Adoption of energy efficient technologies by households – Barriers, policies and agent-based modelling studies

Increasing the adoption of energy efficient technologies by households is one of the formulated strategies to reduce greenhouse gas emissions. This paper presents a systematic review of agent-based modelling studies on the adoption of energy efficiency by households. It starts with an overview of barriers for adoption, of energy efficiency policies, energy efficiency model types. Afterwards, an analysis is given of technologies modelled, policies simulated, decision-making theories included, and the use of empirical data. An overview is presented of how technologies, barriers and policies relate in the models. Furthermore, the core policy recommendations from existing models are presented. The analysis shows that the reviewed studies predominantly focus on a subset of barriers – a lack of capital, a lack of information, high upfront cost, ignorance, inertia and other priorities. So far, agent-based models have focused on how subsidies, technology bans and information cam- paigns influence energy efficiency adoption. There is ample opportunity for future agent-based modelling research on energy efficiency adoption policy by studying other residential technologies, other barriers, and other policies that fit the agent-based modelling paradigm well.


Introduction
Stimulating energy efficiency adoption is one of the strategies formulated by the international community to reduce greenhouse gas emissions and our contribution to climate change [1]. The European Union and national governments therefore seek to increase the adoption of energy efficiency in society [2]. The residential sector is marked as an important sector to contribute to the internationally set climate targets and the increased adoption of energy efficiency by households is needed to mitigate the effects of a globally growing population and increasing energy demand [1]. The desired level of energy efficiency in households has not been achieved yet and has been named the 'efficiency gap' [3].
Energy efficiency can be specified as: 'achieving the same services and performance while using a technology with less energy use' [2]. The increased adoption of energy efficient technologies should thus contribute to achieving the targets set. However, a still open question is: what are effective ways in which this adoption may be achieved in different EU member states? This paper focuses on energy efficiency in households. For the residential sector it is not clear which technologies should be adopted by households, why many people are not adopting the most efficient technologies and how policy makers should stimulate adoption? There are many types of barriers, i.e. structural, economic, social or behavioural barriers that stop households from adopting a new technology.
Policy makers need to design and implement policy that moves the residential sector to become more energy efficient by taking away these barriers. Modelling and simulation is done in order to gain an understanding of what policies can be expected to positively impact the energy efficiency of which households under what conditions. This understanding supports policy makers in their evaluation of policies and helps them to decide on possible interventions.
Mundaca et al. [4] identifies four main methodological categories to create bottom-up energy economy models: simulation models, optimization models, accounting models and hybrid models. Of these four there are two popular classes of modelling studies that provide a descriptive representation of household energy use and technology adoption: the ones using simulation models and those using accounting models [4]. Simulation models, in particular our focus on agent-based models (ABMs), give a quantitative depiction of technology adoption in the context of exogenous scenarios. Accounting models are equation-based and focus mainly on grasping empirical technology adoption data and composing an as accurate representation as possible of the inventory of technologies adopted under the influence of various policy interventions [4]. These accounting models are useful to explain the effects of policies on the adoption of energy efficient technologies, but are limited in how they take heterogeneity of households, household choices and communications into account. Accounting models can describe these aspects on an aggregated level for different types of households. It would dramatically increase the data requirements and computational complexity to explicitly capture interactions on an individual level [5].
Agent-based modelling (ABM) is a method that explicitly addresses the heterogeneous nature of households, diverse preferences and choices that households make, and communication between households [4][5][6]. It can provide insights regarding the effects energy efficiency policy may have on a heterogeneous set of households: by exploring the adoption of energy efficient technologies by households modelled with adoption decision-making on an individual level. ABM can, therefore, generate insights that complement the knowledge acquired from accounting models for the purpose of energy efficiency policy evaluation.
This paper presents a systematic literature review of ABM studies on the adoption of energy efficiency by households. The purpose of this review is to identify the policy recommendations that are produced by ABMs and learn what properties of ABMs contribute to formulation of concrete policy recommendations, in particular with respect to the adoption barriers known in the literature. Additionally this review identifies opportunities for future ABM research that can provide novel knowledge on energy efficiency adoption policies.
This paper is structured as follows: Section 2 explains the literature review methodology that has been applied. Section 3 discusses the energy efficiency adoption barriers, policies and modelling methods. Section 4 then reviews ABMs on the technologies, policies, theories and data they included. The paper ends in Section 5 with conclusions about the insights and opportunities ABM has for energy efficiency policy.

