses of this study. The dependent variable (investment decision in retrofit) is a categorical variable (yes versus no), the independent variables are categorical (social influence versus non-social influence), and ordinal (seven-point Likert scale). First, one-way ANOVA is carried out to determine if the means of the control group and the treatment group significantly differ from each other. Second, a Chi-square test of independence is performed to determine if the dependent and independent variables are associated. Third, correlation analyses are performed to show the correlations between the variables. Fourth, multiple stepwise binomial logistic regressions are conducted to test the hypotheses and to examine the underlying relationships between the dependent, the independent variables and the control variables in multiple steps.
4. Findings
In this chapter, the results from the lab experiment are analysed. First, an overview of the sample is presented. Second, the descriptive statistics and correlations are described. Lastly, the hypotheses are tested by stepwise binomial logistic regressions, and the results presented for insulation and solar panels.
4.1. Sample overview
In Appendix E, table 8, an overview of the sample characteristics (N = 319) is presented. The majority of the sample is female (57.1%). The minimum age is 17 years, the maximum age is 35, where the largest group is between 17 and 22 years old (79.6%). The sample contains 74 different nationalities, with as largest group the Dutch students (28.5%). The largest group in the sample is doing their bachelor-degree (79.6%) and most students study at the Faculty of Economics and Business (85.9%).
19
4.2. Differences between control and treatment group
In this part, the descriptive statistics are presented and analysed. The results are split in the decisions for insulation (table 3), and the decision for solar panels (table 4). The combined descriptives of both groups (N = 319) can be found in table 11 (for insulation) and table 12 (for solar panels) of Appendix E.
4.2.1. Insulation decision
Table 3 shows the descriptive statistics for the control group and the treatment group for the insulation decision. The means of the control group (n = 160) and treatment group (n = 159) differ slightly from one another. One-Way ANOVA is carried out to check whether these differences are significant. Results show that price value insulation is statistically different between the control group and the treatment group: F(1,317) = 6.455, p < .05. Other constructs show no statistically significant difference between both groups (see Appendix E, table 13). Because there is a heterogeneity of variances for PU of insulation, as assessed by Levene’s test for homogeneity of variances (p = .029), a One-Way Welch ANOVA was conducted. The results show that PU of insulation is statistically different between both groups: Welch’s F(1,309.834) = 8.281, p < .005. This implies that for insulation, the means of price value and PU differ significantly for the control and treatment group. Thus, this suggests that the social influence treatment possibly has interfered with the results of the other measured variables.
Table 3: Descriptive statistics insulation decision
Control group
Without social influence
Treatment group
With Social influence
Mean Median SD Min Max N Mean Median SD Min Max N
Insulation decision .54 1 .500 0 1 160 .62 1 .488 0 1 159 Injunctive norms 4.25 4.33 1.285 1 7 160 4.51 4.67 1.309 1 7 159 PEOU 3.58 3.67 1.149 1 7 160 3.61 3.67 1.210 1 6.67 159 PU 5.27 5.33 .834 2.33 7 160 5.56 5.67 .966 3 7 159 Environmental norm 5.18 5.33 1.092 2 7 160 5.38 5.67 1.238 2 7 159 Personal Innovativeness 4.43 4.67 1.269 1.33 7 160 4.42 4.33 1.281 1.33 7 159 Price value insulation 4.46 4.33 1.074 1 7 160 4.76 5 1.069 2 7 159 Risk attitude general 4.42 5 1.146 1 6 160 4.31 5 1.243 1 7 159 Risk attitude financial 3.63 3 1.577 1 7 160 3.78 4 1.598 1 7 159
4.2.2. Solar panels decision
20
Table 4: Descriptive statistics solar panels decision
Control group
Without social influence
Treatment group
With Social influence
Mean Median SD Min Max N Mean Median SD Min Max N
Solar panel decision .64 1 .482 0 1 160 .58 1 .494 0 1 159 Injunctive norms 4.28 4.33 1.380 1 7 160 4.41 4.67 1.397 1 7 159 PEOU 3.79 3.67 1.268 1.33 7 160 3.91 3.67 1.344 1 7 159 PU 5.33 5.33 .972 1.67 7 160 5.45 5.67 .971 2.67 7 159 Environmental norm 5.18 5.33 1.092 2 7 160 5.38 5.67 1.238 2 7 159 Personal Innovativeness 4.43 4.67 1.269 1.33 7 160 4.42 4.33 1.281 1.33 7 159 Price value solar panels 4.82 5 1.161 1 7 160 4.84 5 1.243 1.67 7 159 Risk attitude general 4.42 5 1.146 1 6 160 4.31 5 1.243 1 7 159 Risk attitude financial 3.63 3 1.577 1 7 160 3.78 4 1.598 1 7 159
