What is important for consumers' energy-related decisions? A cross-sectoral systematic review and meta-analysis for the Nordic countries

Review What is important for consumers' energy-related decisions? A cross-sectoral systematic review and meta-analysis for the Nordic countries Behzad Zamanipour * , Ilkka Keppo Department of Mechanical Engineering, Aalto University, Espoo, Finland ARTICLE INFO Keywords: EV adoption Mode choice Heating system adoption Energy-saving measures Systematic review Revealed and stated preferences ABSTRACT The energy transition is shaped by the decisions of individuals. These decisions, in turn, are influenced by a diverse set of factors that have been the object of various studies reported in the literature. Most of these studies have, however, focused on a single decision or a sector. Here, we conduct for the Nordic countries a systematic literature review and a meta-analysis of factors affecting four important consumers' energy-related decisions, namely (1) the choice between an electric or a conventional vehicle, (2) mode choice for personal transport, (3) choice of heating system, and (4) deployment of energy-saving measures at home. We aim to identify the implications that certain factors may have for many of such decisions, potentially encouraging some while discouraging others. Our analysis shows that a group of factors, such as attitude, comfort, costs, living in a detached house, emission implications of the choice, perceived behavioral control, environmental friendliness of the technology, and subjective norm, affect multiple decisions uniformly, in terms of the direction of effect the factor has on environmentally benign decisions. There are, however, also factors for which trade-offs exist, for example, while better public transport infrastructure encourages the use of public transport use, it discourages walking and cycling. We also show differences in outcomes between revealed and stated preferences surveys. Our analysis shows that for electric vehicle adoption, the statistical significance of factors typically increases when excluding the stated preferences surveys. Finally, our findings can aid policymakers in crafting more effective interventions to meet climate targets. 1. Introduction Individuals, through their choices, affect the direction the energy transition will take. As these choices are typically driven by factors going beyond economic and technical ones, scientific discussions have correspondingly gone beyond these disciplines to identify the drivers and barriers of energy-related decisions [1]. These efforts have identified various factors influencing the decision-making process, including financial, contextual, psychological, social, and technical ones [2]. Residential and transport sectors are responsible for more than half of global energy-related carbon dioxide emissions [3,4] and four energyrelated decisions in these sectors have been widely identified as particularly important for the energy transition: (1) Whether to purchase an electric or conventional vehicle, (2) mode choice for personal transport, (3) choice of heating system at home, and (4) implementation of energy-saving measures at home (ref. [5], for example, highlights all four decisions). Various review studies [6–16] have investigated the large body of literature studying the factors influencing these decisions. For example, the acceptance [6] and adoption [10,11,14,15] of electric vehicles (EVs), car and non-car use [13], low-carbon mode choice [12], or mode choice in general [9], and energy-saving measures [7,8,16] have been reviewed. These review studies have, however, been sectoral, typically focusing on a specific, single energy-related decision, and thus have not captured possible wider implications the factors may have across multiple energy-related choices, potentially encouraging some while discouraging others. The review studies have also generally overlooked cultural variety [17] in the evidence, aggregating results from different cultural and geographical areas, sometimes even globally [6], which can lead to conclusions clouded by the cultural variety of the respondents. Also, the previous reviews on this topic have not considered the possible attitude-action gap [18], with much of the evidence reflecting outcomes of stated preferences surveys, rather than revealed preferences ones, and not differentiating the evidence on this criterion. The difference between these two types of surveys stems from the data collection process where the revealed preferences surveys focus on observing actual behaviors or choices that individuals make in real- * Corresponding author. E-mail addresses: [email protected] (B. Zamanipour), [email protected] (I. Keppo). Contents lists available at ScienceDirect Energy Research &Social Science journal homepage: www.elsevier.com/locate/erss https://doi.org/10.1016/j.erss.2024.103861 Received 7 May 2024; Received in revised form 13 October 2024; Accepted 20 November 2024 Energy Research & Social Science 119 (2025) 103861 Available online 28 November 2024 2214-6296/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). world situations, but