Energy Economics 103 (2021) 105563 Contents lists available at ScienceDirect Energy Economics journal homepage: www.elsevier.com/locate/eneeco Reversing impatience: Framing mechanisms to increase the purchase of energy-saving appliances Mariateresa Silvi *, Emilio Padilla Rosa Department of Applied Economics, Universitat Autònoma de Barcelona, Spain A R T I C L E I N F O A B S T R A C T Keywords: Most environmental decisions involve intertemporal trade-offs, in that they require foregoing immediate grati­ Intertemporal choice fication for the sake of future environmental quality. One such example is investing in energy efficiency, which Energy-efficiency entails an initial upfront cost in exchange for a future stream of energy and economic savings. Our experiment Temporal preferences explores the role of individual temporal preferences in the decision to invest in energy conservation. We report Pro-environmental behavior Survey experiment results from a study on a nationally-representative sample of 2010 United States adults. Participants chose be­ Nudge tween appliances that differed solely in price and operating costs. We manipulated the salience of energy costs and primed participants with future-oriented messages. Our treatments increased energy-efficient choices by 24 percentage points compared with the status-quo scenario. Present-oriented individuals are less likely to purchase energy-efficient appliances but loss-framed messages that highlight the opportunity cost of inefficient appliances diminish the effect of impatience on refrigerators choice. 1. Introduction preferences, the tendency of individuals to be either focused more on the present or on the future, are an underexplored dimension in the litera­ The majority of individuals do not invest in energy efficiency even ture on energy conservation. By definition, present-oriented individuals when the long-run economic benefits outweigh the upfront additional discount future well-being at a higher rate; therefore, we expect them to costs of the initial investment (Schubert and Stadelmann, 2015), a be less likely to invest in energy-saving appliances, since that typically paradox known as the energy-efficiency gap. The societal relevance of involves a higher price paid today and lower electricity bills over the the missed potential energy savings is threefold: It represents long-term lifetime of the appliance. loss of available income for households,1 it contributes to global The literature on the energy-efficiency gap has identified the un­ warming,2 and it negatively affects public health.3 derlying causes of the under-adoption of energy-saving appliances: We investigate energy investments as a type of temporal dilemma limited information, cognitive biases, financial constraints, attention since they require individuals to pay a larger sum immediately in order deficits, preferences for other appliance attributes, uncertainties about to enjoy a larger stream of economic savings in the future. In particular, future energy savings, and individual time preferences (DEFRA, 2010; we examine whether individuals can be induced to act more patiently Epper et al., 2011; Newell and Siikamäki, 2015; Allcott and Taubinsky, and choose energy efficient devices through framing messages that 2015; Gerarden et al., 2015; Schubert and Stadelmann, 2015). The highlight the opportunity costs of choosing less efficient electric appli­ majority of empirical studies have focused on information and cognitive ances. We further investigate the role of individual temporal preferences deficits (Schubert and Stadelmann, 2015), but their experimental in energy investment decisions and whether their impact on choice can treatments reached mixed results and interventions which, at best, be influenced through changes in the choice context. Temporal yielded only modest efficiency increases (OECD, 2017). Policy measures * Corresponding author. E-mail addresses: mariateresa.silvi@uab.es (M. Silvi), emilio.padilla@uab.es (E. Padilla Rosa). 1 Granade et al. (2009) estimated the energy-efficiency gap in the United States economy to be worth $1.2 trillion in potential energy savings against an upfront capital cost of $520 billion. 2 Electricity and heat generation from fuel combustion are currently the largest sources of CO2 emissions from fuel combustion globally (42% of the total in 2016) according to the International Energy Agency (2021). 3 Particle pollution from burning fossil fuels for electricity generation in the United States contributes to nearly 15,000 premature deaths a year (Goodkind et al., 2019). https://doi.org/10.1016/j.eneco.2021.105563 Received 10 May 2021; Received in revised form 30 August 2021; Accepted 6 September 2021 Available online 10 September 2021 0140-9883/© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 focused on information and training typically have reached energy appliance categories, from 7% to 17% for lightbulbs to 39%–300% for savings of about 2% (Rivas et al., 2016). In this paper, we focus on in­ refrigerators (Schubert and Stadelmann, 2015). However, such discount dividual time preferences and explore whether their effect on choice can rates may reflect considerations beyond cost optimization, for example, be influenced to induce greater energy savings beyond simple infor­ risk or uncertainty aversion—i.e., to the fact that the future savings from mation provisions. electricity may actually not materialize, the inability to finance the When an inefficient appliance is purchased because the individual is purchase of the more efficient appliance, or preference for other appli­ inattentive to or incapable of computing future operating costs, dis­ ance attributes such as dimensions or appearance as shown in Fig. 1. playing this information alongside the product price can effectively help Hence, these discount rates cannot be considered as reflecting pure increase energy-efficient purchases. However, when the reason for such temporal preferences (Gerarden et al., 2015). In addition, a number of a purchase is that individuals have present-oriented time preferences, deviations from the expectations of neoclassical economics have been displaying such information per se will not produce any effect (DEFRA, found to influence energy efficiency choices, including status-quo bias, 2010). Enzler (2013), Newell and Siikamäki (2015) and Schleich et al. present bias myopia or bounded rationality (Cattaneo, 2019). Therefore, (2019) have found that present-oriented time preferences are associated a number of interesting questions remain unanswered, such as: What is with higher discounting of future energy savings. Yet, the role of pure the contribution of pure time preferences to the energy-efficiency gap? time preferences and the extent to which they can be influenced to help To what extent does myopic preferences drive these abnormally high close the energy-efficiency gap have not been fully explored. discount rates? Can we influence energy-saving product choices by In this paper, we report results from a sample consisting of 2010 manipulating label information on the salience of intertemporal costs United States (US) adults. Participants were subjected to randomized and benefits? treatments showing alternative framings of two refrigerators’ energy Research has shown that individuals largely ignore ancillary costs requirements. Our core treatments display information regarding the (such as maintenance or operating costs) when making investment de­ energy consumed by each refrigerator in either i) kWh/year; ii) esti­ cisions (Allcott and Wozny, 2014; DEFRA, 2010). Half of vehicle buyers mated electricity cost in US$ over the lifetime of the appliance; or iii) admit to not considering fuel costs in their purchase decisions (Allcott, estimated electricity cost in US$ over the lifetime of the appliance with 2011), and buyers often disregard the price of ink cartridge re­ the addition of a message warning about the comparative future eco­ placements when purchasing a printer (Hall, 1997). Individuals are also nomic loss (or gain) compared to the alternative. With the scope of unable to estimate the energy usage of their appliances and the costs testing the power of our treatments in a more realistic choice environ­ associated (Attari et al., 2010; Epper et al., 2011). Even when salient to ment, where a richer array of appliance attributes beyond purchase price the buyer, information regarding the running costs of energy-efficient and electricity consumed are listed, we repeat our core treatments by appliances can be hard to process (OECD, 2017). Most energy labels, adding refrigerator images, total and freezer capacity, and color finish such as the European Union’s energy-efficient rating system, indicate specifications to the choice cards. The purpose is to test whether our core the efficiency of the appliance relative to similarly sized counterparts treatments are equally effective even when this additional information and estimated annual kWh consumption (Rohling and Schubert, 2013). on the choice cards is competing for the participants’ attention. These Evidence from experimental studies, however, shows that consumers differences, which are marginal and such that the participant should be focus mostly on the former. This can induce buyers into the “energy- indifferent to them, are randomly assigned to either the more or the less efficiency fallacy,” the tendency to infer the amount of energy required efficient appliance. by an appliance from its energy rating (Waechter et al., 2015a; Waechter Performing a randomized experiment enables us to infer any cau­ et al., 2015b) rather than from the estimated kWh usage. For the same sality nexus between the treatment and the resulting outcomes. This is reason, consumers may end up buying appliances that are efficient but because the experimental setting allows us to control the exact choice also consume more because they are bigger—which is the “volume-ef­ context and because the sample of participants in each group can be fect” (Stadelmann and Schubert, 2018). Furthermore, the estimated assumed to be homogeneous, thanks to full randomization. In addition, annual kWh usage is not easily translatable into economic terms. This regression analysis helps us to control for other sources of individual would require the consumer to retrieve the price he or she pays for a heterogeneity which may affect the results. kWh of electricity and compute the potential cumulative savings from To date and to the best of our knowledge, no study has focused on the running an appliance over the years relative to the initial price for each temporal salience of operating costs relative to the capital costs. Nor has appliance he or she is considering buying. The non-availability of such the choice context been explicitly framed as one between two sequences: information and the cognitive effort associated reduces the likelihood one that promises long-term energy savings in exchange for a price that energy savings will be considered preeminently in the purchase premium paid in the present and one that grants immediate economic decision (Blasch et al., 2019). For these reasons, kWh per year is savings which are offset in the long term through higher energy con­ considered an opaque characteristic (DECC, 2014). sumption. This framing effect has never been tested in an environmental A few recent field experiments manipulated appliance labels by product choice context, nor has its effect been tested on individual explicitly including the economic cost of operating the appliance. The temporal preferences. We show that the number of energy-efficient effects were, however, inconclusive: this intervention led to a modest appliances chosen increases as a consequence of this framing. increase in the purchase of energy-efficient tumble dryers and washer The remainder of the paper is organized as follows: Section 2 syn­ driers, but it was ineffective or in some cases, it even decreased the thesizes the most relevant empirical studies to date; Section 3 describes purchase of energy-efficient refrigerators in different experiments the data and methods; Sections 4 and 5 show and discuss results; and (Kallbekken et al., 2013; Department of Energy and Climate Change, Section 6 concludes with key messages and policy implications. 