4.2.2. Adoption percentages
Figure 5 shows the percentages of the respondents that invested in insulation and in solar panels under the different conditions. Under the social influence condition, the respondents invested more in insulation (61.64%), than under the control condition (53.75%). But, under the social influence condition, the participants invested less in solar panels (58.49%) than in the control condition (63.75%). The respondents invested relatively more in solar panels (61.12%) than in insulation (57.69%). A chi-square test of association is performed to test whether an association exist between the insulation and solar panels decisions and the social influence decisions. The results show that there is no association between the insulation decision of the control group and the treatment group: χ2(1) = 2.031, p = .154, and that there is also no association between the solar panels decision of the control group and the treatment group: χ2(1) = .928, p = .335.
Figure 5: Results investment decisions in insulation and solar panels
53,75% 63,75% 61,64% 58,49% 57,69% 61,12% I N V E S T E D I N I N S U L A T I O N I N V E S T E D I N S O L A R P A N E L S
INVESTMENT DECISION IN %
21
4.3. Correlations
In order to investigate the correlations between the different variables, the correlation matrix is presented in table 5 for insulation and in tables 6 for solar panels. The separate correlations for the control group and the treatment group can be found in Appendix E (table 16-19). Because none of the variables was normally distributed, as determined by the Shapiro-Wilk’s test (p < .05), the Spearman’s rank-order correlation coefficient is used for the correlations.
Table 5 shows that the insulation decision is positively significantly correlated with all the independent and control variables, except (2) social influence and (10) financial risk attitude. Furthermore, (4) PEOU and (5) PU are both slightly positively correlated with one another, and (5) PU has a moderately-high positive correlation with (6) environmental norms and (8) price value insulation. The latter is not extraordinary, because PU of insulation is related to environmental and price advantages. Surprisingly, (10) financial risk attitude is less correlated with other variables, compared to general risk attitude. This is surprising, because investing in insulation is more in line with financial risks than to general risks.
Table 5: Correlations insulation decision (N = 319)
1 2 3 4 5 6 7 8 9
1. Insulation decision
2. Social influence (dummy) .080
3. Injunctive norms .153** .095
4. PEOU .141* .017 .129*
5. PU .380** .167** .342** .135*
6. Environmental norm .163** .122* .213** .044 .424** 7. Personal Innovativeness .123* -.019 .093 .210** .142* .172**
8. Price value insulation .296** .150** .322** .198** .429** .340** .222** 9. General risk attitude .148** -.049 -.123* .034 -.062 -.036 .259** .021
10. Financial risk attitude .052 .042 -.076 .029 -.061 -.076 .248** .068 .535** * Correlation is significant at the 0.05 level (2-tailed)
** Correlation is significant at the 0.01 level (2-tailed)
22
Table 6: Correlations solar panels decision (N = 319)
1 2 3 4 5 6 7 8 9
1. Solar panel decision
2. Social influence (dummy) -.054
3. Injunctive norms .143* .050
4. PEOU .158** .052 .195**
5. PU .343** .054 .363** .150**
6. Environmental norm .116** .122* .291** .061 .422**
7. Personal Innovativeness .120* -.019 .172** .253** .220** .172**
8. Price value solar panels .338** .010 .359** .317** .392** .252** .292** 9. General risk attitude .078 -.049 -.122* .092 .007 -.036 .259** -.008
10. Financial risk attitude .081 .042 -.048 .068 .016 -.076 .248** .075 .535** * Correlation is significant at the 0.05 level (2-tailed)
** Correlation is significant at the 0.01 level (2-tailed)
For both the insulation and solar panels decisions, general risk attitude and financial risk attitude are highly correlated with each other, which is unexceptional, because they both measure risk-taking attitudes. Also, the risk attitudes show mixed results regarding the other variables in the above correlations. Therefore, the regressions in the next sections are carried out with general risk attitude, because it correlates with more variables than the financial risk attitude. In section 4.6, the regressions are also carried out with financial risk attitude to show differences in results.
