How Viable are Energy Savings in Smart Homes? A Call to Embrace Rebound Effects in Sustainable HCI

Realising the energy-saving potential of smart homes can be challenging: Darby [26] points out the scarce number of evaluations under real-world settings in favour of simulations and evaluations under controlled laboratory conditions. This imbalance can be problematic and lead to an instantiation of the well-documented energy performance gap between ex-ante energy saving predictions and ex-post assessments [36], due to e.g., contextual factors, building and installation design parameters, and occupant behaviour not being taken into account [115]. In 2018, Hargreaves and Wilson [41] reported the results from a study of smart home technology in the real world: based on a 9-month field project with ten UK households, they found that the technology can even lead to overall energy intensification through, e.g., increased comfort expectations and pre-warming of domestic spaces. Additional challenges that may impact the energy-saving potential of smart homes include their limited uptake (due to unawareness or inaccessibility [108, 124] and rebound effects. Tirado Herrero et al. [108] conclude that “in contrast to aspirational claims for a ‘smart utopia’ of greener, less energy-intensive, and more comfortable homes [...] SHTs [smart home technologies] may reinforce unsustainable energy consumption patterns in the residential sector”.

The HCI community specifically has a longstanding record of work that talks in nuanced ways about how smart homes can undermine energy savings by introducing new lifestyle expectations and changing behaviours. This includes the identification of socio-technical challenges around integrating new technologies into everyday routines and practices [24, 49] and the often energy-intensive visions for smart homes. Commonly associated with concepts such as desirability, convenience (including effortless energy savings) and control, these concepts can undermine the sustainability promises of smart homes [51, 104]. Gendered and cultural expectations [25, 105] and individual preferences [50] add another layer of complexity, but are not consistently acknowledged. As Strengers [103] points out, many studies have focused on “resource man”, a default male technology-savvy energy consumer who makes rational decisions about energy use.

From an energy justice perspective, smart energy technology can also be problematic. Research finds that, despite their desire to monitor and regulate their energy usage, many low-income households do not receive feedback on their consumption [27]; a problem that is reinforced through e.g., the high cost of sensing and feedback technologies that can support energy savings [53, 67]. Calls to provide easier and subsidised access to these technologies [82] are thereby potentially in tension with concerns about data privacy and security that relate to their widespread use [53, 67]. Working class people might also end up paying more for their energy despite having e.g., smart meters installed as they cannot adjust their schedule to use energy when it is cheaper [53]. In this context, it will be critical to make energy feedback and energy-saving support accessible and affordable across society. At the same time, care must be taken not to rely on smart technologies and smart homes as the solution. Our work specifically focuses on the limitation of rebound effects (as one of the reasons to challenge this reliance) and calls for a more socio-technical understanding of these effects. This understanding can help pave the way for policies that reduce the ineffective application of efficiency technologies while also reducing inequalities.

The umbrella term ‘rebound effects’ describes a rich class of mechanisms occurring when goods or services become easier accessible or more attractive, leading to their increased demand, therefore their increased consumption and production, and thus ultimately increased energy consumption, material consumption, and greenhouse gas emissions. The first such mechanism yielding rebound effects was described as early as 1865 by British economist William S. Jevons. In his book “The Coal Question”, Jevons notices that although coal extraction was becoming ever more energy efficient at the time, overall not less but more energy was used in coal extraction [52]. The efficiency gains of coal extraction were thus a victim of their own success: by making coal more affordable, they were making it attractive for an increasing number of applications and customers, while at the same time making the process of coal extraction more scalable.

In research fields such as energy policy, rebound effects are a firmly established concern [90, 117]. As the strength of the rebound effect varies significantly across contexts, a key research goal in these fields is to identify causal factors that allow a concrete, context-specific estimation of rebound effects, to then direct and focus possible intervention points [1]. By understanding causal relationships and context-specific factors, it is argued that energy-focused research can be targeted and effective. The methods used to acquire this understanding include modelling and simulation [71, 118] as well as empirical studies [18, 38] that take both macro-level policy perspectives and micro-level household perspectives. Some of these studies specifically highlight the need to consider rebound effects in smart homes. For example, one study of the rebound effect in smart homes suggests that “the rebound effect on energy use should not be underestimated” [18]. It is important to understand what causes rebound effects, and how large they are, to direct research priorities [118]. For all these reasons, researchers should give “due consideration to the full range of behavioural responses to technical efficiency” [1] when conducting energy-focused research on smart homes. After all, the actual energy consumption of a smart home is not determined by its technical components alone but must be understood as a socio-technical system.

