Effectiveness of behavioural interventions to reduce household energy demand: a scoping review

This paper provides a scoping review of behavioural interventions that target household energy demand. We evaluate 584 empirical papers that test the effectiveness of a behavioural intervention to change behaviour associated with household energy demand. The most studied behavioural tools are providing timely feedback and reminders and making information intuitive and easy to access, followed by (in order) communicating a norm, reframing consequences, making behaviour observable, obtaining a commitment, setting proper defaults, and transitions and habit disruption. The most studied demand-side behaviour is electricity use. There is high heterogeneity in effect sizes. We classified the target behaviours of each study as avoid, shift, or improve behaviours and find that avoid behaviours (in particular, reducing electricity usage) are the predominant focus of researchers. The effectiveness of interventions differs across avoid, shift, and improve responses and by the behavioural tool. Specifically, shifting behaviours are less effectively motivated than avoiding behaviours by using an information intervention but more effectively by using a norm intervention. We review the literature to provide further information about which behavioural tools are most effective for specific contexts. The effectiveness of most behavioural tools are augmented when they are used in the right combination with other tools. We recommend that researchers focus future work on high impact behaviours and the evaluation of synergistic combinations of behavioural interventions.


Introduction
Household energy use accounts for approximately two-thirds of global GHG emissions, directly or indirectly (Ivanova et al 2016), with residential buildings responsible for 60% of building emissions (IEA 2019). Thus, the potential for demand-side reduction to shrink household energy use is substantial (Haas et al 2015, Haberl et al 2017, Masson-Delmotte et al 2018, Ivanova et al 2020. This paper focuses on interventions to change behaviour at the household level, where the connection from intervention to target behaviour and energy savings is direct. We do not evaluate non-residential building emissions or behaviours that are predominantly outside the home (i.e. in the workplace or commercial buildings), as those energy-use decisions often involve mediating parties, additional stakeholders, or policy changes.
The present review aggregates and summarizes existing empirical data to evaluate the scope of this body of literature and highlight gaps and future directions in the study of behavioural interventions to reduce household energy demand. A new chapter on 'Demand, services, and social aspects of mitigation' in the Sixth Assessment Report of the Intergovernmental Panel on Climate Change's (IPCC) Working Group III on Mitigation describes demand-side mitigation strategies in the global effort to reduce GHG emissions while attaining well-being for all (Creutzig et al 2022). While not a panacea and best employed in synergistic coordination with economic, technological, and broader socio-cultural interventions, behavioural interventions targeted at individuals and households are an important demand-side lever (Creutzig et al 2018). To summarize current evidence of the effectiveness of behavioural interventions to reduce household GHG emissions for the IPCC's Sixth Assessment Report WGIII demand side chapter, we reviewed all empirical, meta-analytic, and review papers on the topic. The results expanded far beyond the space limitations in the IPCC chapter and are thus reported in this paper. This paper offers a scoping review of behavioural interventions to reduce household greenhouse gas emissions and an overarching synthesis of the themes, gaps, debates, and consensuses in the literature.
There have been other reviews and meta-analyses of some parts of this literature, but none are as broad as the review presented here (see citations noted by an * in the Reference List and throughout the paper). Much of the literature in this area focuses on developing and testing behavioural interventions for specific target behaviours. Few empirical papers compare the effectiveness of different interventions (i.e. McCalley andMidden 2002, Abrahamse et al 2007). More look at the effectiveness of a given intervention across contexts and domains (e.g. the effectiveness of setting choice defaults, Jachimowicz et al 2019 * ) or conduct a meta-analysis of a specific intervention (e.g. providing smart meters, Ehrhardt-Martinez and Donnelly 2010 * ; or providing informational feedback, Zangheri   Far fewer taxonomies categorize by the more abstract nature of the behaviour change that is targeted in the transition. This has been done in a few ways and may predict the success of the behavioural intervention and the stickiness of the change. In this review we classify the targeted behaviour changes as avoid, shift, or improve responses (ASI) (Hidalgo and Huizenga 2013). This classification was originally developed in the context of transportation services, where avoid strategies reduce energy demand by eliminating trips, shift strategies do so by moving to less carbon-energy demanding modes of transportation, and improve strategies improve the energy efficiency of existing modes (Creutzig et al 2016). In this review we generalize these definitions in the following way: a targeted behaviour change is classified as avoid if the target behaviour is reduced or limited in some way (e.g. running the dishwasher less frequently, turning off the lights), shift if the intervention substitutes a behaviour that uses less energy for another (e.g. take public transportation instead of driving to work, change from natural gas-generated electricity to solar electricity), and improve if a behaviour makes a service less energy intensive (e.g. buying a more energy efficient refrigerator, choosing an ecofriendly dishwasher setting).
Behaviour changes to reduce household GHG demand differ in the psychological and economic efforts needed to make and sustain change. The ASI framework can help organize those changes (Creutzig et al 2018). Avoid may be the most difficult type of response as it asks the decision maker to give up something (e.g. a longer shower or a comfortable house temperature); shift responses may be less difficult because the intervention offers a substitute for obtaining the desired goal (e.g. getting to work by bus rather than personal car); and improve responses may be least difficult because it asks for a change in technology that provides the same (or improved) benefits as before (e.g. using the more energy efficient setting on a washing machine) (Creutzig et al 2022). Financial costs of the targeted behaviour changes are also highly variable and this heterogeneity cuts across ASI categories and behavioural tools, from lowering the thermostat to installing solar panels.
Below, we first conduct a scoping review of behavioural interventions to reduce household greenhouse gas emissions. We then qualitatively review the literature to investigate of the overall effectiveness of interventions. Finally, we discuss the implications of this review for researchers and practitioners.