4.4. Relationship social influence and investment in retrofit
23
(injunctive norms and descriptive norms) does not predict the individual’s investment decision in insulation.
Next, the regression for the solar panels decision are performed (see table 21, Appendix F). Again, model 1 contains the control variables, which predicts 67.4% of the cases correctly, and is statistically significant: χ2(4, N = 319) = 36.898, p < .001. Model 2a adds the social influence treatment compared to model 1. This model predicts 69.6% of the cases correctly, but adding the social influence treatment to the model is insignificant: χ2(1, N = 319) = 1.119, p = .290. Model 2b adds the injunctive norms to the control variables: it predicts 67.1% of the cases correctly, but adding injunctive norms to the control variables is also insignificant: χ2(1, N = 319) = .395, p = .530. Model 3 is a combined model, it contains the control variables, the social influence treatment and the injunctive norms. This model correctly predicts 68.7% of the cases, which is a decrease compared to model 2a with the social influence treatment only (-0.9%), but an increase compared to model 2b with the injunctive norms (+1.6%). In model 3, the social influence treatment (B = .264, Wald χ2 (1) = 1.152, p = .283) and the injunctive norms (B = .066, Wald χ2 (1) = 0.433, p = .510) are both insignificant. Furthermore, model 3 accounts for 11.4% till 15.4% of the total variance. These results show that social influence (injunctive norms and descriptive norms) does not predict the individuals’ investment decision in solar panels.
Based on the binomial logistic regressions results, social influence, both the social influence treatment, and the injunctive norms, do not contribute to the models. The inclusion of social influence, injunctive norms and both to the six regression models show insignificant results. Therefore, based on these results, hypothesis 1 must be rejected, thus social influence does not impact the investment in retrofit.
4.5. Relation PU and PEOU with investment in retrofit
This section tests hypothesis 2 and 3 by stepwise binomial logistic regressions. Hypothesis 2 forecasts that PU is positively related to the investment decision in retrofit, whereas hypothesis 3 predicts that PEOU is positively related to the investment decision in retrofit. Again the results are presented for insulation and for solar panels.
24
influences the insulation decision. Finally, model 3 combines the control variables and both PEOU and PU. The Hosmer and Lemeshow goodness-of-fit test shows a significant result (χ2 (1) = 3.843, p < .05), which indicates that model 3 is poor in predicting the categorical outcomes. Therefore, model 3 must be rejected. Based on the above results, PU predicts the individual’s investment decision in insulation, while PEOU cannot predict an individual’s investment decision for insulation.
Also, the regressions for the solar panels decision are performed (see table 23, Appendix F). Again, model 1 (only the control variables) is statistically significant: χ2(4, N = 319) = 36.898, p < .001 and predicts 67.4% of the cases correctly. Model 2a (control variables and PEOU), predicts 68.3% of the cases correctly. Adding PEOU to this model shows an insignificant result: χ2(1, N = 319) = .944, p = .331. Model 2b (control variables and PU) predicts 71.2% of the cases correctly, and adding PU to the control variables is significant: χ2(1, N = 319) = 23.355, p < .001 and significantly improve the model: B = .746, Wald χ2 (1) = 20.801, p < .001. This suggests that an increase in PU positively influences the solar panels decision. Lastly, model 3 (control variables, PEOU and PU) correctly predicts 71.5% of the cases. This model shows that PEOU (B = .103, Wald χ2 (1) = .913, p = .339) is insignificant, while PU (B = .747, Wald χ2 (1) = 20.724, p < .001) is significant. The model’s prediction accuracy improves after adding both PEOU and PU to the model up till 71.5%. Furthermore, model 3 accounts for 17.4% till 23.7% of the variance. From this analysis can be concluded that PU predicts the individual’s investment decision in solar panels, while PEOU cannot predict an individual’s investment decision for solar panels.
