Energy Labels Increase Demand for Compact Fluorescent Bulbs: Analyzing Consumer Preferences for Lighting Technologies Using Discrete Choice Analysis

Size: px
Start display at page:

Download "Energy Labels Increase Demand for Compact Fluorescent Bulbs: Analyzing Consumer Preferences for Lighting Technologies Using Discrete Choice Analysis"

Transcription

1 Energy Labels Increase Demand for Compact Fluorescent Bulbs: Analyzing Consumer Preferences for Lighting Technologies Using Discrete Choice Analysis Jihoon Min, Ines Azevedo, Jeremy Michalek, Wändi Bruine de Bruin, Department of Engineering and Public Policy, Baker Hall 129, Carnegie Mellon University, 5000 Forbes Ave, Pittsburgh, PA 15213, United States Tel (412) , Fax (412) ABSTRACT Lighting accounts for nearly 20% of overall U.S. electricity consumption and 18% of U.S. residential electricity consumption. Incandescent bulbs are responsible for the majority of residential lighting electricity usage. A transition to alternative energy-efficient technologies, like compact fluorescent lamps (CFL) and light-emitting diodes, could reduce this energy consumption considerably. To quantify the influence of factors that drive consumer choices for light bulbs, we conducted a choice-based conjoint field experiment with 183 participants and estimated several discrete choice models from the data. We find that environmentally minded consumers have a stronger preference for compact fluorescent lighting technology, all else being equal, while politically liberal consumers have a stronger preference for low energy consumption. Perceived personal experiences of health issues, previous use or purchase of CFLs, awareness on climate change, income, and education levels were not significant in explaining choices. Greater willingness to pay for lower energy consumption and longer life was observed in conditions where estimated operating cost information was provided. Providing estimated annual cost information to consumers reduces their implicit discount rate by a factor of five, lowering barriers to adoption of energy efficient alternatives with higher up-front costs; however, even with cost information provided, consumers continue to use implicit discount rates in excess of 90%, which is greater than that experienced in most other areas of consumer choices for energy technologies. 1. INTRODUCTION In 2008, residential CFL socket saturation was 21% in California (Canseco 2009) and 11% nationwide (Bickel et al. 2009), with the remainder being almost entirely incandescent bulbs. This shows that further adoption of CFLs or other efficient lighting technologies could achieve considerable energy savings in the residential sector. The main goal of this study is to understand consumers preferences for a specific energy end-use service: general illumination. We start by quantifying the importance of different factors affecting consumers choices using 1

2 discrete choice conjoint analysis. We then unpack the issue of high implicit discount rates. This methodology enables policy decision-makers to understand through choice models why some consumers choose to buy energyefficient light bulbs while others do not by measuring the impact of each relevant factor on consumer choices. These factors are categorized into two parts: factors that relate to individual consumer characteristics and factors that related to the technology characteristics. The former includes income, education, housing characteristics, environmental/political attitudes, or awareness of climate change or toxicity issues, while the latter includes price, wattage, brightness, lifetime, and technology type. From this method, we derived estimates of implicit discount rates independent of other influential factors included in the model. Another goal of this study is to understand how disclosing information on operating costs affects technology choices. Figure 1 shows a new label that will be mandated starting in 2012 by the Federal Trade Commission (FTC 2010). By showing how consumer preferences are shaped and how much each factor contributes to the choices of a certain lighting technology, this study will not just help policymakers design policy options that promote the adoption of energy efficient lighting but also provide a method that can be expanded to be used with other energy efficient technologies. Figure 1. New front label for light bulbs (FTC 2010). The FTC mandates that annual operating cost information needs to be included on packages for general illumination light bulbs, beginning in January By showing how consumer preferences are shaped and how much each factor contributes to the choices of a certain lighting technology, this study will not just help policymakers design policy options that promote the adoption of energy efficient lighting but also provide a method that can be expanded to be used with other energy efficient technologies. 2. METHODS Two main goals of this study are, (1) to quantify consumer preferences for attributes of lighting technologies and to estimate implicit discount rates specifically for lighting sector, and (2) to observe the impact of disclosing operating cost information on consumers behavioral patterns. For these purposes, we create a controlled experiment using choice-based conjoint analysis, and we use the data to estimate several random utility discrete 2