the stated preferences ones try to ask people directly about their preferences in hypothetical situations. The objective of this study is to fill this gap and analyze the role that specific factors play in the different decisions. More specifically, we aim to answer the following research questions: •Are there factors that are consistently significant for multiple decisions? What are these factors? •Is their impact always supportive of environmentally friendly choices, or does this depend on the choices being investigated? •How consistent is the evidence across the studies for each factor and decision, in terms of statistical significance and direction of the influence? •For which factors is there also empirical evidence (i.e. revealed preferences studies) to be found, and which rely purely on stated preference studies? Do the results change, if only revealed preference studies are considered? To answer our research questions, we carry out an interdisciplinary meta-analysis to improve the accuracy of conclusions and to better understand the contrasting findings from individual studies. Meta-analyses play a crucial role in gathering information within a particular research area, offering a valuable tool for summarizing research outcomes [19]. The necessity of such an analysis in our case lies within the complex nature of decision-making, affected by numerous known and sometimes unknown factors, varying from study to study. For example, each questionnaire is likely to have slightly different questions, collected at different points in time and thus reflecting slightly different decision environments. Thus, to assess what can be concluded from the heterogeneous evidence base from the various studies, it is important to gather evidence to draw conclusions from the available studies. As cultural variation across countries affects e.g. purchase behavior [20], we limit the abovementioned context variation by focusing our review on a set of culturally similar countries, the Nordic countries, 1 to improve the robustness of our results. These countries are also further along in some areas of the energy transition, e.g. for electric mobility [21] and heat pumps [22], and thus studies from these countries are more likely to provide also empirical evidence about the diffusion of low-emission technologies. Nevertheless, we hope the results of this analysis can offer insights into the factors influencing energy-related decisions also for other countries, e.g. by providing a reference point against which the other country can be compared. The rest of the paper is organized as follows: Section 2 presents the methodology of this review and meta-analysis. Section 3 provides the results and discusses the findings of this work. Finally, section 4 concludes the paper and suggests avenues for future research works. 2. Methodology In this section, the literature review stages are discussed. First, the relevance and inclusion criteria for the review are thoroughly explained and the number of remaining articles after each stage is shown. Then, the extraction of data, categorization of factors, classification of factor categories, and at the end the super categorization of factor categories are presented to describe the various steps that have been taken to analyze the data. 2.1. The literature review process The systematic review involves multiple stages based on the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) protocol (Fig. 1), which is a recognized framework used in health, medical, and social sciences to conduct systematic reviews and metaanalyses [23]. In the first stage, a search string incorporating the characteristics of the targeted literature has iteratively been developed and refined for each decision by comparing the emerging search results with the set of literature known to be relevant. The search strings consist of three elements, each focusing on a specific aspect of the studies, namely the decision, factor, and case study (see Supplementary Data). Using the search strings on the ninth of August 2022, holding the assumption that peer-reviewed studies are more dependable, a total of 261 English peerreviewed articles were identified through the Web of Science academic database. The search strings were again used in Web of Science and Scopus databases on the third of July 2023 in order to find the newly published papers and expand the search domain. Then, the duplicates were removed. In the second stage of review, the title and abstract of the studies were assessed while considering the relevance criteria (see Table 1) to ensure the article is about the decisions that individuals or households make and also presents factors affecting those decisions. The final review stage was done based on full-text screening. The inclusion criteria by which the articles have been selected check whether the case study of the articles are Nordic countries, whether the studies are based on a survey, whether the articles present the direction of influence of the factors, and whether the statistical significance level of each factor is available. The fifth inclusion criterion was considered since in the presence of a reference mode/heating system, the other choices are compared to the reference mode, and the factors assessed in the study are examined for their effects on the choice of an alternative mode/ heating system over the reference one. Therefore, as the reference modes differ among various studies, the results of such analyses could not be effectively used in our meta-analysis. Following the aforementioned rigorous process, 53 studies were identified as meeting the predetermined criteria [24–76]. See the Supplementary Data for the list of references existing in each stage of the review. We did not limit the search to any specific period, and the studies analyzed were published between 2006 and 2022. 