A substantial part of the literature on smart homes, however, originates in engineering sciences. And although it has been shown that digital technologies are particularly prone to rebound effects—leading even to the emergence of the new term ‘digital rebound’ [22]—it has not yet been thoroughly analysed to which extent the computing communities at large, and that on smart homes, in particular, have engaged with the concept of rebound effects.

In 2011, Kaufman and Silberman [54] encouraged the sustainable HCI community to consider rebound effects to mitigate the problem that “proxies generally used in evaluation (e.g. less energy or water consumption) may be poor indicators of such systems’ effect on emissions because of effects outside the scope of analysis”. At this point, the authors were “not aware of studies examining the degree to which the effects of sustainable HCI technologies rebound”. Assessing whether this has changed over the past decade, and how the level of perception within HCI compares to the one rebound effects receive in computing generally and, even broader, in all energy efficiency-related fields, has been a key motivation behind this work.

Energy efficiency is a vast area of research across all fields, with Scopus returning almost a quarter-million papers in Search 1 alone. In ScienceDirect and IEEE, the numbers are slightly lower but still substantial with around 40,000 and 66,000 search results for ‘energy efficiency’ respectively. In ACM’s full text database, there are significantly fewer results (almost 7,000), while for the SIGCHI library, energy efficiency is clearly not a crucial aspect, with only 33 results in title, abstract, or keywords. Smart home research is more evenly distributed, with approximately 15,000, 1,000, 7,000, and 3,000 papers appearing in Search 1 for the four large databases, and 200 even in ACM’s SIGCHI subset. By contrast, the number of papers mentioning ‘rebound effects’ is limited to a couple thousand at most, almost all of which appear outside of the computing databases: Scopus, ScienceDirect, IEEE and ACM contain 2,714, 1,114, 79, and 7 papers, respectively, and SIGCHI none whatsoever (see Table 1, row 3).

Strikingly, not a single paper in the field of computing (neither IEEE nor ACM) simultaneously encompasses all three topics based on Search 1. Scopus and ScienceDirect yield the small but non-trivial numbers of 4 papers and 1 paper, respectively (see Table 1, row 7). Proportionally, the highest percentage of papers on ‘energy efficiency’ that mention ‘rebound effects’ can be found in Scopus (0.24%) and ScienceDirect (0.84%), covering several hundred publications; with less than 1%, the percentages are still quite low. Not remotely as low, however, as for the computing literature. As Table 1 shows, for general energy efficiency, the ratio of papers focusing on rebound effects is only 0.02% and 0.01% for IEEE and ACM, respectively—one to two orders of magnitude lower than in the two general literature databases.

In principle, this could be an artefact of emphasis: perhaps computing papers engage with rebound effects but consider them important only within the context of discussing work in the body of text? But the results from Search 2 show that this is unlikely. As Table 2 documents, they are broadly consistent with the findings above. For instance, in the thousands of ScienceDirect papers on ‘energy efficiency’, 1.7% mention ‘rebound effect’. In contrast, IEEE and ACM yield fewer papers—251 (0.14%) and 43 (0.23%), respectively (see Table 2 row 5).

As smart homes are an important yet relatively new area of research, it is possible that rebound effects have started to receive attention only in the last couple of years and that we are on a path towards more holistically sustainable research. However, Figure 3 shows that this is also not the case. Although our selected databases depict a net increasing trend in smart home energy efficiency research, most papers do not mention rebound effects. It is evident from Figure 3 that, since 2011, smart home research has not actively engaged with rebound effects, i.e., yielding zero papers annually across large databases; whereas in other cases, the trend in engagement is stagnant. Scopus, for instance, yielded a single paper for each year from 2017 to 2020 in Search 1: clearly, the engagement is not increasing with time. Albeit on quite a low overall level, SIGCHI displays a more consistent (if stagnant as well) rebound engagement. This highlights a critical gap, which is also consistent across Search 2. In particular, Search 2 delineates a significant growth in smart home energy efficiency research in the last ten years; however, such growth is not evident in the ESR intersection. In addition, SIGCHI yielded a greater proportion of research mentioning rebound effects, yet we could not find any signs of growth in attention paid to rebound effects.