Methodology
In order to review the use of ABMs for energy efficiency policy evaluation, a systematic analysis of existing ABMs on the topic is presented. The review focuses on ABM studies that model energy efficient technology adoption. In order to make the review systematic and do an assessment of usefulness, an overview of the barriers to adoption and energy efficiency policies is presented first. Second, popular modelling methods for studying the adoption of energy efficiency are compared. This provides us with a perspective for the review of the ABMs which enables us to identify the insights that ABM studies provide and find opportunities for future research.
An important input is a previously conducted literature review on ABM studies that describe energy efficient technology diffusion, which concludes that ABMs are technology specific, suitable for representing heterogeneous households, supported by empirical data and decision making theories [5]. Our review is structured according to the study by Moglia et al.. However, our review focuses on the conclusions that can be drawn about policy interventions rather than on the elements of technology diffusion ABMs. Therefore, the content of the ABM studies is analysed to identify lessons learned about policy interventions for specific energy technologies.
The literature review consists of three steps, which are now described.

Step 1: overview of barriers, policies and modelling methods
Before conducting the literature review on ABM studies the problem space concerning the adoption of energy efficiency is explored and discussed. This is done by identifying the barriers that obstruct households from adopting a more energy efficient technology and by outlining the range of policy interventions that are available to policymakers. The barriers have been identified be studying literature that lists different types of barriers. The barriers literature has been searched in Scopus using search terms, amongst others energy efficient technology, barrier, energy behaviour and energy efficiency gap. To select the papers that were most relevant and listed barriers only a number of papers were handpicked that explicitly mentioned energy efficiency barriers and had numerous citations. For literature on energy efficiency policies, a similar procedure was followed. The main search terms user were energy efficiency policy or policy instruments, households or residential, energy efficiency, energy behaviour or energy conservation.
Creating this overview of barriers and policies provides us with the necessary context to judge and review the ABM studies on usefulness for policymakers. Also the use of accounting models and ABMs is discussed before conducting the review of ABM papers. This discussion highlights the aspects and insights that ABMs add to knowledge derived from energy efficiency adoption accounting models.

Step 2: definition of search terms and filters for collecting articles
The keywords that have been defined for this literature search have been selected to capture articles that describe ABM studies on the adoption of energy efficiency by households. The search has been conducted on the 17th of March 2017 using the advanced search option of the Scopus scientific library. The following search queries were used: • (ABM OR "agent-based modeling" OR "agent-based modelling") AND ("energy efficiency") AND (("household") OR ("consumer")) • (ABM OR "agent-based modeling" OR "agent-based modelling") AND ("energy technology") AND (("household") OR ("consumer") OR ("residential")) • (ABM OR "agent-based modeling" OR "agent-based modelling") AND "energy efficiency" AND "barriers" AND ("household" OR "residential" OR "consumer") The literature review has only considered scientific research papers from peer-reviewed journals and conference papers published in English. The queries that were used for the literature search provided 83 results. With the later queries, the search term 'residential' was added as an alternative term for households. A check was made that no relevant papers were missing in the result set from the first query. From the list a final selection was handpicked by scanning the titles and abstracts of these papers to only include papers that described studies with results from an ABM (so conceptual articles and articles not discussing ABMs are excluded). A thorough scan of all the papers after this step was needed to sort out papers that referred to ABMs but did not present any model conceptualization or results in the paper. This produced a list of 23 papers that have been reviewed. All of these ABM studies have been published in established energy and/or sustainability related journals. One of the papers was part of a conference proceedings.

Step 3: evaluation of the papers
The 23 papers have been evaluated and analysed by making an overview using a framework of analysis that is similar to the review by [5] that includes modelled technologies, theories used to structure the model and the use of empirical data. The analysis framework therefore starts by discussing the technologies that have been modelled in each study and answering whether the study considers 1) the adoption of efficient energy technologies and/or 2) the more efficient use of technologies by households. The existing review framework is augmented with the policy interventions considered and the outcomes related to those policies in order to highlight the relevance of these ABM studies for the purpose of policy evaluation. The type of policies modelled are discussed together with the degree to which the ABMs are used to test and design policy alternatives and the effects they have on adoption barriers. Next, the theories used to describe the decision making logic of agents in the models are discussed and last the use of empirical data is considered.

Energy efficiency adoption barriers
The 'efficiency gap' is a phenomenon that has been discussed by academia for decades. It can be defined as the slower than optimal adoption of energy efficient technologies [7]. Jaffe and Stavins [7] define optimal adoption in a number of different ways. This paper define the efficiency gap as the hypothetical energy efficiency level that can be achieved by overcoming all barriers that obstruct energy efficiency adoption. In scientific literature a diverse set of barriers have been identified, which all can manifest as cause of obstruction to the adoption of energy efficient technologies by households.
The energy efficiency adoption barriers differ in type and occur to households in varying degrees. Barriers such as split incentives, supply infrastructure limitations and codes and standards are structural in nature and largely outside the sphere of influence of households [9]. Other barriers are economic in nature: Households may lack the capital to invest in a new energy technology or find the upfront cost too high [3,8,9,12]. There are many behavioural barriers that have been identified in literature. House owners can have other priorities than energy efficiency when making a choice, be ignorant about energy efficiency or simply unwilling to change [5,8]. Last, there are barriers that occur due to social behaviour. People may weigh the actions by their social peers when making a decision about adopting a new energy technology or have trust issues with the novel technology [5].
This diversity in adoption barriers and household situations makes the adoption of energy efficient technologies a hard to capture process. Extensive empirical data and modelling is needed to describe all the barriers and complex interactions between households.