3 choice models. We collected data from 183 participants who were recruited randomly in Squirrel Hill, Pittsburgh, PA from September 30 th to October 3 rd, The experiment was performed in a mobile lab equipped with seven laptops running the choice tasks and the survey Experiment Design A choice experiment was carefully designed considering efficiency of the design, cognitive burden for participants, and similarity to the experience of making product choices in the marketplace. We designed a field experiment that consisted of three main parts: 1) choice experiment, 2) choice of a real light bulb, and 3) questions on demographics, experience, knowledge, and attitudes. To observe the effect of disclosing annual cost information, subjects were randomly assigned to one of two groups. Half of them were shown annual operating cost information in their choice tasks while the other half were not. From this point, the group provided with the information is referred to as with-cost group and the group without it as without-cost group Part 1: conjoint choice experiment The choice experiment involved twelve randomized choice tasks and three fixed choice tasks. Figure 2 shows an example of a choice task. Two among the three fixed choice tasks are intended to check whether participants are paying attention to the experiment and have a dominant alternative (a cheaper, more efficient, longer-lasting one). Each alternative in a choice task had six attributes (type, price, power, lifetime, brightness, and color), and a randomly chosen half of the participants were given annual cost information (computed from power). Attribute levels used in the 12 random choice tasks are shown in Table 1. Attribute levels in these choice tasks are generated by Sawtooth software, which created alternatives (three levels for five attributes and two levels for one attribute) and arranged them to generate different versions of choice experiments which were balanced and near-orthogonal (Kuhfeld 1997). Only for the with-cost group, power values were presented together with cost information in parentheses as shown in Figure 2. The third fixed task was a hold-out choice task designed to determine the compensation for participation for participants. A pair of real light bulbs matching the choice that participants made in this task was given to participants as compensation for participation at the end of the experiment. Participants were informed beforehand that they would be compensated with a type of light bulbs decided based on their choices, but they were not told which specific task determined the type. This can potentially incentivize people who might otherwise have chosen lower priced bulbs to choose the expensive bulbs. This would lead to inflated price and life coefficients and deflated power coefficients. However, we assume these effects are not significant. Ding et al. (2005) tested adding an incentive among the conjoint choice tasks and observed that this method helps participants to make choices that are closer to their true preference, reducing the limitation of observing only stated preference. 3

4 Figure 2. Example of a choice task seen by participants. Table 1. Attributes and levels used in choice experiment. Attributes Levels 1. Type CFL / Incandescent 2. Price ($) 0.49 / 2.49 / Power (W) 9 / 27 / Color Soft White / Bright White / Daylight 5. Lifetime (hour) 1000 / 8000 / Brightness (lumen) 500 / 1200 / Part 2: physical choice among market products In addition to the choice experiment on computers, participants were asked to choose one bulb option among five physical products from the marketplace in their original packages. The choices from this step were compared with the predictions from our model to assess external validity of the model Part 3: demographics, experience, knowledge, and attitudes This part contained questions related to demographics, personal attitudes, knowledge, or experience that might influence preference for lighting. Median tiers for income, education, and age are $30-50k per year, bachelor s degree, and between age of 25 and % of the participants were male, 41% owned their houses, and 17% have children. We also asked participants to rank the ten major technical factors that would affect their choice for light bulbs. When rankings of these factors were averaged numerically both with- and without-cost groups showed the same decreasing order: Brightness > Price > Lifetime > Energy Cost > Color > Wattage > Type > Wattage Equivalent > Time to Full Brightness > Shape. 4