The SIGCHI library represents a clear outlier compared to the other databases: from the (relatively few) 278 energy efficiency papers in Search 2, 10 (3.60%) address rebound effects. This trend seems even stronger for SIGCHI’s energy efficiency research in smart homes: From the 41 papers addressing this field, 6 (14.63%) consider rebound effects—a vastly larger percentage than the 0.45% in IEEE and 1.73% in ACM overall. This raises the question how exactly the papers returned in Search 2 for this intersection consider rebound effects. Table 3 summarises the papers from our SIGCHI library that contain the smart home, energy efficiency and rebound effect keywords (ESR intersection) at some point in their full texts². A closer reading reveals that these papers refer to rebound effects but typically do not place them at the core of the analysis. For example, a recent study of smart meters in use in households discusses energy savings at length but only mentions rebound effects in passing: Sometimes “users wanted to heat the house a bit more than they would do normally in order to benefit from low prices (typically termed the rebound effect)” [3]. Pricing was much more on participants’ minds than efficiency. Another study posed an open question about rebound effects as an afterthought: “How to deal with rebound effects, to ensure efficiency gains are not used in additional emissions elsewhere?” [2].

What can we learn from the small number of papers that focus on rebound effects in a smart home energy efficiency context? We first look at the papers from the general literature outside of computing which centrally address rebound effects. These papers, listed in Table 4, have all been published since 2017. Together with the insight that they aim to answer fundamental research questions (e.g., “Does the rebound effect exist in smart homes and what is the size of the rebound effect?” [18]), it becomes apparent that the topic has only recently started to receive attention and that many important, especially more nuanced questions have not yet been addressed. The computing community in particular does not seem to have this on its agenda, as none of the papers in Table 4 was published at a computing venue. The more slippery indirect and structural rebound effects (see Section 2.2.2), however, and their often subtle and intricate consequences (both environmental and societal) remain largely unaddressed across research fields; Section 5.3 touches on this topic.

The papers from Table 4 point to some interesting behavioural dynamics, effect sizes and gaps in the literature, some of them of potentially high relevance for HCI. Walzberg et al. [116], for example, conducted an in-depth analysis of rebound effects, including the distinction between direct and (especially causal factors and dynamics of) indirect effects in smart homes, and the importance of suitable metrics for smart electricity management. The authors suggest that rebound effects do not “cancel out all the benefits of energy efficiency”; in other words, that smart home efficiency gains do not result in backfire. They advise and promote nevertheless the consideration of factors that decrease rebound effects in smart home design, especially related to user behaviours that tend to trigger them during certain periods; considerations that seem highly relevant for the HCI community. In another study that investigates rebound effects in smart homes, student participants were invited to use smart home software simulations [18]. For both of their experimental setups, the results show significant rebound effects. However, as they were using simulations rather than energy data gathered in a real-world smart home context, the ecological validity of this study is low, a fact that the authors seem unaware of; follow-up work is needed to confirm the findings outside of a lab setting. The only study in Table 4 that was carried out with occupants in the context of their everyday life is the one by Beltram et al. [9]. While their iteratively improved algorithms led to better system performance, no experimental conclusions, especially in relation to the rebound effect, were drawn.

To support HCI research, we offer here a taxonomy that synthesises findings from our literature mapping and the wider literature to provide concrete starting points for the HCI community to engage with and further the topic of rebound effects. We envision this taxonomy as the beginning of a much-needed, systematic stream of research to develop guidelines and conceptual frameworks to account for rebound effects both reactively and prospectively:

The taxonomy is shown in Table 5 and takes the form of a systematic array with concrete opportunities for HCI to effectively apply their skillset towards the identification, measurement, explanation and mitigation of direct, indirect and structural rebound effects, which are described in Section 2.2. Examples of such opportunities are gaps and calls in the literature, as well as challenges and relevant methodological suggestions by type of rebound.

Across all types of rebound, Table 5 shows that more empirical research is called for, e.g., [61, 93, 111]. While the complex and systemic nature of rebound makes this a challenging task, the literature offers concrete tools to tackle it, including but not limited to quasi-experimental approaches [98], living lab research [15] and systems thinking [75]. Covering both quantitative and qualitative research skills, these approaches fall under the expertise of many HCI researchers. And the field has, in fact, acknowledged the importance of rebound in its research: even when they are complex and not yet sufficiently understood, HCI researchers have been encouraged to engage with rebound effects when designing, deploying and evaluating ICT systems [72, 87, 120]. So not only can they provide insights for policy makers and other research fields, but they can improve the outcomes of their own projects as well.