Energy efficiency policies
To address the 'efficiency gap' and overcome the barriers to energy efficiency adoption, policies are implemented. Table 2 presents a list of energy efficiency policies that have been discussed in literature. Most of these policies are directed at the adoption decision of individual households. Tax reductions, subsidies, discounts, prohibitions, promotions and awareness raising campaigns are all examples of policies that influence the adoption decision of the household directly. Some regulatory approaches like performance standards, building codes and trade restrictions narrow the available choices of households to more efficient technologies. There are also relatively novel policies, like the energy efficiency tender, tradable white certificates and energy efficiency obligations which are not directed at households but at intermediary parties. Intermediary parties play a significant role in the adoption decision process of households and can be specified as: a party that enables others to adopt energy efficient technologies. Examples of intermediaries are utility companies, electronics stores, web shops or energy service companies. Through policy these intermediary parties can be stimulated to push energy efficient technologies to households.

Descriptive modelling methods
The 'efficiency gap' from the perspective of the policy maker is a particularly interesting avenue for modelling research: the fit of policy alternatives to different adoption barriers may be specific to particular conditions, such as technologies/societal functions, countries, consumer groups, etc. Modelling and simulation can be used to describe and explain the effects of policy alternatives and design promising energy efficiency policies.
An established way of modelling energy demand and the adoption of energy efficient technologies by households is the use of accounting models [4,26,27]. Accounting based models usually focus on the management and use of large amounts of data and have a number of general characteristics: 1) Accounting models provide a detailed representation of energy end-uses, 2) Calculation of energy demand is done on basis of economic, social and technical factors, 3) A scenario approach is used to account the development of adoption under the influence of policy [26]. Taking many factors into account the energy demand is then determined by means of aggregation. The decision frameworks in accounting models are relatively elegant compared to other model types such as optimization and simulation models. The decision framework in accounting models commonly follow the following steps: 1) selection of policy measure, 2) selection of efficient technology, 3) definition of technical performance of the technology and 4) definition of the particular market penetration rate [4]. An example of a decision making framework used in accounting models is the use of net present value calculations. In models using net present value, like some models discussed in [28], the adoption decision of a household is reduced to an investment decision, based on a net present value calculation. At the core of modelling these investment decisions are discount rates. The discount rate helps to represent the perceived value of an investment and expresses this value in monetary form. To capture the influence of other, non-financial, factors and barriers on the investment decision, the implicit discount rate is added to the calculation. The implicit discount rate covers aspects such as preferences in risk and time, personality, habits and biases on top of the regular discount rate [8]. When based on profound empirical data this method is very powerful in explaining the investment decisions households take.
The predictive strength of accounting models also has limitations. Due to the need to aggregate data a substantial amount of data is lost in accounting models and the decision making frameworks are usually more simplified than in other model types. In the case of a model based on net present value calculations the barriers and factors are all reduced to a unidimensional unit: money and captured in an implicit discount rate [8]. In the process of this reduction to unidimensional unit some fundamental aspects of the barriers discussed are neglected. Accounting models can describe barriers on an aggregated level for different types of households, but describing barriers on an individual level would dramatically increase the data requirements and computational complexity [5].
ABM can complement these limitations of accounting models by offering flexibility and the possibility to use a more elaborate decision making framework. Households can be modelled as heterogeneous agents in ABMs and therefore can take into account the large variability in preference of households and also the diverse set of adoption barriers. In ABM the complex relations and interactions between households and household behaviour can also be explicitly addressed and the decision nature of technology choice modelled more specifically and elaborately. This can provide policy makers with knowledge about policy design and analysis that is complementary to the use of accounting models. To learn about what insights ABMs produce about energy efficiency policies the reviewed papers are discussed. Fig. 1 shows that the ABM efforts on energy efficiency studies are relatively young. The earliest published ABM study on the adoption of energy efficient technologies was published in 2010. As can be seen in the figure ABM has gained popularity over the years, with more ABM studies being published in recent years. This fits in the popularity of ABM for studying diffusing in general. This also suggests that energy efficiency adoption by households has mostly been addressed using other modelling methods (i.e. the accounting models discussed before). The review is used to explore where added value of ABM may lie.

Energy efficient technology adoption agent based models
In this section the reviewed ABM studies will be discussed on the technologies that are studied in the paper, policy measures that are explored using the ABMs, theories used to describe decision making of households and the use of empirical data. These angles are used to find what the literature has focused on so far and where opportunities lie for further research. An complete overview of the results is found in Table 5 in the appendix.