2014; Schubert and Stadelmann, 2015). More specifically, Kallbekken et al. (2013) deployed a redesigned efficiency label showing estimated 2. Literature Review lifetime operating costs for tumble dryers and fridges-freezers. They found no statistically significant effect for fridges-freezers and a 4.9% There is an abundant body of literature, starting in the 1980s, that reduction in average energy use for tumble dryers (the energy savings has measured how individuals supposedly trade off upfront capital and were only 3.4% when the effect of staff training was excluded). The future operating costs by deriving implicit discount rates from purchase authors concluded that this type of intervention would be effective only choices. These observed discount rates are typically much higher than for appliances for which the energy cost constitutes a major portion of the interest rate individuals would be charged for borrowing capital, to the total lifetime cost. A field experiment by the UK Government’s the extent that they have been deemed “irrationally high” (Hausman, Department of Energy and the Behavioral Insights Team added a tag on 1979). Implicit discount rates obtained in this way vary widely across top of the EU energy-efficiency label displaying estimated lifetime 2 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 Fig. 1. Causes of the energy-efficiency gap. (Source: own elaboration based on the authors’ literature review) electricity costs for combined washer-dryers, washing machines and As these studies show, providing information on operating costs can, dryers in 19 electronics stores across the country. They monitored sales in some cases, lead to modest efficiency increases, particularly if oper­ in 19 other stores of the same chain that were exposed solely to the EU ating costs are high relative to the total lifetime costs. But it can be energy-efficiency label as the control group (Department of Energy and ineffective or even counter-effective if the operating costs are quite low, Climate Change, 2014). The experiment yielded a 0.7% reduction in the such as for small appliances (e.g., vacuum cleaners) or if they are energy consumption of washer-dryers sold to the treatment group and expressed in annual differences. These limited effects may be due to the no significant effect for the two other appliances. A cost–benefit analysis fact that merely showing operating costs is not enough to influence in­ from a nationwide adoption of the intervention revealed that despite the dividual temporal preferences. To this purpose, we manipulate the in­ limited impact of the treatment, the benefits from avoided CO2 emis­ formation on present and future costs by making the intertemporal sions would vastly outweigh its low implementation costs. In 2015, an dimension of choice more obvious to the prospective consumer. Our online experiment commissioned by the Swiss Federal Office of Energy experiments recreate the status quo—where electricity consumption conditioned the online purchases of freezers, vacuum cleaners, tumble information is expressed in kWh, but we also reproduce the treatments of dryers and televisions with two alternative energy labels. Buyers were the experiments mentioned above—merely translating electricity con­ either exposed to the standard EU energy label or to another displaying sumption in economic terms—in order to use these results as bench­ i) the lifetime operating costs expressed as losses or gains relative to the marks against which to evaluate our own treatments. This enables us to average of all appliances; and the ii) annual electricity cost from running compare with previous experiments but also to investigate whether the appliance, expressed as a color-coded scale comparing products of manipulating the information in a way that makes the intertemporal the same typology. Both labels led to higher purchases of energy- trade-off more explicit can increase the choice of energy-efficient ap­ efficient appliances compared with the baseline scenario with no pliances beyond merely expressing the energy consumed in monetary label. The energy-cost label was slightly more effective in reducing terms. average energy consumption for tumble dryers compared with the EU Behavioral studies aimed at depicting a realistic understanding of energy label (by an additional 1.6 percentage points), but both had no intertemporal choice have highlighted two opposing prevailing inter­ effect on the purchase of freezers. The EU label was instead twice more temporal preferences (Berns et al., 2007). On the one hand, they showed effective than the energy-cost label in reducing energy consumption individual preferences for immediate gratification. On the other hand, a (− 10.2% against − 4.5%) from the purchase of efficient vacuum cleaners lesser-known strand in the literature has shown that individuals have a (Schubert and Stadelmann, 2015). In a hypothetical experiment, preference for improving outcomes, i.e., saving the best for later (Loe­ Skourtos et al. (2021) found that displaying the annual operating costs wenstein and Prelec, 1991; Loewenstein and Sicherman, 1991; on energy labels of refrigerators did not affect choices, because the Chapman, 2000; Frederick et al., 2002). While the former strand of annual differences in operating costs were relatively low. literature depicts individuals as shortsighted and present-biased, the 3 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 latter proposes that they are patient and forward-looking. The implica­ realistic and easier to relate to. The information collected was used to tions for energy-saving investments are straightforward. The former define the choice card design and refine the textual messages. Second, approach suggests that they are unlikely to occur because individuals we tested the two core treatments on energy consumption information positively discount future energy and monetary savings, whereas the on another 224 undergraduate students, the results from this test are latter depicts a more optimistic view, suggesting that people can antic­ included as Annex 1. Third, we run the actual experiment on a nationally ipate the future benefits of energy-saving investments and make sacri­ representative sample of the US population. The experiment was fices in the present with that perspective. While neoclassical economic awarded funding and implemented through the Time-Sharing Experi­ theory postulates that individual time preferences are constant in elic­ ments for the Social Sciences (TESS), a program financed by the US iting the same choice options, regardless of the way they are framed National Science Foundation (NSF).4 The sample used for the experi­ (Tversky et al., 1988), a variety of framing effects have been found to ment consisted of 2010 adults (51.6% female, mean age = 48.3 years) affect intertemporal choice (Lewis, 2018). As Ebert and Prelec (2007) from an AmeriSpeak pre-screened pool of participants who were invited observed, the temporal dimension has an “optional status,” in that it can to respond to a survey online between June and August, 2020. be pushed into the background or become a key concern depending on aspects of the choice situation. This makes sensitivity to time extremely 3.1. Experimental Design and Additional Data Requirements susceptible to manipulations, such that simply drawing people’s atten­ tion to time can eliminate present bias (Goodman et al., 2019; Lewis, Participants chose between two refrigerators, an energy efficient 2018). The variability in reported discount rates between and within one—here in the article referred to as green—and an otherwise iden­ product categories, mentioned at the beginning of this section, would tical, less efficient alternative—gray. We defined the context as a seem to suggest that discount rates are indeed malleable and sensitive to refrigerator replacement decision and the choice as being between two framing (DEFRA, 2010). appliances that the participant had hypothetically pre-selected among In particular, our experiments are inspired by two well-documented various models. Participants were told that the two refrigerators differed framing effects, hidden-zero framing and delay/speed-up asymmetry. The solely in price and energy consumed and that they were otherwise equal. hidden-zero framing explicitly expresses that in the future, the individual The experiment followed a 2 × 3 factorial between-subjects design: one will be getting nothing or even losing from choosing immediate grati­ factor defining alternative framings of the appliances’ energy re­ fication. By unveiling the otherwise hidden opportunity costs of impa­ quirements and the other factor pertained to the number of attributes tience, the effect of this framing was found to reverse temporal shown for each refrigerator. preferences and induce more forward-looking choices in several exper­ The energy consumption factor had three levels: iments (Loewenstein and Prelec, 1993; Read et al., 2005; Magen et al., 2008; Wu and He, 2012; Scholten et al., 2016; Read et al., 2017) and to i. Control: cards showing the purchase price and the annual elec­ reduce implicit discount rates (Faralla et al., 2017). In the delay/speed-up tricity usage in kWh of each appliance. This level reproduces the asymmetry framing, individuals are either asked to delay immediate status quo. gratification or to anticipate a later reward; evidence shows that people ii. Treatment 1 (T1): cards showing the purchase price and the discount more under the first condition (Weber et al., 2007). This strand annual electricity cost for the expected lifetime of each appliance. of research suggests that evoking forward-looking thoughts in the choice This level reproduces what has been done in the earlier literature, process (such as recalling the future electricity savings of a more effi­ in both field and hypothetical experiments. This enables us to cient appliance) can lead to less impatient choices. Frederick and Loe­ compare our own treatment (T2) below, against previous wenstein (2008, p. 233) concluded that individuals possess various findings. cognitive patterns which “may be evoked or suppressed by subtle iii. Treatment 2 (T2): cards showing the purchase price, the annual contextual features” and that pairing events in a sequence may electricity cost for the expected lifetime of each