Several studies mentioned in Table 5 point to the potential to better understand the causal factors of rebound and to identify and evaluate rebound patterns that can inform research and design. For example, rebound effects are enabled and shaped by such features as flexible uses, the potential for new types of uses that permit induced demand, and the savings of resources that can otherwise be redeployed [23]. On the other hand, they are constrained by factors such as natural caps or limits on objectives, such as comfortable temperature zones for human living or existing real-estate costs [22]. After all, rebound is not an arbitrary randomizer. In many cases, natural caps exist that place constraints on the amount of rebound we can reasonably expect. Research in energy policy provides conceptual guidance for methods and macro-level views [95] that can structure and inform HCI research. For example, there is a significant difference between lighting efficiency gains and heating gains in that lighting has additional uses and therefore is subject to induced demand, whereas heating is limited by our temperature comfort zones.

Digitalisation-triggered rebound occurs both within the ICT sector, and also outside it, in society and economy at large. The inner-sectoral rebound is represented by e.g., the induced material and energy consumption of smart technologies such as those deployed in smart homes. It is also reflected in Koomey’s law [59] that shows that the exponential efficiency gains of computation (as reflected by the well-known Moore’s law) have not led to close-to-zero energy consumption for computation, but to an equally exponential growth in the complexity of computations performed—in other words, a direct rebound effect of 100%. Our taxonomy and the HCI opportunities presented in Table 5 are agnostic to whether the rebound triggered by digitalisation occurs within or outside the ICT sector, although the larger need, and thus opportunities, might present themselves across all other sectors.

Following a set of critiques papers, there is evidence that the sustainable HCI community has recently moved away from persuasive and efficiency-focused projects (which accounted for the majority of publications in the field in 2010 [28]) and further embraced third-wave, qualitative work [13]. This work is often more exploratory in nature, with the aim to gather insights rather than to realise pre-determined behavioural or efficiency targets. It also captures the community’s understanding that the deployment of technology does not happen in an (easily modelled) vacuum; instead, it is situated within the complicated everyday lives of people who are themselves shaped by cultures, demands and individual differences—and it often is the situatedness that makes energy efficiency measures backfire. Communicating this understanding to policy-makers, other stakeholders, and potentially more technically-oriented fields could prevent at an early stage misleading research outputs in those fields as well as potentially misled policy measures and incentives as a result of such incomplete analyses. After all, our literature mapping has shown that energy efficiency (in smart homes and beyond) is a vast area of research; Scopus returned almost a quarter-million papers for Search 1 alone.

In particular, the lack of studies that focus on occupants in the context of their everyday life presents a unique opportunity for HCI research that has commonly been described as research ‘in the wild’ [17, 89]; due to the term’s colonial connotations and associated exploitative practices [63, 100], we, the authors, are opting to instead describe such research as in situ, a term also often deployed in HCI research [17, 20]. In-situ HCI research can help other disciplines understand the impact of rebound effects by studying situated behaviours in smart home environments. As established in section 2, HCI researchers come equipped with expertise on both efficiency work and the nuanced ways in which socio-technical factors undermine energy savings in smart homes. Taken together with our finding that HCI shows more awareness of rebound effects than all of the other research communities we included in our search, this puts the HCI community in a position where it can make a real difference: if digital technologies do not (consistently) enable the envisioned efficiency savings, this needs to be acknowledged when climate change mitigation strategies are selected.

For various methodological reasons, assessing the indirect effects of digitalisation (both the environmentally beneficial and detrimental ones) is challenging [21], and rebound effects are no exception: “Environmental impacts that arise when technologies co-evolve with everyday practices are not easily predictable. This seems to be one reason why the existing literature [...] contains relatively few or vague recommendations to policy-makers and other stakeholders” [16]. From the four promising methodological approaches put forward by Börjesson Rivera et al. [16], two are the scenario method and ethnographic methods as well as anthropological theory, respectively. HCI researchers come equipped to apply these methods, arguably better than their colleagues in any other computing discipline.

The many factors that drive rebound effects (and their interplay) add to the complexity of the required research and likely constitute one of the reasons why there exist few empirical studies to date. To deal with such complexity, a shift from computational thinking to system-level thinking has been encouraged among technologists: “the failure to think systemically is a critical weakness in our understanding of the transformations needed to achieve sustainability” [30]. Besides contextually embedded in-situ research, HCI methods can help to explore and speculate about potential impacts of smart home technologies, including their undesirable consequences, within and beyond the constraints of contemporary technologies and social life (e.g., [48, 97, 119]). And they can be used to successfully implement top-down measures (e.g., stricter climate policies as in ‘Green Policy Informatics’ [13]), which will likely have a key role in climate change mitigation.