2.2. Data extraction The following information has been carefully collected from the article after reading them (see Supplementary Data): methods used, year of publication, implementation year of the survey (if available), case studies, type of survey (revealed or stated preferences), number of respondents, the response rate of the survey, technologies and measures, the factors that are not statistically significant, and the sign of the coefficient of the statistically significant factors. A statistical significance level of 0.05 and lower was considered significant. In the process of data extraction, if a source used two different methods with a data set, both were considered separately as individual pieces of evidence. As we aimed to gain a broad insight into the factors for each of the four decisions, and more essentially, find the influential factors that cut across these decisions, every piece of evidence found in the literature that pertains to one of the four decisions is deemed equally important, regardless of the specific choice or population it addresses. For instance, some studies have assessed different populations (e.g. battery electric vehicle users and internal combustion engine users or couples and single-person households) for the EV adoption decision, different trips in the case of modal choice decision (e.g. trips in summer and winter or with different purposes such as work, routine, leisure, shopping, commuting, etc.), and different energy-saving measures (e.g. retrofit measures, energy-saving behaviors, etc.). We included such studies in our body of evidence, just as we did those that did not target specific populations or more narrowly defined decision categories. While extracting the data from the papers, various car and heating technologies and also several transport modes emerge throughout the literature. For EV adoption, the technologies that appear in the literature are EVs in general, Battery Electric Vehicles (BEV), and Plug-in Hybrid Electric Vehicles (PHEV). For this review, we considered all three technologies as electric vehicles but acknowledge that influencing 1 The Nordic countries are: Denmark, Finland, Iceland, Norway, and Sweden. B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 2 factors may differ across the technologies. For mode choice, the evaluated modes in the papers are car, public and active transport in general, internal combustion engine cars in particular, bus, bicycle, and walking. We reflect on the three primary modes of transport: cars, public transport, and active transport, incorporating findings from the other specific transport modes within these categories. For heating system adoption, different names have been used to describe the technologies in the literature. The emerged technologies from the reviewed studies are the following: wood pellets, solid wood-fired, district heating, ground heat, heating by electricity, electric storage heating, exhaust air heat pump, oil heater, heat pump, wood pellet stove, and electric heating. Using the information in the papers, we combined these technologies into five categories of technology which are electric heating (including heating by electricity, electric storage heating, and electric heating), heat pump (including ground heat, exhaust air heat pump, and heat pump), solid heater (including wood pellets, solid wood-fired, wood pellet stove), oil heater, and district heating. Furthermore, two of the papers have discussed the decision to buy a new heating system regardless of the technology type. We evaluated this item along with the others as well. The reason we distinguished between different transport modes and heating system technologies is that the evidence for consumer preferences toward two completely different modes/technologies cannot be aggregated due to their inherent differences, which are arguably larger than differences between various EV types, for example. 2.3. Categorization of factors Having the list of factors emerging from the literature for each decision and observing the similarities between a wide range of factors that have common themes, we categorized the factors affecting each decision Fig. 1. Stages and flow of the literature review. Table 1 Relevance and inclusion criteria for the systematic literature review. Criterion type Criterion Relevance criteria 1. The study should be about the decisions of individuals or households. 