Technologies
From the 23 reviewed studies, 16 study specific adoption of energy technologies, the remaining seven studies focus on the efficient use of energy. 20 papers fixate their ABM around a specific energy technology and this technology specificity seems to pertain to energy efficiency adoption ABMs. Therefore the analysis discusses the technologies that are typically modelled using ABMs and their relation to energy efficiency adoption.
ABMs are suitable for describing and simulating the adoption or diffusion of a technology by a population of household agents over time. A first remark to be made about the modelled technologies is that there are 11 different technologies and/or behaviours modelled (Table 3), despite the small number of papers that have been reviewed. ABMs are thus created to model a comparatively broad set of technologies. Most of the reviewed paper have as primary purpose to gain diffusion or policy insights pertaining a particular technology.
Another general observation is that these models are forward looking and exploring the increase in energy efficiency in the future. A large portion of the ABMs describe technologies that are relatively novel and which have not established themselves as dominant technologies yet. This suggests that these studies focus on the efficiency gains of new technologies. Examples of these are the modelling studies on electric vehicles, solar PV and battery systems and micro CHP units.
In [43][44][45][46] the diffusion of electric or plug-in electric vehicles is explored. Electric vehicles are as a technology not directly linked to the subject of energy efficiency, only for households that produce their own electricity it can be said to be explicitly the case. However, the increase of the electric vehicle fleet can in any case be regarded as a move towards a more sustainable society. By substituting gasoline fuel cars by electric vehicles the possibility of having a fleet run on renewable energy increases. Energy efficiency is thus just regarded as a motive or reason, together with sustainability, to push the diffusion of this technology. This push for electric vehicle diffusion is based on the hypothesis that electric vehicles are a more sustainable technology than gasoline powered vehicles.
Another technology diffusion that is being explored using ABM is the adoption of residential solar PV systems and solar PV with battery systems. In [39,40,42,52] the adoption of these solar PV systems is researched. Solar PV systems are likely to play an increasingly significant role in the energy system as it allows households to produce their own electricity. Solar PV fits in the electrification process and can be seen as an important condition for households to adopt technologies (for instance electrical heating technologies) that use electricity and contribute to a better and more efficient use of energy.
The adoption of different lighting technologies are also explored by multiple ABM studies [29,30]. These ABM studies differentiate themselves from other studies because a variety of lighting technologies are being modelled within the same study. So rather than having a dichotomous model where it is either: adopt or not adopt the technology, multiple technologies that fulfil the same function are compared and judged on their energy efficiency performance. The lighting technologies that are modelled in these studies are: incandescent lighting, light emitting diodes (LED), compact fluorescent lighting (CFL) and halogen lighting. In [31,32] these lighting technologies are modelled as well. However the focus of these studies is on the use of these lighting technologies rather than on the adoption. Hicks and Theis [32] and Hicks, Theis and Zellner [31] conclude that the energy savings accomplished by efficient lighting technology adoption can be diminished and eroded by a rebound effect.
Other residential technologies that have been modelled are: Micro-CHP units [37], wall insulation [38], residential heating technologies like direct electric heating, wood pellet heating stoves and heat pumps [33][34][35]. These papers are exemplar of studies that describe the diffusion of a technology that improves the energy efficiency of a household compared to the current installed technology.
Lastly, in [47,48] the effect of a CO 2 meter and a behaviour-changing feedback device on the energy consumption behaviour of households is explored. These studies differ from the studies that model technology diffusion because the focus is put on the adoption of energy efficient behaviour through these technologies. In [49][50][51] the focus is also not on energy efficiency improvement through technology adoption but rather on more energy efficient behaviour of households and adoption of green eco-innovations.
The reviewed papers study a broad set of energy efficient technologies and behaviours. The majority of studies focus on improvement of energy efficiency and sustainability through technology adoption and therefore use energy efficiency as a reason or motive to push these more efficient technologies. Most of the studies are dichotomous, they compare households that do or do not adopt a new technology. There are some cases where different technologies that fulfil the same function are compared, this is the case for the papers that study lighting technologies. These studies show that ABMs can both explore dichotomous adoption decisions and adoption of multiple competing technologies. The latter option can be identified as an opportunity provided by the ABM method that needs further exploration, since there are other household technologies besides lighting technologies that fulfil a specific function for households. Some examples of household functions that have not been explored are refrigeration, cooking or washing. The adoption of different appliances that fulfil household functions can be compared and studied using ABM.