appliance, and an encourage individuals to consider emotions they may have otherwise additional “patience-inducing” message on each card showing not included in the decision process. These studies suggest that simply the lifetime loss (avoided loss) in electricity cost compared with changing the construal of alternatives without changing their actual the more efficient (less efficient) appliance. The message men­ value has an impact on our ability to make optimal intertemporal de­ tions the date by which such loss (gain) would be realized, cisions (Magen et al., 2008). potentially engaging participants into anticipating their future To our knowledge, only three studies analyzed the relationship be­ emotions in that regard. tween pure temporal preferences and the energy-efficiency gap. In two hypothetical online experiments, Enzler (2013) and Schleich et al. The second factor had two levels: (2019) found that individuals who were more present-oriented were less likely to choose energy-efficient options. In a similar setting, Newell and i. High focus on electricity consumption: cards showed only price and Siikamäki (2015) found that the US energy label that shows the esti­ electricity consumption for each appliance, which reflects the stan­ mated operating costs of running different electrical appliances is less dard practice of hypothetical choice experiments in this field. effective on individuals who discount more future outcomes. These ii. Low focus on electricity consumption: cards showed also additional authors show that present-focused time preferences are associated with appliance features such as total capacity, freezer capacity, color, and higher discounting of future energy savings, and that leads to underin­ an image. These additional features varied marginally such that vestment in energy efficiency. If time preference has a role in deter­ participants would be expected to be indifferent to them, and they mining the adoption rate of energy-efficient appliances, and if it is not an were randomly assigned to either the energy-efficient or inefficient innate and immutable individual characteristic but can instead be option. The purpose of adding these features was to mimic a more influenced (Chapman, 1998), then it should be possible to increase realistic choice environment where several appliance features are energy-efficient investments through framing manipulations. presented to the prospective buyers; and to test whether the effec­ tiveness of our treatments would diminish when more information 3. Methods We developed our hypothetical experiment in three stages. We first 4 TESS contracts NORC at the University of Chicago, a research institution conducted a pilot of the experiment on undergraduate students at a that uses AmeriSpeak, a nationally representative probability-based panel Spanish University to understand how they approached and processed composed of US adult residents. TESS-funded experiments in the economics large purchases decisions and how to make the hypothetical choice more field include Taubinsky and Allcott (2013) and Davis and Metcalf (2016). 4 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 competed for the user’s attention, which is similar to what was done lower than that of the efficient xb, but its yearly operating costs ya are in Andor et al. (2020). higher than those of the efficient appliance yb. Hence, over the lifetime of the appliance, the operating cost savings from running the efficient Participants were randomly assigned to one of the resulting six appliance—discounted by the rate r, the opportunity cost of capital­ groups. Table 1 summarizes the six conditions and illustrates the actual —outweigh the initial difference in their capital costs. choice cards participants were presented with. ∑ T After the experimental session, all participants responded to the xa < xb ya > yb →δt > 1→ (ya − yb )δt > (xb − xa ) (1) same post-experiment questionnaire, which included three sections: t=0 i) a text asking for the percentage of electrical appliances out of the where δt = 1/(1 + r)t is a discount factor representing the value of one total they had bought in the previous three years that had the unit of currency, delayed by one year, given the rate faced by the con­ Energy Star certification5; sumer for borrowing and lending money. To ensure that each refriger­ ii) one question where they had to rate on a Likert scale of 0–5 how ator option is realistic, we determined the range of operating costs and representative five statements were of them. These items were the range of purchase prices in a way that matches the actual range of taken from the Consideration of Future Consequences Scale appliances currently available in the US market. To determine the ap­ (Strathman et al., 1994), a scale which was used to measure in­ pliances’ purchase prices, we selected a mid-sized refrigerator typical of dividual temporal preferences and that was adopted in previous the reference market6 and analyzed the appliances available within this similar studies (Enzler, 2013); and capacity range in the online catalogues of the three chief appliance re­ iii) a matrix where they had to rate how much they agreed on a Likert tailers at the time of the survey (Sears, Best Buy, and Lowe’s). This scale of 0–5 with a statement declaring that individuals can play helped determine an initial range of prices. Within these ranges of an important role in protecting the environment. This question is prices: the same used in the Eurobarometer survey and it is normally interpreted as a measure of ascription of personal responsibility i. The price of the less efficient appliance was fixed at the maximum to take care for the environment. within the 1st quartile of prices observed in the market for that ca­ pacity range. That price was $1099.99. Several variables with socio-demographic data about participants, as ii. The price of the more efficient appliance was calculated as 25% more well as their political views, saving habits, living conditions, and cal­ expensive relative to the price of the other product. That price was culus abilities that had been gathered by the National Opinion Research $1373.99. Center (NORC) in previous surveys were made available to the re­ searchers. The full survey and additional screenshots of the experiment This pricing methodology is similar to a study commissioned by the are included as Annex 2. European Commission (IPSOS, 2014). The two prices obtained with this methodology appear to be good estimates as they fall slightly below and 3.2. Appliance Choice slightly above the national averages, respectively. More specifically, according to the US Bureau of Labor Statistics, the yearly expenditure on We chose refrigerators, among other electric appliances, for the refrigerators per household was on average US$83 in 2018 (US Bureau following four reasons. First of all, refrigerators are among the most of Labor Statistics, 2020). Multiplying this sum by the average lifespan energy-consuming appliances in the household; therefore, operating of non-commercial refrigerators, leads to an average appliance price that costs over the lifespan of the appliance are quite significant. Second, is between the two prices estimated. refrigerators are running all the time, and therefore we can reliably The range of estimated energy consumed in kWh corresponds to the estimate the energy they consume independent of usage patterns, unlike range available on the US Energy Guide label for all similar models air conditioners or dryers, where household size or the local climate available in the market. To calculate yearly operating costs, we multiplied would hamper these estimations. Third, refrigerators are the most the extremes of this range for the most up-to-date national average price common energy-intensive appliances. Hence, this hypothetical experi­ of electricity per kWh in the country at the time of the survey, taken from ment is directly relatable to a high number of participants. Fourth, the US Energy Information administration website (US$0.13/kWh, previous research shows that individuals discount the future cost of March, 2020 figure) (US EIA, 2020). Lifetime operating costs in US$ were operating refrigerators at a much higher rate than any other appliance calculated by multiplying the yearly operating costs by the average (between 39 and 300%) (Train, 1985; DEFRA, 2010; Epper et al., 2011), years of duration. Lifetime operating costs were not discounted, because thus indicating a high potential for behavioral intervention. the intent of the exercise was to let individuals apply their own discount rate in full. We deemed imposing a discount rate and explaining the concept of discounting to laypersons to unnecessarily complicate and 3.3. Appliance Characteristics distort responses as results from focus groups in Kallbekken et al. (2013) also demonstrate. This choice is consistent with the literature (Kall­ In the literature, an energy-efficiency gap exists if there is an energy- bekken et al., 2013; DECC, 2014; Stadelmann and Schubert, 2018). After efficient product that is cheaper in terms of total lifetime costs (adding reviewing different estimates of the average years of duration for resi­ capital and lifetime energy-running costs) than other less efficient dential refrigerators in the US, we chose the estimate from the latest equivalents, yet it is not purchased (OECD, 2017). Consistent with the analysis available on the topic, undertaken by the E.O. Lawrence labo­ literature (Newell and Siikamäki, 2015), to represent the trade-off be­ ratory at the Berkley University (Lutz et al., 2011). Differently from tween a higher purchasing price and lower energy consumption, one earlier estimates which rely on informal manufacturers’ experiences (US appliance has a lower purchase price but a higher energy cost, such that AIS, 2000), Lutz et al. (2011) combined residential survey data with the difference in energy costs is higher than the difference in purchasing manufacturer data on historical shipments. The authors conclude that costs. More specifically, the capital cost of the inefficient appliance xa is the average lifespan of a refrigerator in the US is 19.7 years, which we rounded up to 20 years for simplicity in the experiment. 5 Energy Star is a voluntary certification system in use in the United States. Electrical appliances need to have passed a series of efficiency tests established 6 by the Environmental Protection Agency in order to bear the Energy Star yellow We determined that side-by-side, stainless steel models with ice makers sticker. within 24–25.1 cubic feet would be a good average model of reference. 5 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 Table 1 Matrix summarizing the 6 experimental conditions. 