Based on the regressions results, the inclusion of PU to the three regression models shows significant results (and it contributes to the models). Therefore, hypothesis 2 is supported: PU is positively related to investing in retrofit. Meanwhile, PEOU does not contribute to the models. The inclusion of PEOU to the three regression models shows insignificant results. Therefore, hypothesis 3 must be rejected: PEOU does not impact the investment in retrofit.
4.6. Robustness checks
In this section, a summary of the robustness checks analysis is presented. For a more thorough analysis see Appendices G and H.
Financial risk attitude: in Appendix G (table 24-27), the regressions are also performed with the
25
Treatment checks: several respondents failed one or more treatment checks during the experiment, or
answered the questions very fast. Therefore, an arbitrary choice is made to subtract multiple respondents who failed all treatment checks, or made hasty investment decisions (assuming that they did not read the situation text). This analysis resulted in the exclusion of 25 respondents, resulting in a sample of N = 294. Several statistical analyses are carried out, which show comparable results to the dataset of N = 319, see Appendix H, section 1. Therefore, there is no need to reject the results of the N = 319, as these are more robust than the ‘exclusions’ sample due to the larger sample size and the more equally assigned number of respondents in the different conditions.
Order effects/independencies: multiple chi-square goodness-of-fit tests are conducted to check for
potential order effects for the insulation and solar panels decisions between the different conditions (see Appendix H, section 2). The results show that no decision order effects exist between the different conditions, compared to the total frequencies of the investment decisions. Thus, the frequencies within the conditions are not significantly different from the total frequencies. Also, multiple chi-square tests of independence are performed for the different conditions to check if the insulation decision and the solar panels decision are related. The results show that the insulation and the solar panels decision are significantly dependent on one another (see Appendix H, section 3). Thus, the participants’ decisions in the experiment are not independent of each other.
5. Discussion
In this chapter, the results are discussed. First the results of the relation with social influence are discussed. Second, the results for the relation with PU and PEOU are elaborated upon. Furthermore, this chapter is concluded with the policy implications, limitations and future research sections.
5.1. Social influence and retrofit adoption
In this section the inclusion of social influence is discussed. The question that will be answered is: how does social influence affect the adoption of retrofit measures? In line with the study of Koers (2019), no relation could be found for social influence and the individual’s investment decision in retrofit. The results show that hypothesis 1 has no support, and has to be rejected.
26
investment decision in retrofit. The injunctive norms are also no predictor of the investment decision, although it shows that respondents perceive people who are important to them would neither approve, nor disapprove such investment in retrofit (for insulation: M = 4.38, SD = 1.301 for N = 319, solar panels: M = 4.34, SD = 1.389 for N = 319, measured on a seven-point Likert scale). In the latter case, several studies show that social influence in the form of the subjective norm, (which is comparable with injunctive norms) has an insignificant relationship with the intention, and thus with the actual behaviour (e.g. Yuanquan, Jiayin, Huaying, 2008; Roberts & Henderson, 2000; Wu & Chen, 2005), while several others (e.g. Venkatesh & Davis, 1996; Hu, Clark & Ma, 2003; Taylor & Todd, 1995), found a significant relationship between subjective norm and actual behaviour. The only difference between those studies, and this study, is that above studies are from the information system literature, whereas this study is related to the energy transition. This gives, in advance for the energy transition literature, an impression of the inconstancies about the relation between social influence and technology acceptance. Aforementioned proves that a phenomenon like social influence is very hard to grasp. Furthermore, the reported perceived social pressure (M = 2.85, SD = 1.586, n = 159), and the reported influence of the hypothetical neighbours (M = 3.23, SD = 1.747, n = 159), both measured on a seven-point Likert scale, showed that respondents perceived the social influence treatment as marginal. Therefore, it is highly presumable that social influence in this form alone is ineffective in the case of retrofit investments. It can be concluded that written normative messages suggesting that neighbours invested in retrofit measures only, do not influence one’s actual adoption, and that individuals also did not perceive it as influencing their behaviour. Kiesler, Siegel & McGuire (1984) suggest that digital communication weakens the social influence effect, because of the lack of non-verbal communication, and the lack of social feedback (and impersonal messages). According to Schultz, Nolan, Cialdini, Goldstein & Griskevicius (2007), normative messages have mixed results in changing behaviours. Abrahamse & Steg (2013) add that social influence in the form of descriptive norms seems to be less powerful than face-to-face social interactions. These findings explain why no significant results are found for social influence in written form.