Energy efficiency barriers and policies
The main interest in ABM for policy makers lies in ABM's capability to assist the design and evaluation of energy efficiency policies. This section shows the policies that are included in the studies, the degree to which the studies contribute to policy design and evaluation and the barriers they tackle. This will provide us with insights on the usefulness of ABM for policymakers, the current knowledge derived from ABMs and possible opportunities for further policy exploration and evaluation.
The range of barriers and policies included in the reviewed studies is narrow, compared to the list of identified barriers in Table 1 and policies in Table 2. An overview of the policies that are included, the barriers that are targeted and the policy conclusions is shown in Table 4. Fig. 2 shows two graphs of the policies (left) and the barriers (right) that occur together. Fig. 3 shows links between policies and barriers (left) and technologies and barriers (right). The barriers addressed primarily deal with economic aspects (high upfront cost, lack of capital, uncertainty in fuel prices) and behavioural aspects (inertia, lack of information, ignorance, other priorities). A clear clustering is also formed that link economic and behavioural aspects, which appears to be at the heart of the ABM approach. The explored policy options are limited to a few forms of financial support, regulation and information campaigns. Most of the studies modelled financial support as policy to Table 1 Barriers to energy efficient technology adoption by households.

Structural barriers
Split incentives A circumstance in which the flow of investments and benefits are not properly rationed among the parties to a transaction, impairing investment decisions [3,[8][9][10][11] Uncertainty of prices of energy carriers and other resources Uncertainty about the price of fuels, electricity or other resources defining the production cost of technologies in the future [9,12,13] Supply infrastructure limitations Limitations of the supply infrastructure of energy efficient technologies [3,9] Lack of codes and standards Codes and standards for existing energy efficient technologies, in the field of EE but also other kinds of codes and standards (safety). [9,13] Economic barriers Lack of capital Insufficient capital available to do an investment [8,9,12] High upfront costs High upfront investment cost of energy efficient technologies [3,12] Lack of information Insufficient information available to make an investment decision [8][9][10][11][12][13][14] Transaction costs Cost incurred in making an economic exchange of some sort/the cost of participating in a market (search and information costs, bargaining costs, policing and enforcement costs) [8,11] Behavioural barriers Bounded rationality Decision maker's rationality is limited by the tractability of the decision problem, the cognitive limitations of their mind and the time available to make the decision Satisficing People do not tend to optimize their decision but rather aim to satisfy a small set of criteria, i.e. the minimum requirements [5] Availability bias People primarily draw on knowledge and information that is easily accessible. Lack of information may mean that some opportunities are missed [5] Other priorities Other priorities than energy efficiency when purchasing an energy technology [5,12,13] Ignorance Lack of knowledge, understanding or education about energy efficiency [5,12]

Inertia
People have a tendency to want to stick with the status quo rather than having to change for practical reasons and for convenience; as they like to avoid hidden costs associated with a switch [5,12,13] Persisting with sunk costs Once people have invested in something, in terms of time and/or money, they tend to become fixated on 'recovering losses' [5] Being loss and risk averse People weight losses more than gains when making decisions and people tend to avoid the prospect of a loss even with the prospect of certain gains, and tend to accept a gamble in order to avoid a loss [5,8,9] Irrational response to monetary incentives People's response to incentives are often short-lived and unpredictable and may crowd out intrinsic motivations [5] Free-riding effect People tend to look for ways that they can gain benefits without paying for them [5]