Clearly, the ongoing digitisation seems not to have mitigated our environmental issues, as its globalising and accelerating rebound effects probably outweigh the benefits. As Santarius [91] notices, “Humanity’s ecological footprint keeps growing although we have already digitalized significant parts of our economy and society over the past years. It seems that digitalization is not relaxing but rather reshaping societal metabolism in a way that tends to rebound on global energy and resource demand: Gains in efficiency are more than outweighed by the increase in consumption due to new digital services or falling prices caused by more efficient production processes” [91]. A paper fittingly subtitled Why Digitalization Will Not Redeem Us Our Environmental Sins adds: “Digitalization is unlikely to be the environmental silver bullet it is sometimes claimed to be. On the contrary, the way digitalization changes society, making it ever faster, more connected, and allowing us unprecedented levels of efficiency might in fact lead to a backfire [...] For most manifestations of digitalization, a strong digital rebound seems to be the rule rather than the exception” [22].

Together with other factors, rebound effects obstruct the hopes that digitalisation and other technical innovations can ‘dematerialise’ economic activity from resource consumption so that the economy can keep growing while its material footprint remains constant or shrinks. Other factors include the systematically underestimated impact of services, the limited potential of recycling, as well as insufficient and inappropriate technological change [74]. Rebound effects are one of these central factors standing against the feasibility of absolute decoupling, and they require “an in-depth and systemic consideration and anticipation [...] in the design of sustainability policies” [74, p. 40]. Unfortunately, according to all available evidence, “it is safe to say that the type of decoupling acclaimed by green growth advocates is essentially a statistical figment” [74, p. 31], [32, 39, 44, 45, 114, 122]. Together, these insights reinforce the message that a more profound shift is needed to truly advance toward sustainable societies [43, 58, 66, 68, 72] and confirm how important it is to recognise and evaluate the effects of technology-focused interventions through a systemic view.

In this context, the combination of efficiency and sufficiency strategies could form the basis of the profound shift that is called for: When technical efficiency concurs with limited input resources, real progress could be made towards long-term sustainable resource usage [46]. Digital technologies could then be used to replace or improve existing activities without spurring additional activities. In fact, Freitag et al. [34] argue that “[i]f a global carbon constraint was introduced, we could be certain that rebound effects would not occur, meaning that productivity improvements through ICT-enabled efficiencies both within the ICT sector and the wider economy would be realised without a carbon cost”. Such a nondiscriminatory approach that simply imposes global carbon constraints, however, can perpetuate or even intensify existing disparities, as under-represented or marginalized communities would be disproportionately affected [99]. Well-meant measures to mitigate this risk can themselves have paradoxical counter-effects in turn, such as anti-racist campaigns unwillingly reinforcing the very disparities they set out to address [62] or even the concept of energy justice itself, which often “fails to adequately account for intersecting dimensions of power and inequality, such as gender, race, class, Indigeneity, ethnicity, sexuality, ability status, colonial history, and caste, among other statuses within the world system” [99]. Rebound effects, and in particular the elusive indirect rebound with potential ramifications throughout the entire society, could be particularly prone to amplify existing disparities. The call for sufficiency and effective climate policies is not novel in the computing communities, e.g. [13, 46]. Our call for HCI, however, is to shine light on the critically important topic of rebound effects from a systems thinking perspective [121], but not one focused exclusively on environmental matters, but one with justice considerations at its very core.

Like any systematic mapping study, the scope of ours by necessity exhibits some limitations. The databases do not cover all existing literature on smart homes, energy efficiency and rebound effects. This means that we may have missed some of the published work in these areas. Second, it is unclear to what extent the result sets from the different databases for each search overlap (with the exception of the SIGCHI library, an ACM subset by design). Lastly, due to the reliance of specific keywords as indicators for both awareness and research focus, it is likely that our result sets are affected by a small number of false positives (i.e., papers that include specific keywords, but do not engage in analysis of the subject) and false negatives (i.e., papers that cover the subject area but do not use the search terminology, for example, papers covering energy efficiency and consumption without using the term ‘energy efficient’). For example, one of the 6 SIGCHI results in the ESR set of Search 2 turned out to be the full proceedings of CSCW’17, which contain all keywords spread across different papers. Nonetheless, the covered databases are among the most comprehensive sources available, and combined, the number of missed papers should be minimal. In addition, full-text search and manual verification address false positives, the terminology of rebound effects is well established, and the relative lack of attention is widespread and consistent across all searches. The trends we identified are unlikely to be affected by minor shifts in time or coverage. Therefore, these limitations do not threaten the validity of the findings.