5 2.2. Model Development We adopted a mixed logit model (Train 2003), which models heterogeneity of consumer preferences via random coefficients (i.e. the preference coefficients are random variables distributed over the population) and avoids the restrictive substitution patterns of the multinomial logit model. In our specification, the utility U ij that consumer i draws from product alternative j is modeled as: K N U ij = V ij + ε ij = β k x jk + γ kn z in x jk + ε ij, (1) k=1 n=1 where β k is the preference coefficient for attribute k, x jk is the k-th attribute of alternative j, γ kn is the coefficient for interactions between consumer attribute n and product attribute k, z in is the n-th attribute of consumer i, and ε ij is the random error term, taken as an iid standard Gumbel distribution (Train 2003). The interaction terms z in x jk are intended to reveal how individual characteristics can affect preference for bulb attributes. We first estimated a basic homogeneous part-worth model using only categorical explanatory variables (see Table 13). The resulting coefficients for numerical bulb attributes (price, power, lifetime, and brightness) in this model suggest that we can safely assume that utility is linear in price, lifetime and power without introducing substantial model artifacts. Preference for the brightness attribute appears non-monotonic and calls for a higher order specification. Equations (2) and (3) define Model 1 and Model 2 estimated with continuous specification for all of the numerical (ratio-type) attributes, where explanations for variable names can be found in Table 2; the fit coefficients include all β, σ, and γ terms; ν represents a random variable with an iid standard normal distribution defining general population preference heterogeneity; and ε is a random error variable with an iid Gumbel distribution. We assumed preference for type varies normally in the population and preference for price varies lognormally, since positive coefficients for price would be counterintuitive and theoretically problematic. Considering our focus is on preference heterogeneity for bulb types, we do not estimate heterogeneity distributions for other parameters. We estimated two more models in a similar way including variables related to perceptions on climate change or toxicity, environmental attitudes, and political views. U M1 ij = β 1 + σ 1 ν 1i type ij exp β 2 + σ 2 ν 2i price ij + β 3 life ij + β 4 bght ij + β 5 bght 2 ij + β 6 watt ij 2 + β 7,n color mij m=1 + D ocost,i β 1c type ij + β 2c price ij + β 3c life ij + β 4c bght ij + β 5c bght 2 ij + β 6c watt ij 2 + β 7c,n color mij + ε ij m=1 (2) U M2 ij = U M1 ij + γ 1 expr i + γ 2 buy i + γ 3 health i + γ 4 bchlr i + γ 5 midinc i + γ 6 highinc i type ij + γ 7,n bchlr i + γ 8,n midinc i + γ 9,n highinc i watt ij (3) 5

6 Table 2. Descriptions of variables used in the models Variable Description Value type ij Dummy indicating bulb type 0: incandescent, 1: CFL price ij Price of the bulb j in subject i s choice task $0.49 / $2.49 / $4.49 color 1ij, color 2ij Dummy for color, where color 1ij is bright white and color 2ij is daylight 0: No, 1: Yes life ij Lifetime of the bulb j in subject i s choice task 1,000/8,000 /12,000 [hours] bght ij Brightness level of the bulb j in subject i s choice task 500/1,200/1,800 [lumens] watt ij Power consumption of the bulb j in subject i s choice task 9/25/75 [watt] D ocost,i Dummy indicating whether annual operating cost information is provided to subject i 0: No, 1: Yes expr i Dummy indicating whether subject i has used CFLs before 0: No, 1: Yes buy i Dummy indicating whether subject i buys light bulbs sometimes 0: No, 1: Yes health i Dummy indicating whether subject i has experienced any health issues related to CFL use 0: No, 1: Yes bchlr i Dummy indicating whether subject i has a bachelor s degree 0: No, 1: Yes midinc i, highinc i Dummy indicating subject i's annual household income, where midinc i is between $30k and $75k and highinc i is above $75k 0: No, 1: Yes Implicit discount rates that can be estimated from the model above by using a conventional method as was used by Hausman (1979), is not useful, because it is not appropriate to assume a common lifetime of a CFL and an incandescent bulb, considering the vast difference between lifetimes of the two types. Instead, we could estimate the implicit discount rate explicitly from a specification using annualized cost. The basic specification for estimating implicit discount rate is β 1 U ij = β 0 price 1 ij + Ocost ij + ε ij, (4) 1 (1 + β 1 ) life ij where β 0 represents consumer sensitivity to annualized cost of ownership and β 1 represents the consumer s implicit discount rate. The variables price ij and Ocost ij represent price and operating cost of alternative j that consumer i sees. The life ij term is expressed in years. In order to control effects from other relevant factors, we also estimate two more specifications: β 1 U ij = β 0 price 1 ij + Ocost ij + β2 type ij + β 3 color 1ij β 1 life ij +β 4 color 2ij + β 5 bght ij + β 6 bght 2 ij + ε ij, (5) 6