2. The study should present factors that influence the decisions. Inclusion criteria 1. The case study should be Nordic countries. 2. The study should be based on a survey. 3. The factors should be presented with their direction of influence. 4. The statistical significance level for each factor should be available. 5. An additional criterion for the mode choice and heating system adoption is that there should not be a reference mode or technology to which the transport mode or heating system is compared. B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 3 to obtain a holistic view of what groups of factors influence the decisions. We categorized some factors into a single category when they are conceptually similar, proxy of a variable, or just worded differently across the studies. For instance, “charging time”and “charging speed”have a similar concept but are worded differently in two different studies, or for example, “price”and “operation cost”both reflect the costs related to car use. The name of a category was chosen in such a way that it can represent the similar concept of the underlying factors. Considering the mentioned criteria, the extracted factors for each decision were categorized (see Supplementary Data). Some factors did not fit into any category and thus formed their own. The evidence for a category is the aggregation of the evidence of each factor within the category. Some factors are defined in such a way that the sign of their coefficient should be reversed to match the other factors in their category. For instance, the category of “charging time”includes two factors: “charging time”and “charging speed”. As the charging speed increases, the charging time decreases, thus, the sign of the coefficient of the charging speed was reversed. The factors that the sign of their coefficient was reversed are tagged with a star in Supplementary Data. We acknowledge the subjectiveness of this process. Tighter scrutiny in the categorization would lead to more factor categories, and to more factor categories constituting just one factor. This would, on average, reduce the amount of evidence for factor categories, making it more difficult to draw conclusions about them. On the other hand, a more relaxed categorization could affect the consistency of the evidence, i.e. the aggregation process itself may show up as inconsistency in the evidence base for a factor category. The factors considered in this study for each factor category are presented in Supplementary Data. 2.4. Classification of factor categories We used the factor categories to analyze the evidence using a classification based on four indexes. The first index is the amount of evidence (i.e. number of studies), assisting us to identify whether the evidence is enough to decide on factor effectiveness. The second index is “statistical insignificance”which is equal to the number of evidence indicating the factor category is non-significant divided by the total number of evidence. This index shows to what degree a factor category has been found non-significant in the surveys. It should be noted that statistical significance is a different term and has its own definition relating to the significance of a variable in a study. Our “statistical insignificance”index, however, shows summary information about the significance of a factor category among the reviewed body of literature. The third index is “directional consistency”which is an index to display the consistency in the direction of influence of a factor category in the body of literature. Using the number of evidence indicating a positive effect (P) and the number of evidence indicating a negative effect (N), eq. (1) shows how this index is calculated. For example, the index will be equal to 1, 0.9, 0.8, …, 0.2, and 0.1 if 100 %, 95 %, 90 %, …, 60 %, and 55 % of the evidence that shows a direction of influence agrees on a specific direction respectively. The index will be equal to zero for the factor categories that the number of evidence showing positive effect is equal to the number of evidence showing negative effect. Therefore, as the index increases, the evidence for a factor category is more directionally consistent. The fourth index is whether there is at least one piece of evidence that is based on a revealed preferences survey. This index informs us if the evidence contains information obtained from empirical data or is only based on the stated preferences of individuals. Directional Consistency =     P−N P+N     (1) Based on the four described indexes, we then divided the factor categories into six classes as follows: Class 1 Influential with also empirical evidence: All evidence shows the factor categories as statistically significant, with a consistent direction of influence throughout the surveys, of which at least one is a revealed preferences survey. Class 2 Influential without empirical evidence: As class 1, but all surveys reflect stated preferences. Class 3 Inconclusive: More than half, but not all, evidence shows the factor categories as statistically significant, and/or not all evidence agrees on the direction of the influence (see Supplementary Note 1 for the exception). Class 4 Non-influential: Half or less of the evidence shows the factor categories as statistically significant. Class 5 Influential, but lack of evidence: Statistically significant, but only one study reports on the factor category. Class 6 Non-influential, but lack of evidence: Not statistically significant, but only one study reports on the factor category. See Supplementary Note 1 for the exact conditions the factor categories should satisfy to be included in each class and further discussion on the classification process to illustrate why such rules have been considered for the classification. 