Social behavioural barriers Social comparisons
People tend to follow the behaviour of others, i.e. following the norm [3,5,13] Trust People seek information and judgement from those that they trust. People may also trust information from specific people or institutions more than others. [3,5,13]  Households [16] stimulate the adoption of a technology. Twelve studies did this by including a subsidy or tax exemption policy in their models. Five studies modelled a form of regulation such as a technology ban or adoption obligation. Five studies considered informational instruments such as a campaign or promotion in their modelling studies. Last, nine studies did not examine any policies.
The most modelled policies are the economic, financial and market based instruments (Table 2), typically in the form of a subsidy or tax exemption. They aim to tackle a variety of barriers with a focus on high upfront costs and a lack of capital. The conclusions about the effectiveness of these policies is found to differ across the studies as is shown in Table 4. In the studies about lighting technologies subsidies are found to be less effective at increasing energy efficiency adoption than other policies. Chappin and Afman [30] find this to be the case and Hicks et al. [32] conclude that a subsidy in combination with other policies can be beneficial as long as households do not increase their energy consumption due to lowered electricity cost caused by having more efficient lighting technologies. These results suggest that for lighting technologies the barriers of high upfront cost and lack of capital do not apply. Studies on heating technologies are more optimistic about the benefits of financial incentives in the form of subsidies. Snape et al. [35] finds that the UK Renewable Heat incentive has already been effective in increasing the adoption of heat pumps and that more policy action is needed to keep this adoption rate from falling. Faber et al. [37] researched the adoption of micro-CHP units and found subsidies based on decreasing price differences between technologies to be much more effective than applying a purchase subsidy. Sopha et al. [33] and Sopha et al. [34] conclude that support for technical development in heating technologies such as wood-pellet stoves are, in combination with other policies, necessary to increase the adoption of these technologies. Palmer et al. [39] finds that current solar PV scheme of Italy is already effective and that at the current level of subsidies the adoption rate will probably slow down. Zhang et al. [41] states that incentives have little effect on increased solar PV adoption and instead suggests another policy to increase adoption: seeding, which is providing a number of households with free solar PV systems and then relying on peer effects to increase adoption. Alyousef et al. [42] concludes that support for battery efficiency research is needed together with a higher electricity price to increase adoption, which can be seen as a subsidy to electricity producers and battery owners. For the solar PV technology high upfront cost and a lack of capital are thus considered adoption barriers that can be addressed with different sorts of subsidies. The studies on electric vehicle adoption see subsidies for electric vehicles as a necessity to stimulate adoption. Noori and Tatari [44] and Silvia and Krause [46] suggest a combination with information instruments and Wolf et al. [45] conclude that an exclusive zone for electric vehicles in cities boost adoption more effectively as opposed to financial incentives alone. For the adoption of electric vehicles these studies see as most prevalent barriers high upfront cost, lack of capital and a lack of information about electric vehicles. Subsidies are modelled most often of all policy alternatives. A possible explanation for this popularity is the flexibility of this policy. Governments can incentivize adoption with different types of subsidies, like subsidizing technology use, technology adoption or the fuel of a technology. Moreover, in these studies subsidies are combined with other policy alternatives to address a multitude of adoption barriers, as Fig. 2 and Fig. 3 suggest. Subsidies seem most effective in the studies on technologies where high upfront cost and a lack of capital are the main adoption barriers, such as the studies on electric vehicles and the heat pump [35,[44][45][46]. In the studies on solar PV, lighting technologies and Micro-CHP's [30,32,39,41,42] subsidies seem less effective. This suggests that for these technologies other barriers, like inertia and social comparisons are more relevant than the high upfront cost or lack of capital barriers.
Five studies investigate regulatory approaches (Table 2) in the form of a ban, obligation or price regulation. These policies are in all cases found to be effective at increasing the adoption of more energy efficient technologies. Cao et al. [29] and Chappin and Afman [30] both conclude that banning incandescent lamps is a most effective policy to increase the uptake of more energy efficient lighting technologies and note that increasing the share of high-efficient lighting technologies is essential. These results suggests that household's inertia and having other priorities are important barriers that block adoption of efficient lighting technologies. Chappin and Afman [30] also note that a tax at a sufficiently high level can be effective, Hicks et al. [32] also find this but warn about a rebound effect. Friege [38] researches the adoption of wall insulation and finds that obligating new homeowners to insulate the walls has a great potential to increase the insulation rate. Sopha et al. [33] find that establishing a stable price for fuel for wood-pellet stoves and other heating technologies, is one of the requirements to stimulate adoption. This can be accomplished through a price regulation or other policy forms. Fig. 3 shows that regulation and taxation address adoption barriers like inertia, other priorities and ignorance. In the studies where they are researched they seem have a positive effect on the adoption of energy efficient technologies by households. Regulation also seems more effective at overcoming inertia and other priorities than the use of subsidies in these studies.
Last, five studies examined the effects of informative and voluntary schemes ( Table 2) on the adoption of energy efficient technologies. These instruments were either an information campaign or promotion in the reviewed studies. These policies aim at barriers such as a lack of information, inertia, ignorance and other priorities, as indicated in Fig. 3. Cao et al. [29] (lighting technologies), Sopha et al. [33] (heating technologies) and Friege [38] (wall insulation) all found these information campaigns not to be as effective as other available policies. This indicates that the barrier of lack of information may be high or that there are other barriers that obstruct the adoption of energy efficient technologies by households. Only in studies about electric vehicle adoption were information campaigns found to be effective, given that these were combined with a subsidy policy [44,46].
The limited diversity in policy measures explored in the reviewed studies is noteworthy, compared to the relatively diverse set of technologies and behaviours modelled. There are more types of financial, regulatory and informational instruments available that could be explored. Also policies that include the role of the intermediary have not been found in the reviewed studies, which is a clear opportunity for future ABM studies. Policies, such as tradable white certificates, energy efficiency obligations and energy efficiency tenders, include different  Table 4 Policy conclusions and barriers.

Policies included
Barriers targeted actors in the diffusion process of energy efficiency. Since these are relatively novel policies the data on interactions between intermediaries, households and policy is not as readily available, which can be marked as one of the explanations why it has not been done before. The effectiveness of these policies has not been explored yet and ABM is a suitable method to do this because it allows the modelling of complex relationships and interactions between multiple and heterogeneous agents [5].