7 β 1 U ij = β 0 price 1 ij + Ocost ij + β2 type ij + β 3 color 1ij β 1 life ij 2 +β 4 color 2ij + β 5 bght ij + β 6 bght ij +β 7 type ij NEP i + β 8 watt ij liberal i + ε ij, (6) where NEP i and liberal i represent subject i's New Ecological Paradigm (NEP) score 1 and whether the subject is politically liberal, which were the most significant factors in Model 4 in Table. Through maximum likelihood estimation, we can estimate β 1 representing the population s implicit discount rate employed when making purchasing decisions for any lighting products. 3. RESULTS AND DISCUSSION Fifteen among the 183 subjects were removed from the analysis who did not pick dominant answers from either of the two choice tasks designed to check whether participants were paying attention. Thus, the data collected from a total of 168 subjects were used for this analysis. We first estimate four models based on Equation (1), and we later estimate three nonlinear models based on Equation (2) which incorporate implicit discount rates. First, we fit a basic model including only bulb-related factors and operating cost information. In the second model, we add interactions with basic demographic information (income and education) and variables that explain perceptions and experience regarding lighting technologies. In the third and fourth models, we retained only those factors which appeared significant in the previous model. The third model includes variables related to perceptions on climate change and toxicity of lighting technologies; and variables that measure participants awareness of the relationships between bulb attributes. For this model, we introduced a linear scale index ranging from 0 to 4 which corresponds to the number of correct answers among the four questions on basic knowledge. The last model included variables regarding personal attitudes: New Environmental Paradigm (NEP) score and political orientation. These four models provide our main results. We present only the statistically significant factors in Table 3. Multi-collinearity was not a concern because the variance inflation factor was around three for all model variables. Across the four models, log likelihood values increased only slightly (from to -2292). This implies that more complicated models do not necessarily have higher predictive capability. For that reason, it will be enough to use the first model to predict collective market status or market behavior. 1 We conjecture that personal values pertaining to environment or energy issues have influence on preference for energyefficient products. As a measure of general environmental orientation, we adopted New Ecological Paradigm (Dunlap et al. 2000), which is widely used in environmental sociology. However, because its recent version consists of 15 questions which can be too demanding for participants, instead we used an abbreviated version with five questions, which is reported to have a considerable internal consistency (Cronbach α = 0.72) and a large correlation (0.78) with the regular NEP scale (Cordano et al. 2003). 7