Inclusion criteria
The search engines used to identify papers were Web of Science, PsychINFO, and Google Scholar. In addition, existing meta-analyses and reviews were used to identify papers to be screened for inclusion. The full search string is reported here: [(home OR household OR residential OR individual OR transp * ) AND (energy OR greenhouse gas OR electricity OR heating OR cooling OR drying OR efficien * ) AND (demand OR reduc * OR conservation OR decreas * OR mitigat * OR low * OR limit * OR) AND (intervention OR intervene OR) OR (field stud * ) OR random OR (control group)]. The initial literature search was conducted in June-July 2020 and two subsequent suppletory searches were conducted in June 2021 and November 2021. Papers in this review were all published in or before November 2021. Papers were included in the review if they (a) contain a behavioural intervention, (b) are published in a peer reviewed journal, highly cited, or included in a previous meta-analysis (including unpublished studies), (c) are original, experimental work, and (d) the targeted behaviour change impacts household energy demand.

Coding
Papers that passed the screen were coded for the following characteristics: (a) the behavioural tool employed, using the eight categories identified by Yoeli et al (2017): setting a proper default, reaching out during a transition, providing timely feedback and reminders, making information intuitive and easy to access, making behaviour observable and provide recognition, communicating a norm, reframing consequences in terms people care about, and obtaining a commitment; (b) whether the study was conducted in a developed country, using the Human Development Index; (c) the energy demand behaviour that was targeted, with the following categories: buying carbon offsets, choice of energy source, electricity use, investment in energy efficiency, or choice in mode of transportation; (d) the targeted type of behavioural change (ASI); and (e) whether the intervention employed an economic incentive with the behavioural tool. Papers were counted for a given behavioural tool if at least one study in the paper used the tool as the primary intervention. Papers that studied the effect of multiple behavioural tools were coded and included in the count for each tool. To distinguish between types of behavioural tools, information interventions must provide some information that goes beyond what would fall into the feedback or norm categories (e.g. an energy saving tip is considered an information intervention, descriptive norm information is a norm intervention, and individual energy usage is a feedback intervention). Table 1 provides an overview of the 584 empirical papers included in the scoping review. The most studied behavioural interventions are providing timely feedback and reminders (258 papers) and making information intuitive and easy to access (246), followed by communicating a norm (158). Choice architecture interventions, probably because they have been identified as tools for behaviour change more recently (Johnson 2021), are the least studied: 29 papers look at defaults and 11 papers focus on transitions and habit disruption. Electricity use is the most studied target behaviour (439 papers), followed by investments in energy efficiency (94), choice in mode of transportation (41), choice of energy source (17), and buying carbon offsets (4). Avoid responses are examined in the most papers (415 papers), followed by shift (112) and improve (77) responses. 72 of the 584 papers included an economic incentive; the choice architecture interventions of reframing consequences (26%) and setting proper defaults (24%) are most frequently combined with an economic incentive.

Scope of the literature
Two 'heat' maps illustrate information patterns from table 1. Figure 1 shows the frequency with which the eight behavioural tools were used to affect either avoid, shift, or improve responses. The tools were not used to the same extent to affect all three types of behaviour change (X 2 (14, N = 584) = 226.15, p < 0.0001). To affect avoid responses, the tools most often used were information, making behaviour observable, norms, reframing consequences, and defaults, whereas the tools most used to affect shift and improve responses were information, norms, reframing consequences, and defaults. Figure 2 shows the frequency with which behavioural tools have been used to target specific energy behaviours. The tools were not used to the same extent for all energy behaviours (X 2 (28, N = 584) = 264.95, p < 0.0001). Electricity use, the most studied behaviour, is most frequently addressed by providing feedback, making information accessible, and communicating a norm. The next most studied target behaviour, investment in energy efficiency, is most often addressed by providing information, making behaviour observable, reframing consequences, and communicating a norm. To influence mode of transportation, the most studied tool is providing information. Figure 2 also shows which combinations of tools and target behaviours have been studied least.
Figures 3(a) and (b) summarize information about the joint application of behavioural tools in the reviewed papers: 57% of papers examined one tool, 34% examined two tools, and 9% examined three or four tools (figure 3(a)). The most common pairs of tools were feedback and information (109 papers), feedback and norms (62 papers), norms and information (40 papers), and feedback and commitment (30 papers) (figure 3(b)).