27
the control group. In the experiment, the social influence treatment stressed the fact that most of the hypothetical neighbours have installed solar panels. The participant could have perceived this message as a reason to not invest in retrofit, because most neighbours already did (free-riding effect), which eventually led to lower adoption in solar panels within the treatment group. Future research should clarify if the boomerang effect exists in the adoption of solar panels under social influence.
Lastly, perceived price value is related to one’s investment decision for retrofit: someone who perceives the price value as higher, will more likely invest in retrofit. These results are comparable to the study of Girod et al. (2017). They emphasize that the willingness to pay (price value) is a determinant for intention in adopting intelligent heating controls (Girod et al., 2017). This study adds that perceived price value for retrofit is also a determinant for the adoption of retrofit. Risk attitudes (or risk aversion), show mixed results, whereas the logical determinant for an investment decision would be the financial risk attitude (the perception of one’s own risk-taking attitude in financial decisions). However, financial risk attitude showed no relation at all with the investment decision. The ‘general’ risk attitude showed a positive relationship with insulation, but not with solar panels. Fischbacher et al. (2015) show that for their study, general risk-taking attitude and financial risk-taking attitude showed both similar results for energy efficient renovations, which is a contrary finding compared to this study. It can be concluded that people who tend to take more risk in general are more likely to invest in insulation only.
5.2. PEOU/PU and retrofit adoption
In this section, the inclusion of PU and PEOU is discussed. The sub-question that will be answered is: what are the relationships of PU and PEOU with the actual investment in retrofit measures? Hypothesis 2, the prediction that PU has a positive relationship with the investment in retrofit, is supported. Meanwhile, hypothesis 3, the prediction that PEOU has a positive relationship with the investment in retrofit could not be found from the current results. Therefore, hypothesis 3 needs to be rejected: PEOU cannot be seen as a determinant of investments in retrofit.
28
context of the energy transition is a determinant of the intention towards adoption and acceptance of energy-related technologies. This study adds to the literature that PU is a determinant for the investment in retrofit, besides all the other factors of importance, such as the household characteristics, retrofit characteristics and home characteristics (Mulder, 2018; Koers, 2019).
An important side note, with regard to the above explanation, is that insulation and solar panels are not technologies that need to be used multiple times to determine the usefulness or easiness. In comparison to technologies such as smartphones or computers, which are used several times a day, solar panels or insulation are just a ‘plug and play’ technology. In general, after installation, no further user interaction is needed, besides some maintenance or removal of the solar panels. In the case of insulation, it is even not possible to interact with the product, because it is not visible for the adopter. Therefore, it is hard to determine whether the used classification for PEOU is an appropriate determinant in the case of retrofit. This is also the case for other similar energy-related technologies that do not need user interaction to function. The frequency of interacting with the retrofit measure itself is deficient, and therefore PEOU cannot be a determinant of investing in retrofit. A study of Röcker (2010) even suggests that the construct of PEOU, which was a predictor of the acceptance of a technology for decades, may become obsolete in the near future. Röcker (2010) suggests that due to more autonomous behaviour of information systems (such as machine learning, smart systems and systems that can act independently from the user), could lead to new factors of importance to study acceptance of technologies. This finding, although related to the information systems literature, is in line with the finding that PEOU for retrofit (which operates autonomously) is not a determinant for adoption of retrofit.