2.5. Super categorization of factor categories Following the categorization and classification processes, the factor categories in the first four classes, which are the factor categories that do not lack evidence, were further categorized into a higher level of categories called super categories (see Fig. 2 and Supplementary Data). The super categorization is conducted across all the energy-related decisions to assess the factor categories that are important for various choices. The aim is to broaden the evidence base, as we use the super categories to analyze the impact of factors on multiple different decisions simultaneously. Similarly to the initial categorization, the factor categories that appear for different decisions and are conceptually close, proxy of a variable, or just worded differently, have been considered within a super category which is named to represent the concept of the underlying factor categories. In addition, as some factor categories show different aspects of a single categorical variable, we created some super categories that represent such categorical variables. For example, “being in Denmark”,“being in Sweden”, and “being in Norway”are considered within a super category of “region and area”, or for instance, because different existing heating technologies are found to be important for choices of heating system, we put all the existing technologies in a single super category called “existing technology”. There are factor categories that did not fall into any super category and formed their own. The factor categories within each super category are presented in Supplementary Data. The review and meta-analysis process is developed by the authors to ensure the replicability of the analysis using transparently defined criteria and indexes. However, there are steps in the categorization and super categorization processes that are subjective in nature. It should also be noted that the screening, data extraction, factor categorization, and super categorization were done by one of the authors, which enabled methodological consistency, however, the other author validated the process through several reviews and examinations. 3. Results and discussion In this section, the results of the systematic literature review are presented. In sub-section 3.1, we lay out the extracted data from the literature and discuss the outcomes of the classification of factor categories to assess how consistent the evidence is across the studies for each factor category both in terms of statistical significance and direction of influence. Then, in sub-section 3.2, we assess the impact of survey type on the effects of several factor categories on the decisions, and how they differ if the survey is based on revealed versus stated preferences of respondents. Finally, in sub-section 3.3, super categories are used to examine their influence on various decisions to find factors that are consistently significant for multiple decisions and understand whether B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 4 Fig. 2. Categorization, classification, and super categorization of the factors. Fig. 3. The evidence base of the review. The evidence is disaggregated in terms of decision (inner ring), geographical scope (middle ring), and survey type (outer ring). RP and SP represent revealed and stated preferences survey respectively. EVA, MC, HSA, and ESM are EV adoption, mode choice, heating system adoption, and energy-saving measures respectively. B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 5 their impact is always supportive of environmentally friendly choices or not. 3.1. Laying out the factors The data extracted from the literature shows how numerous factors have been examined for their possible effect on electric vehicle adoption (EVA), mode choice (MC), heating system adoption (HSA), and energysaving measures (ESM). We found 158, 172, 49, and 68 factors investigated for EVA, MC, HSA, and ESM respectively, followed by the extraction of 1492 data points on the significance and direction of the effect of factors from the body of evidence: 337, 521, 239, and 395 data points for EVA, MC, HSA, and ESM respectively. The evidence base is mainly focused on Sweden (N=53), Norway (N=38), and Denmark (N =14) and to a less extent on Finland (N=7) and Iceland (N=2) (Fig. 3). The distribution of evidence between the decisions indicates the lack of surveys on heating system adoption. Furthermore, although the body of evidence based on empirical data is strong (65 out of 114 pieces of evidence), there is only one revealed preferences study on heating system adoption, highlighting