8 Table 3. Estimation results from the four models. This table shows only significant variables. Variables Model 1 Model 2 Model 3 Model 4 color: base=soft white color=bright white * (0.0889) * (0.0890) * (0.0889) * (0.0890) color=daylight (0.0862) (0.0861) (0.0861) (0.0862) watt ** ( ) ( ) * ( ) ( ) watt*d ocost *** ( ) *** ( ) *** ( ) *** ( ) life (x10 3 hours) *** ( ) *** ( ) *** ( ) *** ( ) life*d ocost ** (0.0113) ** (0.0113) ** (0.0113) ** (0.0113) brightness (x10 3 lumens) 1.621*** (0.417) 1.623*** (0.417) 1.620*** (0.417) 1.604*** (0.417) brightness *** (0.178) *** (0.178) *** (0.178) *** (0.178) (type=cfl)*mid_income * (0.220) * (0.225) * (0.220) (type=cfl)*high_income (0.246) (0.238) (0.237) (type=cfl)*(toxic=very serious)*toxiccfl * (0.604) * (0.598) (type=cfl)*(toxic=very serious)*toxicboth (0.870) (0.860) watt*(cc= not very serious) ( ) ( ) watt*(cc= somewhat serious) ( ) ( ) watt*(cc= very serious) ** ( ) ( ) watt*(cc= not aware) ( ) ( ) (type=cfl)*nep score ** (0.0238) watt*liberal *** ( ) type=cfl 0.528*** (0.136) 0.937*** (0.306) (0.584) (0.497) std. dev *** (0.0987) 0.993*** (0.0977) 0.964*** (0.0971) 0.960*** (0.0982) price *** (0.278) *** (0.278) *** (0.270) *** (0.274) std. dev *** (0.189) 1.228*** (0.192) 1.190*** (0.182) 1.178*** (0.177) Log likelihood Observations 7,560 7,560 7,560 7,560 Standard errors in parentheses *** p<0.01, ** p<0.05, * p< Discussion How do bulb-specific factors affect consumer choices? From Model 1, we observe that, all else being equal, consumers prefer lower power, longer life, lower prices, CFL technology, and a relatively high level of brightness. Consumers appear to prefer soft white color over bright white color with p<0.1, but preference for color are otherwise not significant. Preference for low power and long life increase significantly when operation cost information is provided on the label. There is high variation in preference for bulb type in the population, with some consumers preferring CFL and others preferring incandescent, all else being equal. There is also substantial variation in price sensitivity in the population. On average, (using the mean of the lognormal distribution for price coefficients) subjects are willing to pay $2.30 more for CFL bulbs than incandescent bulbs, all else being equal; however, there is considerable variance, with some consumers willing to pay more for incandescent bulbs. Consumers are willing to pay $0.43 more for every 1,000 hours of lifetime increase within the range tested in the experiment (1,000 ~ 12,000 hours), and that amount increases by $0.12 when they are shown annual cost estimates. Consumers are willing to pay $0.12 per 10 watt decrease in power requirements, all else being equal, and that number increases by $0.32 when they are shown 8

9 annual cost estimates. Consumers are willing to pay $1.70 for a 500 lumen increase from 500 to 1,000 lumens but only $0.50 to increase from 1,000 to 1,500 lumens. Considering that the attribute values used in the choice experiment cover most common ranges of attribute values available in the market, the factor that can provide the widest range of consumer utility is lifetime, followed by price, power, type, brightness, and color, when the operating cost is provided. Throughout the four models, we observe that the significance of most coefficients for main technical features of bulbs does not change. The only change was that the main effect of power became statistically insignificant as we add consumer-specific covariates. This is because, in higher level models, preference for low energy use is explained by other individual factors such as a more liberal political view. We should note that, because power information in our design was independent of brightness, we intended that it does not represent brightness but just electricity consumption of a bulb How do consumer-specific factors affect consumer choices? When the demographic and experience variables were added in Model 2 in Table 3, we observed lower preference for CFLs among middle-income class ($30k-$75k/year) than in other income brackets. In Model 3, two additional terms were significant: 1) subjects who respond that CFLs contain toxic materials and think the toxicity problem is very serious are, as expected, very unlikely to choose CFLs, 2) those who think climate change is a very serious issue do not like higher energy consuming light bulbs. However, the second point becomes not significant in Model 4, where attitude variables of environmental or political orientation are controlled for. This last model shows that a higher NEP score is significantly associated with higher preference for CFLs and that the participants who are politically liberal prefer a bulb consuming less energy. Importance of attitude variables in consumer decision making has been emphasized in multiple studies especially in the transportation sector (Vredin Johansson et al. 2006). We observe that those findings apply to lighting purchase decisions as well. We found participants whose NEP scores are 5 points higher than others will have a 27% higher odds ratio of choosing CFLs over incandescent bulbs. When a liberal consumer is choosing between two bulbs, one of which consumes 20W less electricity than the other one, he is 10% more likely to choose the one with lower energy use than a nonliberal consumer, all other things being equal. In the last model, the part-worth of the type variable becomes statistically insignificant suggesting that preference for bulb types is mainly induced by different levels of toxicity or environmental awareness How does disclosing anual operating cost information impact choices? Based on the first model we test the impact of disclosing operation cost information. We show that having operating cost information is related to higher preferences for longer lifetime and lower power, while it lacks significant influence on choices for color, brightness, type, and price. For example, when the operating cost information is given, a consumer is willing to pay $0.12 more for a 1,000-hour increase of lifetime and $0.33 less for a 10W increase of power, compared to the case where he does not see the information. A potential explanation for this can be that when the annual operating cost information is given, consumers tend to pay more attention to the implications of lifetime and power on future savings. The fact that lower power and longer lifetime affect consumer choices less when operating cost information is not shown can be a potential reason why CFLs are still underperforming in the market. The impact of operating cost information might potentially be due to just the fact that there is more imformation for each subject to process. However, if the latter is the case, we expect that it will have similar impact on other attributes such as color and brightness, which was not observed. Thus, our results suggest that disclosing 9