3.4. Hypothetical Bias the behavioral mechanisms defining individual choices. It is possible that choices made in a hypothetical environment may 3.5. Outcome Variables and Analysis differ from choices in the real world. For instance, individuals may be unable to accurately represent their preferences if given a hypothetical Differences between the choices of the treatment groups revealed the scenario they had never experienced. Alternatively, they may not effectiveness of performing the two treatment manipulations against the engage enough with experimental choices when they are not going to baseline scenario in the promotion of energy-saving investments. The bear any consequences in real life. Experimental economics typically main outcome variable is the percentage of individuals choosing to addresses this limitation by adding real incentives to experiments, purchase the energy-efficient (green) appliance over the less efficient thereby linking real consequences to the hypothetical environment. (gray) one. Based on these percentages, we are able to calculate the Unfortunately, compensating participants with economic sums propor­ estimated energy savings generated from each treatment, had such ap­ tional to the amounts mentioned in our experiment was financially pliances been purchased in the real world. unfeasible with such a large number of participants. To establish the determinants of energy-efficient choices, we esti­ However, there is no reason to believe that the hypothetical bias mate logit regressions of the form: would have affected a specific treatment group more strongly. Rather, we think that if there was any hypothetical bias, it should have affected Y(i) = β∙Treatment1(i) + γ∙Treatment2(i) + δ∙FOS(i) + θ∙Treatment1(i) ∙FOS(i) all groups equally. In this sense, we expect the relative differences in the + ϑ∙Treatment2(i) ∙FOS(i) + μAttr(i) + a(i) + ε(i) (2) outcome variables of the different groups to be potentially comparable with what they would be in a real world context. Likewise, studies where Y(i) is our dependent variable. It is a 0–1 dummy variable set to 1 comparing hypothetical choice settings with revealed preference ap­ participant i chose the green appliance, and 0 otherwise. proaches have shown that treatment effects tend to be of the same size FOS(i) stands for Future Orientation Scale. It is a 5–25 scale (Carson et al., 1996; Ebeling and Lotz, 2015) or significantly correlated measuring individuals’ temporal orientation; the higher the value, the (Attanasi et al., 2018). Finally, to check for hypothetical bias, we more future-oriented the individual is. Additionally, we look for inter­ analyze answers to the question regarding how many energy-efficient action effects between the treatments’ variables and FOS(i).7 electrical appliances participants had purchased in the last five years. μAttr(i) is a 0–1 dummy variable set to 1 if the participant could see Comparing these answers with experimental choices allows us to detect additional appliance attributes in the choice cards, such as image and potential hypothetical biases. An additional advantage of survey ex­ total capacity, and 0 otherwise. periments, compared with a field experiment, is that they allow us to a(i) is a vector consisting of individual-level and state-level socio- tightly control the decision environment, and they provide a vast array of information on our participants. This enables us to identify and disentangle various elements that concur with the end result and unveils 7 FOS was constructed by summing answers on a 1–5 scale to 5 questions measuring individual future orientation. 6 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 demographic controls as well as a control for individual environmental and the field experiments mentioned in the literature section, we orientation. These controls increase the precision of our estimates and compared the decrease in kWh consumption they achieved with our correct for the slight socio-demographic imbalances observed between T1—which reproduces the treatments provided by these experiments in groups. The controls included are: a hypothetical scenario—. The decrease in kWh achieved in our T1 totals 5% (when additional appliance attributes are included) or 4.5% (when i. Demographic controls: sex, age, race, education, employment choice-cards only mention price and electricity cost). In the field ex­ status, marital status, State, religion, church attendance, ideol­ periments testing the same treatment as in T1, the decrease in kWh ogy, party ideology, and financial literacy. obtained ranged between 0 and 4.9%, depending on the experiment and ii. Household characteristics controls: home type, household size, the appliances being considered. So, while we cannot reliably predict telephone service type, metropolitan area dummy, and internet the size of the effect that T2 would have in a field or real world scenario, availability. we can nonetheless observe that the effect of our T1 falls withing the iii. Economic indicators: household income scale, three dummy range observed in field experiments. As such, we could expect that T2 variables capturing whether the person expects their economy to could produce effects that are commensurate in size to our experimental worsen in a year, whether the person saves systematically a results. portion of income, and whether the person pays rent for his or her The results of our econometric estimations are reported in Table 3. In home. column 4, we find that both T1, i.e., merely expressing energy cost in iv. Environmental orientation: whether the person feels he or she total € over 20 years of use, and T2, i.e., adding a message highlighting can play a role in caring for the environment, on a 1–5 scale. the relative future losses compared with the other appliance, are both statistically significant at the 99% confidence level (controlling for de­ 4. Results mographic and household characteristics). The coefficients show that being in either group increases the likelihood of choosing the green In this section we include results organized by topic, combining re­ appliance, compared with the control group. These results are robust sults from our econometric analysis with qualitative analysis. regardless of whether socio-demographic controls and household char­ While random assignment of participants to groups presumably leads acteristic controls are included (columns 2–4). In column 8, we compute to homogenous groups, we calculated descriptive variables of the sam­ marginal effects based on the equation in column 7, and we find that ple by group to ensure that these groups were balanced (Table 2). being in T1 increases the probability of choosing the green appliance by Groups were homogenous, as the p-values of the ANOVA test of equal 16.6%, while being in T2, increases the likelihood of choosing the green variances (for the continuous variables) and the p-value of the Kruskal- appliance by 19.7%. Both estimates are statistically significant at the Wallis test (for the ordinal variables) show. 99% confidence level. Pseudo-R2 reaches 10%. This level of R2 is similar to that of other papers in the field and is considered an acceptable level, 4.1. Effect of Treatment on Appliance Choice given the complexity of human behavior (Langbein, 2015). Based on the qualitative and quantitative evidence presented, we are able to conclude We start with a graphical representation of our key results. The that T2 is statistically more effective than T1 in increasing energy- variable of interest across all groups is the percentage of participants efficient choices. picking the green appliance. As shown in Fig. 2, the proportion of par­ Participants who visualized additional information on the refriger­ ticipants choosing the green appliance increases from 57% (in the con­ ators—beyond price and electricity requirements—were less likely to trol group) to 72% in T1 and to 75% in T2 when the survey displays only choose the efficient appliance. We thus interpret this result as the the appliance prices and electricity requirements. consequence of a distraction from electricity consumption information. The same upward trend from Control to T2 is repeated when addi­ The additional refrigerator characteristics (image, total refrigerator ca­ tional appliance features are shown to the participants (from 47% to pacity, and color) varied minimally between appliances and were 71%). However, the percentage of green choices is always lower randomly associated with either the green or gray appliance. We thus compared with the case in which these additional features are not expected that participants would be indifferent to these differences. To shown. Introducing more appliance features clearly reduces the partic­ check for this hypothesis, in column 5, we include dummy variables ipants’ focus on energy efficiency, leading to a greater proportion of capturing whether individuals chose refrigerators with Image A, stain­ energy-inefficient choices. Even in this context, which mimics more less steel color, a marginally smaller total capacity, and smaller refrig­ closely a real-life choice environment, T1 and T2 appear to be particu­ erator capacity. We find that only the coefficient for smaller freezer size larly effective, with a 17- and 24-percentage point increase in green is statistically significant and has a negative sign, signaling that a pref­ refrigerator choices, respectively, compared with the control group. A erence for a slightly larger freezer explains a small part of the above Kruskal-Wallis H test was conducted to determine if the likelihood of result. Yet, we notice that adding these controls to the regression does choosing the green appliance was different for the three conditions. The not alter the findings previously discussed; the other coefficients main­ test showed that there is a significant association between treatment and tain their sign and size. Electricity prices in the US, vary widely across appliance choice. There is a statistically significant difference in the States. At the time of the survey they ranged between US$0.09/kWh in likelihood of choosing the green appliance between the three groups (χ2 Oklahoma to US$0.32/kWh in Hawaii. We thus added State level kWh (5) = 57.756, p = 0.0001). electricity price as an independent variable in our model but did not find We calculated the total kWh/year that would be consumed under the a statistically significant effect, and we thus did not report this regres­ Control and T2 scenarios (with additional appliance features) by taking sion. It is to be expected, however, that the effect of T1 and T2 would be into account the percentage of individuals choosing green appliances in even more pronounced in countries with higher electricity prices, should those conditions, and we found that total kWh consumption decreases operating costs be calculated using local prices as opposed to the na­ by 7.15%. Considering that 12.4 million refrigerators were sold in the tional average, an expectation which is line with findings from Davis and US in 2019 (AHAM, 2020), we calculated the potential energy savings Metcalf (2016). that could be generated on a national scale, assuming that choice was restricted to these two models. We estimated energy savings worth 4.2. Effect of Temporal Orientation on Appliance Choice 642.6 million kWh/per year for the refrigerators sold in a given year. Hypothetical experiments are believed to be better suited to provide The temporal preferences survey allowed us to construct a contin­ qualitative rather than quantitative insights (Epper et al., 2011). In uous scale FOS, with higher values indicating that participants are more order to establish a comparison between our hypothetical experiment future-oriented. Fig. 3 below shows the proportion of energy efficient 7 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 Table 2 Main descriptive statistics of