29
incorporates besides the household characteristics (psychological factors), also retrofit characteristics and home characteristics, as determined by the studies of Mulder (2018) and Koers (2019).
5.3. Exclusion of intention testing
In this study, the focus was only on the actual behaviour of investing in retrofit, instead of all behaviour towards investing in retrofit. Behaviour itself is characterized and shaped by attitudes, intentions and other beliefs, preliminary to the actual behaviour or acceptance (see, Fishbein & Ajzen, 1975; Davis, 1989; Ajzen, 1991). In this study, the attitude and the intention are not tested, and their relation towards the actual investment in retrofit stays unknown. The general assumption is made that if a respondent invests in retrofit, their attitude and intention are also positive and prior to the decision phase. The technology acceptance literature (Davis, 1989; Venkatesh & Morris, 2000), suggest that PU, PEOU, and social influence affect the attitude or the intention directly, and the actual behaviour indirectly. In this study, it is assumed that PU, PEOU and social influence directly influence the decision, which could not have caused the expected effect as proposed in the literature, because neither the attitude, nor the intention are tested. Also, due to the value-action gap, one cannot conclude that social influence had no effect on an individual. Social influence, as investigated in this study, could have had a significant effect on someone’s intention, and it may or may not have led to actual adoption, due to the gap between the attitude and behaviour (Frederiks et al., 2015). Furthermore, the literature suggests that social influence leads to higher adoption, while in the case of solar panels it has led to lower adoption under the social influence condition. Due to the aforementioned situation, and the value-action gap (intention differs from actual behaviour), this could have led to a stronger contrast between the intention under the social influence condition. Future research, which includes testing whether the intention differs from the actual behaviour, should clarify this matter.
5.4. Policy implications
30
policymakers could make use of information campaigns to shape the social perception of households positively, targeting the price value and PU of retrofit. If households perceive the price value and PU of retrofit as higher, it is more likely that those households will invest in retrofit. Also, as price value is related to the price of retrofit, the perceived price value of households can be improved by providing financial incentives either for the purchase of those retrofit measures, or incentives for the reduction of household use of fossil fuels. Households that perceive higher value for the price will more likely invest in retrofit. Albeit, the multiple factors that affect adoption, the focus should not be on one single policy to induce households to invest in retrofit. This is in line with the advice of Brown & Wang (2015) that policies need to be multifaceted and that pricing, financing and information aspects need to be combined into a new policy mix. Only with an integrated approach, fruitful results can be achieved to seduce households in adopting energy efficient technologies.
5.5. Limitations
First of all, the lab experiment as carried out now is possibly flawed, because the respondents could have made their investment decisions dependently on one another. This could be the case, because for the whole experiment a respondent gets €8 for participating anyway, and up till €3.50 for the two decision a respondents makes. Considering that the extra payment was per investment decision, and therefore independent is not true. A respondent could have responded dependently on his or her previous choice in order to distribute the risk over both choices. Randomization of which decision was presented first to the participant had ruled out this bias, but one cannot conclude that the decisions of the participants were totally independent. Results have shown that the insulation decision and the solar panel decisions in the different condition groups, but also combined are dependent on each other (see Appendix H, section 3). Therefore, it can be concluded that this dependency may have flawed the study results.
Secondly, the fact that the investment decision was presented as the first question (due to the set-up of the experiment to frame people in a certain situation to make the decision), the questions related to the independent and control variables are presented after the decision was made. This could have led to an unintended reverse causality of the results. For example, the questions related to PEOU are asked after the decisions are made, and therefore the decision to invest or not invest could have unintentionally biased the results of the variable. While, if the PEOU questions were asked before the investment decision, it could only have influenced the decision, and not the other way around.