a relative research gap. This also hinders us from effectively assessing the differences between the results of stated and revealed preferences surveys on this decision. This categorization process yields 102, 113, 42, and 57 factor categories for EVA, MC. HSA, and ESM respectively. Subsequently, the classification to the six factor category classes has been done, as described in the methodology section. The influential factor categories (Class 1 and Class 2 as described in Section 2) are shown in Table 2, shown for the four different decision categories, with mode and heating system choice further split based on modes and technologies, respectively. A considerable number of the factor categories have not been considered in many studies (see Supplementary Fig. 1), and this applies also to the factor categories that are influential. Further investigations on the factors with a lower number of evidence would increase confidence in the interpretation of outcomes and could also yield different results. There are, however, some factor categories in Class 1 and Class 2 with more evidence supporting the conclusion about them. The factor categories that have “high”or “very high”number of evidence (>6 pieces of evidence) –and include also empirical evidence –are functional barriers for EV adoption, and having access to car and public transport infrastructure (e.g. bus, metro, and train stations), affecting the choice of public and active transport respectively (see Supplementary Data for the factors considered in each factor category). The evidence base is also strong for the cost of heating system adoption, but without any empirical data. Although many factor categories are supported by the empirical data, there are also a few that have not been investigated by the revealed preferences surveys. Table 1 excludes factor categories that were considered inconclusive (Class 3). Considering the rather strict criteria we have established for our influential factor categories, with all the evidence having to be fully supportive of the conclusions about significance and direction, and taking into account the effect our aggregation of factors into factor categories may introduce to these metrics, we analyze the factor categories in this class further. Some of these factor categories may, in reality, be influential, but the processes of data extraction (which might have been affected by the potential Table 2 fallacy in the underlying studies [77]) and categorization have affected the outcome, consumer behavior and perception of technologies may have changed over the time range covered by the studies, attitude-action gap may affect the classification, and biases such as publication bias [78] might impact the statistical significance of factor categories (see also Supplementary Note 2 and Supplementary Table 2). Due to this, we want to further investigate this class and its evidence base. In Fig. 4, we depict metrics for the inconclusive factor categories, for “directional consistency”and “statistical insignificance”. The closer the directional consistency and statistical insignificance indexes are to one and zero respectively, the closer we are to the factor categories being classed as influential. On the other hand, the closer the statistical insignificance index is to 0.5, the more likely it is to be considered noninfluential. Fig. 4 depicts the inconclusive factor categories for EV adoption, with information also about the size of the evidence base, presence of empirical data and direction of the influence. Similar figures for the inconclusive factor categories of the other decisions are available in Supplementary Data. The results show that directional consistency is not the main reason that prevents most of these factor categories from being considered influential, but the statistical significance is. The direction of effect for many factor categories has been consistent in the literature, however, there is a considerable amount of evidence suggesting statistical nonsignificance for the inconclusive factor categories (see Supplementary Fig. 2). For instance, in Fig. 4, the directional consistency for the majority of factor categories is equal or close to one, meaning that most of the evidence base is in favor of single direction of effect. However, except for charging infrastructure, education, incentives, personal norm, car costs, and environmental behavior factor categories, the statistical significance of none of them has been approved by >70 % of the evidence base. Also, the figure shows that the factor categories demonstrating higher significance and consistency have usually been investigated by many studies, unlike the others with a weaker evidence base. This suggests that further studies on the latter could move these factor categories toward the left –or drop them to Class 4 (NonInfluential). As for the non-influential factor categories (Class 4), most of them have at least one piece of evidence suggesting their statistical significance, but with the majority of evidence showing the opposite. Several factor categories have five or more pieces of evidence, suggesting that they have broadly been considered