Economic, financial and market based instruments
A substantial number of studies do not take policies or policy scenarios into account at all. Azar and Menassa [36] and Rai and Robinson [40], aim to illustrate that the use of ABM is useful to model dynamics and heterogeneous agents. Other studies that do not include policies are by Chen et al. [49], Kowalska-Pyzalska et al. [50] and Byrka et al. [51], operationalize theories using an ABM to illustrate how these theories explain a phenomenon. For instance in Kowalska-Pyzalska et al. [50] the ABM offers a hypothetical explanation for the evaluated discrepancy between consumer opinions in surveys and the actual participation in pilot programs and the adoption of dynamic tariffs.
Ten of the reviewed studies analyse the adoption of energy efficient technologies, these studies do not specifically test different policies and policy scenarios, but rather study the diffusion of a particular technology. The effects of policy are included in these studies, among other external factors that influence diffusion, but are not the central focus of the ABM study. Examples of these studies are Hicks and Theis [31] and Hicks et al. [32] who study the adoption of efficient lighting technologies in different scenarios of consumer light consumption. They conclude that all savings acquired through policy can be eroded and negated by a rebound effect. Other examples are Snape et al. [35], who conclude that the current incentives program for heat pumps is successful but insufficient for the future and Palmer et al. [39] who find that the current subsidies in Italy have already accomplished a significant adoption rate. These conclusions are certainly relevant to policymakers, but policy testing is not the focus of these studies. Seven studies conduct more focused policy testing. Examples of these are: Cao et al. [29] and Chappin and Afman [30] who test different policies for efficient lighting technology adoption. Sopha et al. [33] do a policy scenario analysis for a few heating technologies and compare how different combinations of these policies affect adoption. Friege [38] compares policies like new homeowner obligations for wall insulation and the effect of information campaigns on wall insulation adoption.
What we can conclude from this review from a policy perspective, is that the majority of studies aims to analyse the adoption of energy efficiency and policy testing and is therefore of interest to policy makers. The explored policies are however, limited in scope. Mostly financial incentives are modelled in these ABMs, whilst more policies constructs have been identified in literature, like policies that include an intermediary. To gain insight on what aspects of ABMs allow modellers to do policy testing and design, the decision making theories and input data used need to be examined.

Decision theories
The decision making logic of agents in ABMs are often rooted in theory, i.e. a particular decision making theory is explicitly used to describe the adoption decision of a household in the model and we explored how the choice of theory affects the barriers modelled.
The most often applied theory is the Theory of Planned Behaviour by Ajzen [53]. This theory states that the behaviour of an individual is determined by the intention and perceived behavioural control of this individual. The intention in turn is influenced by the attitude and subjective norm of the individual. This provides a basis for where barriers come from and how they may be taken away at the individual level and therefore fits well with ABM in general and in particular those that focus on energy efficiency. Seven of the studies use this decisionmaking theory to describe the decision making of households [33,34,40,47,48,50,51]. A reason for the popularity of the Theory of

Policy conclusion
Informative and voluntary schemes [29] Lighting Planned Behaviour could be its flexibility and elegance. The theory is relatively easy to operationalize. Household intentions can be modelled as threshold values for different behaviours. Moreover, the attitudes, norms and intentions of individual households can be broken down into more specific attitudes, norms and intentions. This makes it very suitable for including multiple barriers in the ABM. Households can for instance have a certain attitude towards a new technology and a certain preference for known technologies, making it possible to include the inertia barrier in the ABM. The Theory of Planned Behaviour is used to model a variety of barriers (high upfront cost, lack of capital, uncertainty of fuel prices, lack of information, and inertia, and other priorities as barriers) and a range of policies (economic such as subsidies, but also targeted at providing information). This theory may also fit a variety of other barriers affecting intentions and perceived behavioural control and policies that may be used to overcome these barriers.
A number of studies make use of utility functions to structure the decision making logic of agents and form the agent theory. Five studies describe the decision of households to adapt a technology by calculating the maximum utility [31,32,39,43,44]. The use of a utility function is also common in other economic models [27] and allows multiple aspects of the adoption decision to be expressed as utility for the household. This implies that using utility functions enables ABMs to be connected to other types of models. Though utility functions can represent different types of aspects, they are used to model economic barriers such as high upfront cost and lack of capital, essentially focussing on economic aspects of the household decisions. This also translates to economic policies (with a focus on subsidies).
Other theories that are regularly used in these ABMs are network or social network theory [30,33,38,49,51] and the diffusion of innovations theory [45,46]. These theories do not necessarily describe the decision making of households but rather how information, perceptions or innovations spread through a network or population of households. It is possible to combine these theories with others that describe different aspects of the household adoption process. The diversity of theoretical angles and the diversity in purposes (The ranging from exploring the usefulness of ABMs to the tests of particular policy instruments) shows that ABM is a versatile method. It also suggests there still are many opportunities for improving the understanding of energy efficiency policies.