the sample by group. Variables Control. with Control T1. with additional T1 T2. with additional T2 p-value form additional features features features F test (1) (2) (3) (4) (5) (6) (7) Median age 46.95 (0.905) 48.74 47.91 (0.923) 47.60 49.76 (0.949) 49.05 0.294 (0.923) (0.951) (0.939) Female (%) 51% 53% 54% 49% 50% 53% 0.8572 Income (1 v. low–18 v.high) 9.75 (0.229) 9.75 9.71 (0.230) 10.26 10.10 (0.248) 10.16 0.322 (0.214) (0.211) (0.220) Education 9.57 (0.150) 9.55 9.42 (0.159) 9.62 9.61 (0.173) 9.45 0.934 (0.153) (0.166) (0.169) Future Orientation Score (FOS) 17.87 (0.198) 17.78 17.77 (0.205) 18.15 18.29 (0.183) 18.24 0.192 (0.198) (0.196) (0.179) Individual Responsible for the 4.29 (0.278) 4.35 4.37 (0.292) 4.40 4.32 (0.296) 4.10 0.973 environment (1 agree–5 disagree) (0.271) (0.291) (0.052) In the next years my income will worsen 56% 56% 57% 58% 53% 58% 0.888 (%) Energy Star purchases (%) 39.63% (2.53) 35.86% 38.75% (2.52) 40.32% 34.68% (2.47) 38.11% 0.575 (2.42) (2.56) (2.49) N. participants 344 352 327 328 323 336 Additionally, we conducted a one-way ANOVA analysis to stablish whether there is a statistically significant difference in the FOS score between individuals who choose the green appliance and the gray 85% appliance in each treatment group. Table 4 shows average values (for treatments excluding additional appliance features) and standard errors 75% in parentheses. In all groups, individuals choosing the green appliance 65% have higher FOS scores on average. However, the FOS score differentials are significantly narrower in T2, 46% and 42% less compared with the 55% differences in the Control and T1 groups, respectively. These differences are statistically significant at the 0.01 and 0.1 significance level, as the p- 45% values from F-tests in the last column indicate. This is also consistent with results from the Spanish pilot. We consider the narrow gap in FOS 35% scores for T2 as an indication that reminding individuals about the future implications of the available choices with a priming message 25% manages to partly control for pre-existing individual temporal prefer­ ences, possibly by convincing some impatient individuals to choose the Control T1 T2 With addional appliance features Without addional appliance features efficient, more expensive refrigerator. This effect, however, is not confirmed when additional appliance Fig. 2. Proportion of energy-efficient choices per condition. attributes are displayed, as Fig. 5 below graphically illustrates. This (Source: own computations based on own experimental data on 2010 US adults. suggests that the presence of a wider array of appliance features may Error bars show confidence intervals at 95% level.) water down the effects of T1 and T2 on temporal orientation. We also estimated interaction terms between FOS and T1 and T2. In both cases, choices made by individuals clustered by their FOS quartiles. Data show however, the coefficients were not statistically significant (Table 3, that higher FOS scores (more patient individuals) are associated with a columns 10 and 11). greater proportion of energy efficient choices. This pattern is, however, more obvious for individuals in the control group and in T1. From Fig. 3, we can also conclude that T2 is the most effective treatment for the most 4.3. Effect of Treatment on Other Individual Characteristics present-biased and the most forward-looking individuals alike (Q1–Q3). However, for individuals in the top quartile, T1 is slightly more effective In column 6 of Table 3, we include the four economic controls and than T2 when the additional appliance features are not present. the environmental-orientation control previously mentioned, plus two Econometric analysis in Table 3 further confirms that more future- dummies equal to 1 if the participant identifies to some extent with the oriented individuals are more likely to choose the efficient appliance Republican Party and if he or she could solve a basic percentage cal­ for every additional point in the FOS scale, the coefficients are statisti­ culus. Of these additional variables, only identifying with the Repub­ cally significant with positive sign at 0.01 significance level. lican Party has a statistically significant coefficient, with negative sign. Wondering whether the effect of individual temporal preferences This might be due to the well-documented climate-change skepticism of depended on condition, we calculated the marginal effects of being conservatives and a possible politicized perception of energy savings as a included in T1 or T2 for each additional unit increase in the FOS score Democratic Party trait. This interpretation is consistent with Gromet (Fig. 4, calculated in the equation in column 7, Table 3). The effect of et al. (2013) who found that politically conservative individuals are less being included in T1 and T2 positively contributes to the likelihood of in favor of energy-efficient investments. choosing the efficient appliance (compared to the Control group). T2 is We estimated a minimal response time for understanding the whole more effective than T1 for each FOS score level. The effect of both survey as requiring about 4 min. We thus repeat the regression in col­ treatments is stronger on individuals with lower FOS scores and it di­ umn 6 to exclude participants who completed the survey in less than 4 minishes as FOS increases. This finding matches the intuition that in­ and 9 min, respectively (columns 7 and 9). If we restrict the analysis to dividuals who are already future-focused do not need as many reminders answers that took more than 4 min, the coefficient for environmental of the future consequences of choosing immediate gratification. orientation is statistically significant at the 0.05 significance level and has a positive sign. The R2 also increases, and the model contributes to 8 M. Silvi and E. Padilla Rosa Table 3 Regression output. Variables Appliance choice (Logit) (1) (2) (3) (4) (5) (6) (7) dur > 4 (8) Marg.eff. (9)dur > 9 (10) (11) min. min. Treatment 1: operating costs info 0.647*** 0.649*** 0.671*** 0.701*** 0.710*** 0.713*** 0.835*** 0.166*** 0.192 (0.365) 0.106 (0.694) 0.702*** (0.148) (0.147) (0.151) (0.151) (0.150) (0.150) (0.177) (0.032) (0.151) Treatment 2: operating costs 0.877*** 0.882*** 0.877*** 0.904*** 0.885*** 0.926*** 0.997*** 0.197*** 0.729*** 0.906*** 1.213 (0.771) info+patience-inducing message (0.151) (0.152) (0.155) (0.155) (0.155) (0.155) (0.180) (0.032) (0.047) (0.154) FOS 0.053*** 0.051*** 0.059*** 0.057*** 0.058*** 0.056*** 0.053*** 0.011*** 0.146*** 0.046** 0.062** (0.018) (0.018) (0.019) (0.019) (0.019) (0.019) (0.021) (0.004) (0.047) (0.023) (0.022) FOS*T1 0.033 (0.038) FOS*T2 − 0.017 (0.041) Additional appliance attributes − 0.455*** − 0.467*** − 0.477*** − 0.478*** − 0.313*** − 0.066** − 0.821*** − 0.475*** − 0.475*** (0.124) (0.128) (0.128) (0.128) (0.145) (0.030) (0.335) (0.128) (0.128) Image A dummy 0.005 (0.167) Stainless steel dummy − 0.250a (0.159) Smaller freezer − 0.329** (0.162) Smaller total capacity 0.181 (0.167) Pessimist beliefs about future 0.090 (0.132) 0.101 (0.152) 0.021 (0.032) 0.436 (0.312) income (dummy) 9 NoSave (dummy) − 0.259* − 0.198 − 0.043 − 0.207 (0.160) (0.184) (0.040) (0.364) Income 0.007 (0.020) 0.004 (0.023) 0.001 (0.004) 0.003 (0.045) Rent (dummy) 0.002 (0.204) 0.008 (0.211) 0.001 (0.044) − 0.282 (0.438) Republican Party (dummy) − 0.368** − 0.493*** − 0.107*** − 0.135*** (0.175) (0.200) (0.044) (0.444) Environmental Values − 0.004 0.177** 0.037** 0.059** (0.009) (0.075) (0.015) (0.024) Financial literacy (dummy) − 0.193 − 0.226 − 0.048 − 0.390 (0.181) (0.204) (0.204) (0.395) Demographic controls N N Y Y Y Y Y Y Y Y Households characteristics controls N N N Y Y Y Y Y Y Y Observations 2010 2010 2006 2006 2006 2006 1576 476 2006 2006 Pseudo R-squared 0.032 0.041 0.094 0.100 0.102 0.105 0.140 0.290 0.102 0.099 p-value F test 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Standard errors in parentheses. Energy Economics 103 (2021) 105563 Demographic controls include sex, age, race, education, employment status, marital status, State, religion, church attendance, ideology). Households characteristics controls include home type, household size, telephone service type, metropolitan area dummy, and has internet dummy. a p-value F test nearly statistically significant at 0.111. * Significant at <0.10. ** Significant at <0.05. *** Significant at <0.01. M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 70% 65% 60% 55% 50% 45% 40% 35% 30% 1st quart. 2nd 3rd 4th quart. 1st quart. 2nd 3rd 4th quart. 1st quart. 2nd 3rd 4th quart. FOS FOS FOS FOS FOS FOS Control T1 T2 Fig. 3. Proportion of energy-efficient choices by FOS quartile, per condition. (Source: own computations based on experimental data) explaining 14% or 29% of the observed variations in the response var­ iable in columns 7 and 9, respectively. We further investigate whether being in a particular treatment group affects the relevance of other variables. This can help explain the un­ derlying mechanisms that lead one treatment to be more successful in promoting energy efficiency, compared with others. We thus rerun the regression in (6), splitting the sample by treatment condition. We first run the regressions split by the three conditions, Control, T1, and T2, without the additional appliance attributes (Table 5, columns 12–14). We start by noting that R2 rises from 10.5% (6) up to 40% with split regressions (14). This indicates that there are significant changes in the variables affecting choice in each treatment group. Indeed, FOS is the only variable that is statistically significant, with positive signs across all conditions while a series of coefficients changes sign or loses/gains statistical significance. The relevance of the economic variables varies widely in the three treatments. Having negative expectations about future income decreases Fig. 4. Marginal effect of being included in T1 and T2 on the likelihood of the chances of choosing green in the control group (statistically signif­ choosing the energy-efficient appliance (compared to the Control group), by icant coefficient with a negative sign); but it is not statistically signifi­ FOS score. cant in T1, and it increases the likelihood of green choices in T2 (Source: own computations based on experimental data. Predictive margins (statistically significant coefficient with a positive sign). We observe the with 95% confidence levels.) same pattern for individual who live in a rented home (Rent). Likewise, people who do not save systematically are less likely to buy the green appliance in T1 but are more likely to buy it in T2. Higher income levels Table 4 are associated with more green choices in T1 and T2 (statistically sig­ Average Future Orientation Scores FOS per condition and choice (without additional features). nificant coefficient with a positive sign) but not in the control group. It thus appears that the loss-framed message in T2 nudges individuals who Participants Participants Difference Average F test choosing choosing FOS per and its are worried about their income or who are usually unable to save to­ gray green condition p- wards the concrete possibility of saving in the future through an energy- value efficient appliance. Some of these coefficients are, however, significant Control 16.98 18.37 1.39 17.78 11.71 only