as potentially significant for driving decisions, but in the end, mostly turned out not to do so. For example, environmental concern and income for EV adoption and energy-saving measures both. All such factor categories can the found in the Supplementary Data. There are also some factor categories in the non-influential class that are never found to be statistically significant. Examples include, for instance, gender for using public transport and house size for energysaving measures, with >4 pieces of evidence supporting them. Supplementary Table 1 shows all non-influential factor categories that are statistically non-significant throughout the data. Finally, the fifth and sixth classes encompass the factor categories that only have one piece of evidence supporting or rejecting their significance, which we consider in our classification as inadequate. These classes can, in our categorization, be considered as experiments conducted in only few studies, but which require repetition to increase the confidence in the findings. The full list of factor categories for these classes is available in Supplementary Data. 3.2. Revealed versus stated preferences The difference between what people say and what they actually do [14] may explain some of the contradictions observed for the inconclusive and non-influential factor categories throughout the literature. To investigate this hypothesis, we assessed the difference between directional consistency and statistical insignificance derived from (1) only empirical evidence and (2) all evidence for the inconclusive and non-influential factor categories (Class 3 and 4 above) that have at least two pieces of revealed preferences evidence and one piece of stated preferences evidence (Fig. 5). As shown, using the evidence from the revealed preferences studies generally improves directional consistency. This is, however, not the case for all factor categories, meaning that empirical observations sometimes provide more contradictory evidence about the direction of the effect than the combined data does. The general direction of the evidence from empirical and all evidence remains the same, except for B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 6 Table 2 Influential factor categories (Class 1 and Class 2, see Section 2.4). The multiplicity of evidence (i.e. unweighted number of studies) is shown by four levels based on the number of evidence; Low (2–3), Medium (4–6), High (7–9), and Very high (10−12). Decision/Choice Factor category Direcon of effect Empirical evidence Mulplicity of evidence Electric vehicle adopon Access to second home + 9 Low Business or economics college degree-male (interacon factor) + 9 Low Detached house + 9 Low Immigrant parents + 9 Low Male government employee (interacon factor) + 9 Low Male municipality employee (interacon factor) + 9 Low Region of residence +/-† 9 Low Technical college degree-male (interacon factor) + 9 Low Travel to work 15-100 km + 9 Low Travel to work 5-15 km + 9 Low Wealth + 9 Low Female government employee (interacon factor) - 9 Low Female municipality employee (interacon factor) - 9 Low Male immigrant from high income country (interacon factor) - 9 Low Travel to work >100 km - 9 Low Atude + 8 Medium Boot size (trunk space) + 8 Low Being in Sweden + 8 Low Environmental norm + 8 Medium Personal Innovaveness + 8 Low Status and excitement + 8 Medium Subjecve norm + 8 Low Charging time - 8 Low CO2 emission - 8 Low Funconal Barriers - 8 Very high Mode choice-Car Comfort + 9 Medium Habit + 9 Medium Number of children 7-12 years old + 9 Medium Number of legs in trip chain + 9 Medium Perceived behavioral control + 9 Medium Subjecve norm + 9 Low Winter + 9 Medium Minimum density at desnaon - 9 Medium Age squared - 9 Medium Mode choice-Public transport Atude + 9 Low Public transport infrastructure + 9 Low Car access - 9 High Travel me - 9 Low Mode choice-Acve transport Affecve moves + 9 Low Atude + 9 Low Mix in the main uses of buildings-home - 9 Low Mix in the main uses of buildings-school - 9 Low Number of housing units per hectare-home - 9 Low Number of housing units per hectare-school - 9 Low Number of residents per sq km-home - 9 Low Number of residents per sq km-school - 9 Low Proporon of land covered by public buildings-home - 9 Low Proporon of parks and recreaon areas-home - 9 Low Proporon of parks and recreaon areas-school - 9 Low Public transport infrastructure - 9 High Rao of signalized intersecons to total number of street interseconsschool - 9 Low B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 7 the factor category “number of cars in household”, which affects EV adoption positively based on empirical data and negatively with all evidence combined. Use of empirical evidence alone generally improves the statistical insignificance indicator (lowering it) for factor categories, especially for the choice of EV adoption. Three factor categories of EV adoption, namely incentives, personal norm, and children in household, are found to be influential (Class 1) after excluding the evidence from the stated preferences studies. This is also the case for perceived behavioral control for choice of public transport. At least for these factors, the revealed preferences studies have thus been able to identify the influence of factors more effectively. Naturally, the empirical evidence base for such factor categories is now even lower than the previously combined body of evidence. In terms of our key interest, i.e. factor categories affecting multiple decisions, age, gender, and income are more statistically significant for EV adoption and energy-saving measures, when only revealed preferences studies are considered. However, this is not the case for the other factor categories (see Supplementary Data). 