Empirical data
There are two main uses of empirical data in agent-based models: 1) to structure the decision making logic of agents and 2) to provide specific data on properties of barriers, technologies, households and policies. Models that use empirical data to structure decision making logic are different from models that use a theory because the particulars in the dataset make the modelling study highly specific to a technology, arising barriers, and a population of households (and this effect is less prominent when a theoretical basis is used). There are five studies that do not mention the use of a theory to structure the decision making rules [29,[35][36][37]52]. These studies either make their own assumptions about the decision making behaviour of households or base the decision making purely on empirical data.
The use of empirical data to determine the specific properties of households, technologies and policies seems to lead to concrete results about energy efficiency adoption and policy testing. Fourteen of the studies use data from a complementary survey or empirical research for this purpose and eight of the studies extract data from an available dataset. The studies that draw policy conclusions all use data from a survey or statistical study to substantiate their ABMs and results. For instance [30,33,38] all make use of survey data and can compare the effectiveness of different policies on the adoption of lighting technologies, residential heating technologies and wall insulation. Chappin and Afman [30] point out that a ban of incandescent lighting is much more effective than subsidies and Sopha et al. [33] propose a package of policies to stimulate wood pellet stoves and recognize that information campaigns are only effective when combined with subsidies and regulation on wood-pellet fuel prices. Other studies [39,41,42,44,46] use survey or statistical data test the effectiveness of policies in different scenarios. Palmer et al. [39] for instance conclude that the current subsidy scheme is already effective at stimulating the diffusion of solar PV and will need to be extended to push adoption further.
There are also studies in this review that do not make use of empirical data at all. Kowalska-Pyzalska et al. [50] and Byrka et al. [51] do not make use of empirical data and focus on generating insight in how the theory can be translated into an ABM. Rather than focussing on generating insights in the effectiveness of particular energy efficiency policies, these models have as goal to operationalize certain theories. This operationalization of the theory can then be used for academic discussion.
These findings suggest a variety in modelling purposes and that the studies are set up differently according to their purpose. We observe that only papers with empirical data (through a survey) provide concrete suggestions for policy interventions. Decision making theories can be used to shape the structure of the model and also helps focus the relevant empirical data for the issue that is modelled. Using decision theories also enables the reuse of models or model pieces and a comparison of different modelling studies.

Conclusion
In this paper, barriers to the adoption of energy efficiency by households are identified and four clusters have been identified: structural, economic, behaviour, and social barriers. Accounting models and agent-based models (ABMs) are models used to describe the effects of different policy alternatives that address these adoption barriers. ABMs complement accounting models by offering the possibility to model more elaborate decision making logic of households. The literature review resulted in 23 studies on energy efficiency adoption using ABM. The following barriers were addressed in existing modelling studies: a lack of capital, a lack of information, high upfront cost, inertia, ignorance, uncertainty of fuel prices, and other priorities. The set of modelled policies is restricted to subsidies, regulation and taxation, technology ban, household adoption obligation and various information campaigns. These policies directly influence the adoption decision of households and are therefore effectively operationalized in the studied ABMs. Policies that indirectly influence the adoption decision of households, such as policies affecting intermediary parties, have not been modelled before. Many of the models are rooted in the Theory of Planned Behaviour, use utility functions, and/or use empirical data. Overall, the studies show that subsidies help to stimulate the adoption of electric vehicles and alternative heating technologies. Banning incandescent lamps is the most effective policy to increase the adoption of efficient lighting. An obligation for new homeowners to insulate their houses effectively helps to increase the adoption of wall insulation. And informational instruments are not as effective as other policies to stimulate electric vehicle adoption unless combined with a subsidy scheme. These conclusions translate into concrete policy advice specific to a particular technology. Many of the adoption barriers, of the energy efficient technologies and of the energy efficiency policies have not been researched yet with ABM. This seems to be the case because the decision making theories used in the reviewed ABM studies are best suited to describe economic or behavioural barriers. ABM has the potential to provide insight in many more energy efficiency policy questions.
The literature review shows that most of the ABMs on energy efficiency in households rely on empirical data from public databases or survey studies to use as input for the model parameters and agents. Decision making and network theories structure the model logic and directs and focuses the empirical data needed. Models that have a theoretical basis and use empirical data show the strongest ability to test policies on a complex population of heterogeneous households. The value of ABM, for energy efficiency policy making, constitutes of its capacity to produce policy recommendations specific to the adoption of particular energy efficient technologies in the context of heterogeneous populations and specific energy efficiency policies. This capacity complements the policy recommendations that are derived from other types of models like accounting models. Both accounting models and ABMs are founded in extensive empirical data and are therefore suitable to complement each other for the purpose of policy testing and design. In this collaboration, ABMs can specifically contribute to the policy research by focusing on elaborate decision making modelling and the heterogeneity of households.
One pathway for further ABM research is evaluating policies with a more elaborate institutional setting, such as policies that involve multiple stakeholders. Prime examples are policies aimed at intermediary parties, like tradable white certificates, energy efficiency obligations and energy efficiency tenders. By including the complex interactions between intermediary and households in these models, novel and distinctive policy insights about these policies can be derived. Another research opportunity pertains the modelling of other household energy technologies and other adoption barriers than the ones already studied.

Table 5
Energy Efficiency Adoption ABMs.