at the 0.1 significance level confidence level, probably due to the (0.331) (0.248) (0.203) 0.000 smaller sample size for these split regressions. Treatment 17.23 18.52 1.29 18.16 8.95 Identification with the Republican Party negatively affects green 1 (0.338) (0.235) (0.196) 0.003 choices in the control group and in T1 only. It is possible that the loss- Treatment 17.67 18.43 0.75 18.24 3.33 2 (0.380) (0.202) (0.179) 0.068 framed message in T2 might dilute the perception of energy efficiency Totals 17.23 18.44 1.21 18.05 25.62 as a political matter by reframing it more strongly as a purely economic (0.205) (0.131) (0.112) 0.000 matter. Ascription of personal responsibility to care for the environment is statistically significant only in T2. Perhaps, mentioning a date in the future in T2, induced individuals who held pro-environmental attitudes 10 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 Control T1 T2 Control T1 T2 Without addional appliance features With addional appliance features Fig. 5. Difference in average FOS between gray/green choices, by condition. (Source: own computations based on experimental data) Table 5 Split regression output. Appliance choice (Logit) Variables (12) (13) (14) (15) (16) (17) Contr, no attr. T1, no attr. T2, no attr. Control T1 T2 FOS 0.183*** (0.052) 0.128*** (0.057) 0.195** (0.101) 0.055* (0.034) 0.101*** (0.035) 0.137*** (0.042) Additional appliance attributes − 0.545*** − 0.680*** − 0.344 (0.263) (0.216) (0.247) Pessimist beliefs about future income − 0.649* (0.391) − 0.425 (0.489) 1.048* (0.562) (dummy) NoSave (dummy) − 0.096 (0.475) − 1.312* (0.526) 1.172* (0.648) Income − 0.078 (0.067) 0.133* (0.077) 0.148* (0.086) Rent (dummy) − 0.850* (0.459) 0.036 (0.622) 2.313*** (0.771) Republican party (dummy) − 1.326*** − 1.616*** − 0.324 (0.691) (0.535) (0.622) Environmental values 0.050 (0.123) 0.295 (0.264) 0.650** (0.288) Financial literacy (dummy) − 0.128 (0.520) 0.904 (0.622) 1.044 (0.817) Demographic controls Y Y Y Y Y Y Households characteristics controls N N N Y Y Y Observations 329 286 272 679 634 618 Pseudo R-squared 0.312 0.338 0.400 0.165 0.214 0.180 p-value F test 0.000 0.000 0.000 0.000 0.009 Missing Standard errors in parentheses. Demographic controls include sex, age, race, education, employment status, marital status, State, religion, church attendance, ideology). Households characteristics controls include home type, household size, telephone service type, metropolitan area dummy, and has internet dummy. * Significant at <0.10. ** Significant at <0.05. *** Significant at <0.01. to anticipate also the non-financial consequences of choosing the gray participants choosing the green appliance8 slightly increases with the appliance. We hypothesize that this may have contributed to activate the percentage of Energy Star appliances they had previously bought. For effect of pro-environmental orientation in the decision process. When we run regressions by Control, T1, and T2 conditions without dis­ tinguishing whether the participants visualized additional attributes 8 (columns 15–17), we find that in T2, the coefficient for Additional At­ This graph includes only choices made in the Control group and in T1, since they represent more closely the status quo in US retailer shops, and they are tributes is not statistically significant. The result suggests that the loss- thus the experimental conditions that are most comparable to the conditions US oriented message in T2 helps to contrast the cognitive noise intro­ residents encounter when buying appliances. While electricity requirement duced by the additional appliance features. information in shops is usually presented in kWh (mirroring the control group), As an external validity test, we checked whether hypothetical all electrical appliances also come with a mandatory Energy Label that shows choices made in the experiment were associated with participants’ real- the estimated cost of operating the appliance for a year in US$ (somewhat life choices by asking them to report the percentage of Energy Star-rated similar to T1). Participants who did not buy any appliance in the previous three appliances they had bought out of their total relevant purchases in the years, who replied with an absolute number rather than a percentage, or who previous three years. Fig. 6 below shows that the proportion of skipped the question were excluded from the count, thus reducing the count to just 606 observations. 11 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 While T1 reproduced the treatment from previous field experiments 70.0% in the literature, T2 goes beyond the simple provision of energy- efficiency information, as it nudges individuals by making the hidden 65.0% temporal component of the choice context a key concern in the decision process. T2 reaches the highest proportion of individuals choosing 60.0% efficient appliances. The incremental effect compared with T1 is 7 per­ centage points. This results were in line with results from the Spanish 55.0% pilot (in Annex 1), where T2 outperformed T1 by 6 percentage points. The fact that T2 outperforms T1 in both experiments, despite the use of 50.0% two very different samples, suggests that nudging temporal preferences rather than merely informing about energy-cost implications works 45.0% better for increasing energy-efficient choices. It also signals that there are concrete opportunities to increase energy-efficient purchases with 25% En.Star 50% En.Star 75% En.Star 100% En.Star simple and cost-effective framing interventions. Our treatment simul­ Without addional appliance features (C+T1) taneously lightens the cognitive load faced by perspective buyers—by With addional appliance features (C+T1) computing the relative long-term convenience of the efficient appli­ ance—; it leverages loss aversion, it promotes identification with one’s Fig. 6. Proportion of energy-efficient choices and Energy Star purchases, by future self—by mentioning a date several years into the future—. While condition. the experimental design does not allow us to disentangle the single (Source: own computations based on experimental data) contribution of each of these components to the increase in energy- efficient choices, we do observe that our treatment affects the extent example, 64% of the people who purchased less than 25% of Energy Star to which present-focused preferences are activated in the decision products chose the green appliance in the experiment compared with process. 66% of those who bought only Energy Star appliances (for treatments We separately assessed individual temporal preferences using excluding additional appliance features). Results from a Kruskal-Wallis research-validated scales that employ a series of qualitative questions to H test, however, showed that the association between how many En­ construct an index identifying individuals on a spectrum between ergy Star rated appliances had been purchased in the past and appliance present-oriented and forward-looking. We used elements of the Zim­ choice was not statistically significant. Nonetheless, we cannot rule out bardo Future Orientation Score for the Spanish sample and the that the hypothetical decisions that subjects made in our hypothetical Consideration of Future Consequences Scale for the US sample. Quali­ scenario are related to some extent to their real-life behavior in similar tative and econometric analysis showed that in both samples, individual decision domains. temporal preferences influence energy-efficiency choices. Present- oriented individuals were less likely to choose efficient appliances, 5. Discussion whereas people who chose energy-efficient appliances had higher FOS scores. However, this difference did not equally apply to all treatment These experiments analyzed the role of pure time preferences in the groups. The difference in FOS between those who chose the green and decision to invest in energy efficiency. They tested whether individuals those who chose the gray appliance was nearly half in T2, compared could be nudged to be more patient and make more energy-efficient with the control group. This finding was consistent in both the Spanish choices, which typically deliver higher environmental and economic and the US sample. In addition, in the Spanish sample, econometric benefits in the long run. outputs showed that temporal orientation affected refrigerator choice A nationally representative sample of 2010 US adults faced a hypo­ only in the Control group, although this result was not replicated in the thetical purchase choice consisting of two refrigerators that differed US sample. We interpret these findings as an indication that the treat­ solely in capital price and electricity consumed. The more efficient ments, T2 and partly T1, reduced the influence of pre-existing temporal refrigerator was more expensive to purchase but cheaper to operate preferences on energy-efficient choices. T2 was the most effective compared with the inefficient alternative. However, the overall total treatment for both the bottom and the top FOS quartiles of the US lifetime cost was cheaper for the efficient appliance. The treatments sample when several appliance attributes were shown. These results consisted of alternative framings of the appliances’ energy re­ would seem to suggest that there is relationship between temporal quirements, showing the information expressed in either i) kWh/year orientation and energy-efficiency decisions. Pre-existing individual (Control); ii) total electricity cost in US$/€ for the product’s lifetime temporal preferences affect the likelihood that the individual will buy (T1); or iii) adding to lifetime cost a loss-framed message about the efficient appliances. However, this paper reinforces the idea in the relative loss (or savings) of each appliance at a given date in the future literature that individual temporal preferences can be activated or (T2). pushed to the background by the choice architecture in place. The literature on hidden-zero framing and delay/speed-up asymmetry Hypothetical experiments in energy efficiency typically elicit pref­ states that individuals’ temporal preferences can be influenced by erences by showing only price and electricity requirements. This may contextual features that highlight the hidden opportunity costs in the raise the question of whether individuals can focus on energy efficiency choice situation or by evoking forward-looking thoughts. These features as much in a real-choice environment where appliances are described can potentially inspire alternative cognitive patterns that can improve with a wide array of features (appliances are described with a range of individuals’ ability to optimize their intertemporal decision making. Our 40–70 features on US retailer websites and 10–30 on Spanish retailer core treatment T2 leverages these findings by highlighting the long-term websites). To increase the resemblance of our experiment to a real- economic losses with a message. choice setting, in the US experiment, we repeated the experiment by Previous experiments that displayed only electrical appliances’ adding additional appliance features, such as color, total capacity, yearly or lifetime operating costs found either limited increases or even freezer capacity, and images. These features were randomly assigned decreases in energy-efficient