Effectiveness of behavioural interventions
Behavioural interventions vary greatly in their effectiveness to reduce household GHG emissions. Setting proper defaults is one of the highest impact Table 1.
Counts of papers that experimentally tested the effectiveness of different behavioural tools to reduce household energy demand. Paper counts are broken down by: the behavioural intervention(s); the type of country in the sample; the targeted energy demand behaviour(s); the avoid, shift, or improve response type; and whether an economic incentive is also included in the treatment. In the energy demand behaviour column, the number of papers that targets each behaviour is provided in parentheses. Carbon offset program (1) Energy source (7) Electricity use (123) Investment in energy efficiency (17) Mode of transportation (12)   and note that when people accept the goal or commitment that the experiment proposes, the effect size is larger (d = 0.48). The relatively small number of studies that examine the impact of reaching out during transitions report mixed results ranging from 3% Many interventions employ social elements in the form of norms or peer information, but with mixed results that suggest such tools should be deployed strategically. The Opower study, perhaps the prototypical study of the impact of social norms on household energy consumption, finds 2% reduction in total household electricity use (Allcott 2011). The decrease in energy consumption with repeated norm feedback is sustained even after the intervention ends (

General discussion
A large body of literature investigates behavioural interventions to reduce household electricity use. This scoping review and the many meta-analyses and reviews that have come before show the interest in this topic. What is clear from our scoping review and prior reviews (Pettifor et al 2017 * , Zangheri et al 2019 * , Nemati and Penn 2020 * , Sanguinetti et al 2020 * ), is that some interventions (i.e. feedback, information, and social norms) have been studied more than others (transition and default interventions). The technological and strategic innovations in these highly studied areas and the relatively low cost of their implementation have contributed to their popularity (i.e. digital in-home feedback displays, normative feedback and energy saving information on monthly electricity bills). However, we find that these interventions are often used on behaviour changes that require consistent, repeater behaviour change (i.e. reducing total household electricity).
Many high-impact avoid behaviours (e.g. reducing or eliminating meat consumption or air travel) would be considered very difficult by many Western consumers, and perhaps for that reason are rarely targeted for behavioural interventions. The most common example of avoid responses in our reviews was reduction in electricity use (table 1; figure 2). The nature of avoid change required consistent change, which can be psychologically fatiguing and people can lose interest or motivation. As discussed in the scoping review, more specific intervention targets have larger energy reductions. Previous metaanalyses for practitioners provide additional guidance on how best to develop and deploy feedback, information, and norm interventions to reduce electricity use (Šćepanović et al 2017(Šćepanović et al * , Zangheri et al 2019. Interventions that target general or ambiguous behaviours should be specific and temporally close. There are also sizable gaps in the literature on high-impact emissions behaviours (Wynes et al 2018 * ). Most studies focus on low-impact behaviour changes, such has household electricity use (interventions have found an average 149 kgCO 2 e/year/household reduction, 0.8% reduction of the average American's emissions). There is less work that studies personal vehicles; those interventions that have been studied measure a 571 kgCO 2 e/year/driver reduction, 3.2% reduction of the average American's emissions (WRI 2014, Wynes et al 2018. Studies that examine energy decisions related to personal vehicles mostly examine only the effectiveness of economic incentives (Wynes et al 2018 * ). More work is needed to compare economic incentives to behavioural tools, either on their own or in combination.
Few studies focus on interventions that shift consumers to green electricity (see Wynes et al 2018 * for a discussion of what has been studied), even though this is the household energy behaviour with the highest impact on GHG emissions that people can make (Wynes and Nicholas 2017).
To avoid the worst impacts of climate change, we must reduce global GHG emissions by much more than can be achieved with just behavioural change (Creutzig et al 2018). Behavioural science interventions work most effectively in conjunction with changing the physical infrastructure and political landscape, as a multiplier and facilitator of these interventions. New technologies-rooftop solar or electric vehicles, for example-are only effective if they are adopted and used by a lot of people. The technology and cost of production is undeniably important in reducing GHG emissions, but social influence, for example, can trigger peer effects that expedite adoption and lead to greater GHG reduction (Wolske et al 2020). Feedback and reframe consequences tools are present in many of the effective combinations. Economic incentives have the highest individual average effect, and they appear in the two most effective intervention combinations. Iweka et al (2019) * also find economic incentives, or rewards, to be an effective addition to a behavioural intervention, but the effect only lasts while the reward is being distributed.