3.3. Common factors across the decisions Our initial factor categorization is decision-specific, so that we can capture also factors that are relevant only for certain decisions. In order to assess the factor categories that may be important for various decision contexts, we have further aggregated them into “super categories”(see Supplementary Data) as described in Section 2. Doing this allows us to identify broader super categories that affect decisions for more than one choice (Fig. 6). Specific super categories are represented in Fig. 6 in both, panel a and panel b. This means that while the super category is influential (Class 1 and Class 2) for some decisions (shown in panel a), it isn't that (Class 3 and Class 4) for others (those in panel b). In panel b the direction of the effect is defined only from the majority of pieces of evidence that are statistically significant. There are super categories of factors in panel a of Fig. 6 affecting multiple choices uniformly in terms of direction of influence and statistical significance, namely attitude, comfort (e.g. ease of taking action and/or use of the technology), costs, living in a detached house, emissions following the choice made, perceived behavioral control, environmental friendliness of the technology, and subjective norm. Perceived behavioral control and subjective norm were found to be influential in the transport sector decisions, affecting EV adoption and mode choice for car use. Furthermore, while costs and environmental friendliness are found to be important in the residential sector for the choice of various heating systems, attitude, comfort, residing in a detached house and tendency to adopt a technology with lower emissions encourage the environmentally benign choices cross-sectorally. As before, the number of studies is, however, fairly low for some of these, Rao of 4-way intersecons to total number of street interseconsschool - 9 Low Summed floor space in all the buildings divided by buffer area-home - 9 Low Summed floor space in all the buildings divided by buffer area-school - 9 Low Total road lengths divided by the drawn school journey length-school - 9 Low Buying a new heang system Buyer of environmentally friendly products + 9 Low Age of residence + 9 Low Apartment - 9 Low Level of sasfacon with current heang system - 8 Low Electric heang Comfort of use + 8 Low Electricity is exisng technology + 8 Low Environmental friendliness + 8 Low Outside air heat pump is available + 8 Low Heat pump Comfort of use + 8 Medium Environmental friendliness + 8 Medium Ground heat is exisng technology + 8 Low Solar panel/solar water heater is available + 8 Medium Cost - 8 Very high Repeon decision strategy - 8 Low Emissions - 8 Low Indoor air quality - 8 Low Solid heater Comfort of use + 8 Medium Environmental friendliness + 8 Medium Living in a sparsely populated area + 8 Low Solid wood fired is exisng technology + 8 Low Oil heater Cost - 8 Medium District heang Comfort of use + 8 Low Environmental friendliness + 8 Low House is in district heang area + 8 Low Cost - 8 High Energy-saving measures Atude + 9 Medium Being in Denmark + 9 Low Detached housing + 9 Medium Number of household members less than 12 years old + 9 Low General environmental atude + 8 Low Personal norm + 8 Medium Residence in urban area - 8 Low †Depending on the region of residence it can either have a positive or negative effect. B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 8 Fig. 4. Inconclusive factor categories for EV adoption. The diameter of the circle illustrates the size of the evidence base, letters p and n are for positive and negative effects in the majority of the statistically significant evidence base for the factor category (in turn shown by the number after the comma). Red text means there is no empirical evidence for the factor category, and black that there is at least revealed preference study for the factor category. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Fig. 5. Comparison of results between empirical and full evidence bases for the inconclusive and non-influential factor categories. The axes show the difference between revealed preferences evidence (+) and combined evidence (−). For three factor categories that have no statistically significant evidence based on the empirical data, namely social effects for EV adoption, subjective norm for mode choice-public transport, and age for energy-saving measures, only the change in the statistical insignificance is shown. B. Zamanipour and I. Keppo Energy Research & Social Science 119 (2025) 103861 9