purchases. This might be due to the fact and varied marginally between appliances, such that the participant that energy-inefficient choices are in part driven by temporal prefer­ should have been indifferent to them. However, we found that in­ ences rather than by an information deficit. For example, Newell and dividuals in these latter treatments were less likely to choose the energy- Siikamäki (2015) found that displaying yearly operating costs has less efficient appliance. Nonetheless, we noted that T2 was still the treatment effect on present-biased individuals. with a higher proportion of energy-efficient choices. Displaying 12 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 additional appliance features diminishes the individual’s attention to socio-cultural characteristics, different electricity prices and appliance energy efficiency, but T2 succeeded in giving visibility to the conse­ markets. quences of choosing inefficient appliances. This finding is particularly relevant for policy: Giving prominence to the opportunity costs of 6. Conclusions appliance energy requirements might counter the distraction repre­ sented by multiple appliance features. However, the extent of this The experiments conducted in this paper prove that there is a strong distraction may be higher in markets where a high number of appliance correlation between individual pure time preferences and the likelihood attributes are shown, such as the US market. they will invest in energy efficiency. Results suggest that pure temporal Economic indicators clearly influenced choice: household wealth, preferences play a role in explaining the energy-efficiency gap, and they the availability of savings and difficulty paying bills were determining can help to clarify why field experiments that provided information on factors in appliance choice. This was somewhat predictable, since in­ the running costs of different appliances have not been as effective as dividuals need the financial means to pay the price premium that comes had been hoped for in closing the energy-efficiency gap. with energy efficiency9. However, qualitative analysis of people’s ex­ While the literature has traditionally considered temporal prefer­ planations for their refrigerator choices in the Spanish pilot revealed two ences as given and constant across domains, more recent experimental opposing attitudes towards economic constraints and appliance choice. evidence suggests that intertemporal choice is sensitive to subtle vari­ Some participants felt that since they feared their future income might ations in the choice architecture (Frederick and Loewenstein, 2008). worsen, it would be best to save money immediately by buying the More specifically, within the same individual, there coexist contrasting cheaper appliance. Other participants felt that saving in the long term sets of preferences and cognitive patterns that are activated by specific through lower electricity bills would help them cope with a possibly cues. Pairing two events as part of a sequence (such as the time of lower income in the future. We thus investigated, whether the effect of purchase and future electricity payments) invites individuals to evoke individual economic situation on appliance choice was mediated by the emotions they may otherwise not have experienced and may induce treatments participants had been subjected to. them to make more forward looking choices by shifting their psycho­ Split regression analysis in the Spanish and US sample showed that logical perspective forward (Loewenstein and Prelec, 1991). people that had no savings, who were economically worse-off, who lived Findings in this paper corroborate the idea that temporal preferences in a rented home or who expected their income to decrease in the future can be activated or attenuated within a given choice context and that were less likely to choose the green appliance if they had participated in framing techniques can be leveraged to induce intertemporal choices the control group or T1. While participants in T2 with the same char­ that are both economically optimal in the long term for the individual acteristics were more likely to choose the green appliance compared to and for the climate. In this paper, simply highlighting the long-term the others. In the Spanish sample, the interaction term between eco­ hidden costs of choosing an energy-inefficient appliance by a given nomic pessimism and T2 was close to statistical significance (p = 0.111) date into the future with loss-framed language increased energy- and had a positive sign.10 This result suggests that in the status quo, efficient choices up to 24 percentage points when compared with the where electricity requirements are expressed in kWh, people worried baseline scenario and by 7 percentage points when compared with just about their future income may be tempted to follow a saving-now displaying lifetime operating costs, as done in the previous literature. strategy and choose the cheaper, less energy-efficient appliance. In We hypothesize that the effectiveness of the treatment is delivered contrast, participants in T2 received a loss-framed message which through two mechanisms: It lowers impatient choices by lowering the reframed the more expensive, efficient appliance as an opportunity to effect of impatient individuals’ default temporal preferences, but it also lower their future electricity expenses and to better cope with a lower redirects the attention of income constrained individuals towards income in the future. energy-efficiency as a money saving strategy. For the US market, part of In the US sample, we extended the analysis to political identities and the effectiveness of the treatment was also due to the fact that found that identifying to some extent with the Republican Party was mentioning the economic convenience of energy-efficient products negatively associated with the likelihood of choosing the efficient neutralized the negative effect of conservative political views of climate- appliance, even when controlling for income and future orientation. friendly initiatives. However, this relationship was not statistically significant when the The main policy implication of this paper is the suggestion to analysis was restricted to individuals in T2 and individuals who had introduce temporally oriented nudges to increase energy-efficiency up­ been subjected to treatments displaying additional appliance features. take and foster other pro-environmental behaviors. Our nudge can Climate change is currently a highly politicized topic in the US, and it is support economically constrained individuals to make choices that will possible that energy saving is seen as a Democratic Party interest, which help them to save money in the long term. would explain the Republican Party subsample’s reluctance to choose Random assignment in the experimental setup ensures there is a the efficient appliance. However, we interpret our result as an indication causal relationship between our treatments and the percentage of that when the energy-efficiency information is framed in terms of per­ energy-efficient appliances chosen. The findings were consistent across sonal economic losses (T2) or when energy efficiency information is two very different population samples and different appliance markets. dispersed among other appliance features (color, size, etc.), political This suggests that this simple and cost-effective nudge may be applicable views are not activated in the decision process. It suggests that labeling to different socio-cultural contexts. The main limitation in our experi­ energy efficiency as climate-friendly may cause counterproductive re­ ment concerns the lack of variation in the refrigerators’ price and actions from individuals identifying with the Republican Party, and it operating costs and the limited variation of the other attributes. While highlights the potential for future research. This view is supported by our choice was dictated by parsimony—i.e. the need to contain the total Gromet et al. (2013), who found that conservatives were less likely to number of permutations within a given sample size—we recognize that buy energy-efficient lightbulbs if they were labelled with environmental varying the price and the operating costs would have enabled us to messages. Overall, findings in the two experiments were consistent with define marginal effects and to assess the effectiveness of our treatment each other despite obvious differences in the sample demographic and on a wider range of appliances. Another limitation to our research is the lack of real incentives for participants. Future research could address both limitations by replicating this experiment in a field setting, it would 9 In addition, also renting one own’s living quarters impacted choice, which validate our findings and effect sizes while extending the number and is also to be expected considering that homeowners may expect to spend more the levels of appliance attributes considered. Our nudge was designed time in their homes and hence exploit their appliances for a longer time. for application in an online environment. As an extension of this paper, 10 Results from the Spanish pilot are included as part of Annex 1. one could test its introduction at different steps in the purchase process 13 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 to identify the moment that makes it most effective. Additional future DEFRA, 2010. Behavioural Economics & Energy Using Products: Scoping Research on Discounting Behaviour and Consumer Reference Points. Final Report. London. research avenues in relation to this paper include the application of a Department of Energy and Climate Change, 2014. Evaluation of the DECC/Jhon Lewis similarly inspired nudge towards other pro-environmental behaviors by Energy Labelling Trial. DECC, London. Retrieved from. http://files/2081/Joh highlighting their opportunity costs. (International Energy Agency, n_Lewis_trial_report_010914FINAL.pdf. 2021) Ebeling, F., Lotz, S., 2015. Domestic uptake of green energy promoted by opt-out tariffs. Nat. Clim. Chang. 5, 868–871. https://doi.org/10.1038/nclimate2681. Ebert, J.E.J., Prelec, D., 2007. The fragility of time: time-insensitivity and valuation of the near and far future. Manag. Sci. 53 (9), 1423–1438. https://doi.org/10.1287/ Declaration of Competing Interest mnsc.1060.0671. Enzler, H.B., 2013. Future Orientation and Energy Saving - Subjective Discount Rates and The auhtors declare no compeiting interest. Consideration of Future Consequences as Predictors of Environmentally Responsible Behavior. Zurich. Epper, T., Fehr-Duda, H., Schubert, R., 2011. Energy-Using Durables: The Role of Time Acknowledgements Discounting in Investment Decisions. Faralla, V., Novarese, M., Ardizzone, A., 2017. Framing effects in intertemporal choice: a nudge experiment. J. Behav. Exp. Econ. 71, 13–25. https://doi.org/10.1016/j. This research was supported by the Spanish Ministry of Economics socec.2017.09.002. and Competitiveness and ERDF under grants RTI2018-095484-B-I00 Frederick, S., Loewenstein, G.F., 2008. Conflicting motives in evaluations of sequences. and BES-2016-078156 and by the Time-sharing Experiments for the J. Risk Uncertain. 