31
consistency in comparison to the other used constructs in this study. Future research should adjust these items of PU for investment in retrofit to achieve higher internal consistency and thus a higher reliability. Lastly, the sample consists of students only, who have no (or little) experience with the framed situation (having their own home, investing in general, having a full-time job). It could have been hard for the students to imagine themselves in a situation in which they are inexperienced, and therefore made decisions which can be different from the general population. Schepers & Wetzels (2007) suggest that TAM studies that use student samples showed stronger relations and stronger effect sizes, compared to non-student samples. Also, the information that was provided was as minimal as possible to make the decision as simple as possible, but in the ‘real world’ such investment is very complex and needs a lot of information. In practice, an investment in retrofit would have taken more time, and would have not been made on the spot: a non-investor would also have a sort of ‘waiting’ option, to invest in the nearby future or to gather more information to make the best decision. Although, the experiment does not completely represent a real-world case, it gives valuable insights for future research.
5.6. Future research
32
experiments should take into account possible boomerang effects of social influence treatments, especially in the case of solar panels.
Future research could also carry out a controlled field study with a survey as used for this lab experiment. In a controlled field study the respondents can have a two-part survey, in which the first time the intention is tested, and by a follow-up survey actual adoption is tested. By studying a more heterogeneous sample, with participants which actually have a house and a job (without assuming it), with different amounts of savings and different backgrounds, can even give better insights in the effects of social influence, compared to student samples. Such a study will also give better insights into the value-action gap under social influence.
Lastly, future research could focus on the development of a TAM framework for retrofit measures (and other energy technologies that do not need user interaction). Such framework should include all different and important factors that influence the investment decisions of energy technologies, as proposed in previous research. This framework should also take into account possible moderating effects of social influence on the other constructs, if no direct effect can be found. Besides that, future research should also focus on developing more reliable measurements for PU and eventually PEOU of investment, in order to get reliable constructs with higher internal consistency.
5.7. Conclusion
33
References
Abrahamse, W., & Steg, L. (2013). Social influence approaches to encourage resource conservation: a meta-analysis. Global Environmental Change, 23 (6), 1773-1785.
Achtnicht, M., & Madlener, R. (2014). Factors influencing German house owners’ preferences on energy retrofits. Energy Policy, 68 (5), 254-263.
Ahmad, S., Mat Tahar, R. B., Cheng, J. K., & Yao, L. (2017). Public acceptance of residential solar photovoltaic technology in Malaysia. PSU Research Review, 1 (3), 242-254.
Ajzen, I. (1991). The theory of planned behaviour. Organizational Behaviour and Human decision processes, 50 (2), 179-211.
Ajzen, I. (2002). Perceived behavioural control, self-efficacy, locus of control and the Theory of Planned Behaviour. Journal of Applied Social Psychology, 32 (4), 665-683.
Allcott, H. (2011). Social norms and Energy conservation. Journal of Public Economics, 95 (9-10), 1028-1095.
Ambrosius, F. H. W., Hofstede, G. J., Bokkers, E. A. M., Bock, B. B., & Beulens, A. J. M. (2019). The social influence of investment decisions: A game about the Dutch pork sector. Livestock, 220, 111-122.
Archibald, R. W., & Newhouse, J. P. (1980). Social experimentation: some whys and hows. Retrieved from: https://www.rand.org/content/dam/rand/pubs/reports/2007/R2479.pdf
Armitage, C. J., & Connor, M. (2001). Efficacy of the Theory of Planned Behaviour: a meta-analytic review. British Journal of Social Psychology, 40, 471-499.
Barnow, B. S. (2010). Setting up social experiments: the good, the bad, and the ugly. Zeitschrift für ArbeitsmarktForschung, 43 (2), 91-105.
Beerepoot, M., & Beerepoot, N. (2007). Government regulation as an impetus for innovation:
evidence from energy performance regulation in the Dutch residential building sector. Energy Policy, 35 (10), 4812- 4825.
Benbasat, I, & Barki, H. (2007). Quo vadis, TAM? Journal of the Association for Information Systems, 8 (4), 211-218.