Combinations of interventions
Previous meta-analyses have found that interventions that employ a combination of behavioural tools are the most effective. Osbaldiston and Schott (2012) * offer one of the first meta-analyses to evaluate combinations of interventions; they find the following six combinations to be particularly effective: rewards and goals, instructions and goals, commitment and goals, prompts and making it easy, prompts and justifications, and dissonance and justifications. More recently, Grilli and Curtis (2021) * find that the most successful combined interventions include outreach and relationship-building. These interventions fuse information sharing with the social influence of teaching people in a group, they make behaviour observable by intervening on the whole community. Chatzigeorgiou and Andreou (2021) * emphasize the importance of clarity and specificity when designing feedback interventions, both to improve the effectiveness of interventions and to extract concrete measurements of effect from each type of intervention.

Developed countries vs. other countries
This scoping review shows that a large majority of research has been conducted in developed countries, and that there is a paucity of empirical work that investigated the effectiveness of behavioural change to reduce household GHG emissions in developing and pre-industrial countries. A few notable exceptions to this are the following studies of behaviour change in developing countries and emerging economies: (Xu et al 2015, Chen et al 2017, Mi et al 2019, 2020a, 2020b. The field's focus on the developed world is reflected in the behaviours that are targeted for change, most of which are predicated on an unacknowledged set of parameters. Few studies consider households that are not in the affluent West, let alone homes without stable electricity or the financial means to invest in new appliances. There is some justification for focusing this work on countries that are responsible for the highest household GHG emissions-mitigation efforts should be placed at the source of emissions. However, this leaves a large portion of the world to which research insights gathered in developed countries are unlikely to transfer and where research does not exist to instruct policy makers on how to improve household behaviours. There is mixed to negative support for how well household energy interventions work when Climate change mitigation research that has focused on developing countries has largely centered around the development of new technologies, for example cookstoves (Hanna et al 2016). However, without adequate adoption, technological improvements will not have a significant or lasting impact. Even a well-designed, affordable new product is not guaranteed to be adopted, and regional expertise and understanding is critical to developing solutions that people will use (Hanna et al 2016).

Avoid-shift-improve responses
Most work on the effectiveness of behavioural interventions for energy demand reduction has focuses on avoid responses, albeit in relatively easy energy contexts (like reducing household electricity consumption by a small percentage vs. foregoing flying), and relatively less work on shifting and improving responses (table 1; figure 1). Electric vehicles and rooftop solar panel adoption are the main behaviours that have been studied as shifting or improving behaviours. The focus on avoiding behaviour might, in part, explain the temporal decay that is observed in many papers. Many avoid behaviours (e.g. turning off lights or computers) require repeated action and sustained attention and have a low impact for each individual action, making the transition cognitively taxing for a relatively small and unsustained impact (Reeck et al 2022). Interventions that focus on habit change have been most successful for avoiding responses; these techniques have also been applied to shifting responses (Wynes et al 2018 * ).
Shift and improve responses have higher behavioural plasticity, the fraction of nonadopters who could be induced to change their behaviour (Dietz et al 2009). Improving behaviours, such as weatherization and fuel-efficient cars, have the highest potential impact, in terms of emission reduction, and highest behavioural plasticity. Shifting behaviours, such as driving behaviour and car-pooling, have high potential emission reductions and lower plasticity making them more promising targets for intervention studies. . Future work should incorporate methodological best practices of large sample sizes, random assignment, control groups, and field applications. This suggests to practitioners that they should seek out meta-analyses, specifically those that correct for publication bias and review grey literature.

Methodological discussion
Too few studies examine the long-term impact of interventions. Meta-analyses indicate length of the intervention and measurement period impact overall effect size (Khanna et al 2021  *  ). Most studies evaluate the effect of intervention at the end of its deployment, which is perhaps more descriptive of the transitional impact that the complete impact (Zangheri et al 2019 * , Grilli and Curtis 2021 * ). Studies that do evaluate long-term and post-intervention impacts find that effect size decays over time (Allcott and Rogers 2012, Nemati and Penn 2020). The rebound effect, the unintended consequence of people increasing their energy consumption in response to an intervention, should be evaluated by measuring energy consumption for longer periods of time, especially after the intervention has been executed (Alvi et al 2018). There is some evidence to suggest some types of interventions become more effective over time, for example choice architecture interventions and energy audits (Iweka et al 2019 * ). Future research ought to routinely measure longer term impacts.
The behavioural interventions literature has focused on the aspects of household energy demand that are most easy to study. Future work should prioritize high impact behaviours, for example air travel or diet. These are behavioural changes that require avoiding behaviour change and are some of the most difficult to address.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// osf.io/h69bx/?view_only=2847e0b5b42b4aae9758a2 74c14f0cee. Data will be available from 1 January 2022.