37 (2–3), 221–235. https://doi.org/10.1007/s11166-008-9051-z. Social Sciences, NSF Grant 0818839, Jeremy Freese and James Druck­ Frederick, S., Loewenstein, G.F., O’Donoghue, T., 2002. Time discounting and time preference: a review. J. Econ. Lit. XL, 351–401. man, Principal Investigators. Gerarden, T., Newell, R.G., Stavins, R.N., 2015. In: Stavins, Robert N. (Ed.), Assessing the The authors wish to thank Jeremy Freese and James Druckman for Energy-Efficiency Gap (Faculty Working Paper Series No. 2015–04). https://doi.org/ agreeing to financially support this project; the project managers at TESS 10.2139/ssrn.2554735. Cambridge, Mass. Goodkind, A.L., Tessum, C.W., Coggins, J.S., Hill, J.D., Marshall, J.D., 2019. Fine-scale and NORC’s team at the University of Chicago for coding and carrying damage estimates of particulate matter air pollution reveal opportunities for out the final survey on our behalf. We are grateful to professors at the location-specific mitigation of emissions. PNAS 116 (18), 8775–8780. https://doi. Department of Applied Economics at Universitat Autònoma de Barce­ org/10.1073/pnas.1816102116. Goodman, J.K., Malkoc, S.A., Rosenboim, M., 2019. The material-experiential lona for allowing us to conduct a pilot with their students and, and to our asymmetry in discounting: when experiential purchases Lead to more impatience. colleagues together with Pietro Guarnieri and Tommaso Luzzati for J. Consum. Res. 46 (4), 671–688. https://doi.org/10.1093/jcr/ucz017. inspiring conversations and for helpful comments received on an earlier Granade, H.C., Creyts, J., Derkach, A., Farese, P., Nyquist, S., Ostrowski, K., 2009. Unlocking Energy Efficiency in the U.S. Economy. McKinsey & Company, Stamford, version of this article. We also thank the editor and two anonymous CT. reviewers for their valuable comments and suggestions. Gromet, D.M., Kunreuther, H., Larrick, R.P., 2013. Political ideology affects energy- efficiency attitudes and choices. Proc. Natl. Acad. Sci. U. S. A. 110 (23), 9314–9319. https://doi.org/10.1073/pnas.1218453110. Appendix A. Supplementary data Hall, Robert, 1997. The Inkjet Aftermarket: An Economic Analysis. Stanford University. Hausman, J.A., 1979. Individual discount rates and the purchase and utilization of Supplementary data to this article can be found online at https://doi. energy-using durables. Bell J. Econ. 10 (1), 33–54. https://doi.org/10.2307/ 3003318. org/10.1016/j.eneco.2021.105563. International Energy Agency, 2021. CO2 Emissions Statistics. Retrieved March 8, 2019, from. https://www.iea.org/statistics/co2emissions/. References IPSOS & London Economics, 2014. Study on the Impact of the Energy Label and Potential Changes to it – on Consumer Understanding and on Purchase Decisions ENER/C3/ 2013-428 Final Report (October). (141). Retrieved from https://ec.europa.eu/ Allcott, H., 2011. Consumers’ perceptions and misperceptions of energy costs. Am. Econ. energy/sites/ener/files/documents/Impact of energy labels on consumer behaviour. Rev. 101 (3), 98–104. https://doi.org/10.1257/aer.101.3.98. pdf. Allcott, H., Taubinsky, D., 2015. Evaluating behaviorally motivated policy: experimental Kallbekken, S., Sælen, H., Hermansen, E.A.T., 2013. Bridging the energy efficiency gap: a evidence from the Lightbulb market. Am. Econ. Rev. 105 (8), 2501–2538. https:// field experiment on lifetime energy costs and household appliances. J. Consum. doi.org/10.1257/aer.20131564. Policy 36 (1), 1–16. https://doi.org/10.1007/s10603-012-9211-z. Allcott, H., Wozny, N., 2014. Gasoline prices, fuel economy, and the energy paradox. Langbein, L., 2015. Public Program Evaluation: A Statistical Guide (First). Routledge, Rev. Econ. Stat. 96 (5), 779–795. https://doi.org/10.1162/REST_a_00419. London. Andor, M.A., Gerster, A., Sommer, S., 2020. Consumer inattention, heuristic thinking and Lewis, A., 2018. The Cambridge Handbook of Psychology and Economic Behaviour, 587. the role of energy labels. Energy J. 41 https://doi.org/10.5547/01956574.41.1. Cambridge University Press. mand. Loewenstein, G.F., Prelec, D., 1991. Negative time preference. Am. Econ. Rev. 81 (2), Association of Home Appliance Manufacturers (AHAM), 2020. Total Unit Shipments of 347–352. Retrieved from. http://www.jstor.org/stable/2006883. Refrigerators in the US. Retrieved July 8, 2020, from. https://www.statista.com/st Loewenstein, G.F., Prelec, D., 1993. Preferences for sequences of outcomes. Psychol. Rev. atistics/220111/unit-shipments-of-refrigerators-and-freezers-in-the-us/. 100 (1), 91–108. Attanasi, G., Georgantzís, N., Rotondi, V., Vigani, D., 2018. Lottery- and survey-based Loewenstein, G.F., Sicherman, N., 1991. Do workers prefer increasing wage profiles? risk attitudes linked through a multichoice elicitation task. Theor. Decis. 84 (3), J. Labor Econ. 9 (1), 67–84. Retrieved from. http://www.jstor.org/stable/2535114. 341–372. https://doi.org/10.1007/s11238-017-9613-0. Lutz, J.D., Hopkins, A., Letschert, V., Franco, V.H., Sturges, A., 2011. Using national Attari, S.Z., DeKay, M.L., Davidson, C.I., Bruine de Bruin, W., 2010. Public perceptions of survey data to estimate lifetimes of residential appliances. HVAC R Res. 17 (5), energy consumption and savings. Proc. Natl. Acad. Sci. 107 (37), 16054–16059. 726–736. https://doi.org/10.1080/10789669.2011.558166. https://doi.org/10.1073/pnas.1001509107. Magen, E., Dweck, C.S., Gross, J.J., 2008. The hidden zero effect: representing a single Berns, G.S., Laibson, D., Loewenstein, G.F., 2007. Intertemporal choice – toward an choice as an extended sequence reduces impulsive choice. Psychol. Sci. 19 (7). integrative framework. Trends Cogn. Sci. 11 (11), 482–488. https://doi.org/ Newell, R.G., Siikamäki, J., 2015. Individual time preferences and energy efficiency. Am. 10.1016/j.tics.2007.08.011. Econ. Rev. 105 (5), 196–200. https://doi.org/10.1257/aer.p20151010. Blasch, J., Filippini, M., Kumar, N., 2019. Boundedly rational consumers, energy and OECD, 2017. Tackling Environmental Problems with the Help of Behavioural Insights. investment literacy, and the display of information on household appliances. Resour. OECD Publishing. Retrieved from. http://www.oecd-ilibrary.org/environment/tackl Energy Econ. 56, 39–58. https://doi.org/10.1016/j.reseneeco.2017.06.001. ing-environmental-problems-with-the-help-of-behavioural-insights_9789264 Carson, R.T., Flores, N.E., Martin, K.M., Wright, J.L., 1996. Contingent valuation and 273887-en. revealed preference methodologies: comparing the estimates for quasi-public goods. Read, D., Frederick, S., Orsel, B., Rahman, J., 2005. Four score and seven years from Land Econ. 72 (1), 80–99. https://doi.org/10.2307/3147159. now: the date/delay effect in temporal discounting. Manag. Sci. 51 (9), 1309–1448. Cattaneo, C., 2019. Internal and external barriers to energy efficiency: which role for Read, D., Olivola, C.Y., Hardisty, D.J., 2017. The value of nothing: asymmetric attention policy interventions? Energy Efficiency 12, 1293–1311. https://doi.org/10.1007/ to opportunity costs drives Intertemporal decision making. Manag. Sci. 63 (12), s12053-019-09775-1. 4277–4297. https://doi.org/10.1287/mnsc.2016.2547. Chapman, G.B., 1998. Sooner or later: The psychology of intertemporal choice. In: Rohling, M., Schubert, R., 2013. Energy labels for household appliances and their Medin, D.L. (Ed.), The Psychology of Learning and Motivation: Advances in Research disclosure format: a literature review. In: IED Working Paper, 21 (10). and Theory. Academic Press, San Diego, CA, US, pp. 83–113. Schleich, J., Gassmann, X., Meissner, T., Faure, C., 2019. A large-scale test of the effects Chapman, G.B., 2000. Preferences for improving and declining sequences of health of time discounting, risk aversion, loss aversion, and present bias on household outcomes. J. Behav. Decis. Mak. 13 (2), 203–218. https://doi.org/10.1002/(SICI) adoption of energy-efficient technologies. Energy Econ. 80, 377–393. https://doi. 1099-0771(200004/06)13:2<203::AID-BDM317>3.0.CO;2-S. org/10.1016/j.eneco.2018.12.018. Davis, L.W., Metcalf, G.E., 2016. Does better information lead to better choices? Evidence from energy-efficiency labels. J. Assoc. Environ. Resour. Econ. 3 (3), 589–625. https://doi.org/10.1086/686252. 14 M. Silvi and E. Padilla Rosa Energy Economics 103 (2021) 105563 Scholten, M., Read, D., Sanborn, A., 2016. Cumulative weighing of time in intertemporal US Appliance Industry Saturation, 2000. 23rd Annual Portrait of The U S Appliance tradeoffs. J. Exp. Psychol. Gen. 145 (9), 1177–1205. https://doi.org/10.1037/ Industry Saturation. In: Appliance Magazine, 57, pp. 83–110. Retrieved from. xge0000198. https://www.tib.eu/en/search/id/BLSE%3ARN084175500/23rd-Annual-Portrait-o Schubert, R., Stadelmann, M., 2015. Energy-using durables: driving forces of purchase f-The-U-S-Appliance-Industry/. decisions. Front. Energy Res. 3 (7), 1. https://doi.org/10.3389/fenrg.2015.00007. US Bureau of Labor Statistics, 2020. Consumer Expenditures in 2018. Retrieved July 8, Skourtos, M., Damigos, D., Tourkolias, C., Kontogianni, A., 2021. Efficient energy 2020, from. https://www.bls.gov/opub/reports/consumer-expenditures/2018/hom labelling: the impact of information content and style on product choice. Energy e.htm. Efficiency 14, 58. https://doi.org/10.1007/s12053-021-09950-3. US Energy Information Administration, 2020. Electric Power Montly. Retrieved July 8, Stadelmann, M., Schubert, R., 2018. How do different designs of energy labels influence 2020, from. https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=ep purchases of household appliances? A field study in Switzerland. Ecol. Econ. 144 mt_5_6_a. (August 2017), 112–123. https://doi.org/10.1016/j.ecolecon.2017.07.031. Waechter, S., Sütterlin, B., Siegrist, M., 2015a. The misleading effect of energy efficiency Strathman, A., Gleicher, F., Boninger, D.S., Edwards, C.S., 1994. The consideration of information on perceived energy friendliness of electric goods. J. Clean. Prod. 93, future consequences: weighing immediate and distant outcomes of behavior. J. Pers. 193–202. https://doi.org/10.1016/J.JCLEPRO.2015.01.011. Soc. Psychol. 66 (4), 742–752. https://doi.org/10.1037/0022-3514.66.4.742. Waechter, S., Sütterlin, B., Siegrist, M., 2015b. Desired and undesired effects of energy Taubinsky, D., Allcott, H., 2013. The Lightbulb paradox: using behavioral economics for labels - an eye-tracking study. PLoS One 10, e0134132. https://doi.org/10.1371/ policy evaluation. In: NBER Working Papers, 19713, pp. 1–51. Retrieved from. htt journal.pone.0134132. p://www.stanford.edu/group/SITE/SITE_2013/2013_segment_6/2013-segment Weber, E.U., Johnson, E.J., Milch, K.F., Chang, H., Brodscholl, J.C., Goldstein, D.G., _6_papers/allcott.pdf%5Cnfile:///Files/77/77A46F55-D2CA-4553-AFEC-2F1CF4A2 2007. Asymmetric discounting in intertemporal choice. Psychol. Sci. 18 (6), 9C30.pdf. 516–523. https://doi.org/10.1111/j.1467-9280.2007.01932.x. Train, K., 1985. Discount rates in consumers’ energy-related decisions: a review of the Wu, C.-Y., He, G.-B., 2012. The effects of time perspective and salience of possible literature. Energy 10 (12), 1243–1253. https://doi.org/10.1016/0360-5442(85) monetary losses on intertemporal choice. Soc. Behav. Personal. Int. J. 40 (10), 90135-5. 1645–1653. https://doi.org/10.2224/sbp.2012.40.10.1645. Tversky, A., Sattah, S., Slovic, P., 1988. Contingent weighting in judgment and choice. Psychol. Rev. 95 (3), 371–384. https://doi.org/10.1037/0033-295X.95.3.371. 15