10 operating cost information will have a significant impact on inducing more users to buy light bulbs with lower energy consumption and longer lifetime What are the implicit discount rates that consumers use when making choices for lighting technologies? We fit three different nonlinear models, as represented by Equation (2), separately for with- and without-cost groups. The discount rate estimates from the three models are presented in Table 4. The first model ( Basic column in Table 4) only includes the most basic bulb characteristics related to cost calculation: price, operating cost, and lifetime. In the second model ( Bulb column), we include other bulb characteristics (type, color and brightness). Then, in the last model ( Attitude column), individual attitude variables of NEP and political view that were significant in the fourth model above are included. We find that implicit discount rates range from 96% to 120% for with-cost group and from 550% to 580% for without-cost group. The results suggest that participants value future savings more when they were given the operating cost information than when they were not. Considering the Energy Information Agency s National Energy Modeling System (NEMS) adopts discount values mostly under 100% for various residential end-uses (Brown et al. 2000), the relatively large implicit discount rates estimated for lighting can be explained by the fact that savings from individual energy efficient light bulbs are normally smaller than savings from other energy efficient appliances. Discount rates for the smaller rewards were reported to be larger (Green et al. 1997). This finding suggests that lighting can face a higher barrier than other technologies with regard to the perception of operating cost information and potential reductions in energy bills. Table 4. Estimates of implicit discount rates depending on the availability of operation cost information. Model type Implicit discount rates Basic Bulb Attitude Operating cost shown 112% (21%) 119% (22%) 96% (22%) Operating cost not shown 564% (81%) 553% (67%) 576% (75%) Note: standard errors in parentheses 3.2. Model validation through physical choice observations To examine the predictive accuracy of the estimated model, we first calculated population-wide choice probabilities of the three alternatives in the hold-out task. Concurrent to this, we used our model to predict choice probabilities for the physical choice task which was the second part of our experiment. From these calculations, we intend to see if the model can go beyond what it was designed to do and predict choices in a more complex situation with additional unobserved attributes. Table 5. Summary of aggregate prediction accuracy. The first column shows how well the estimated model fits with the observed data. The second column is about the predictive performance of the model. The last column indicates how well this model behaves in a realistic setting with additional unobserved attributes. Estimation data Hold-out task Physical choice Model Random Model Random Model Random Log-likelihood Avg. share prediction error 3% 10% 6% 10% N 2520=168*

11 Table 5 summarizes model fit (estimation data) and predictive accuracy (compensation and physical choice tasks) compared with a random model that treats all choice alternatives as equally likely. The average share prediction error provides a metric for summarizing aggregate prediction accuracy. In the hold-out conjoint task, the model predicts share with an average of 3% error, compared to 10% error for a random model where a choice is made randomly with equal probabilities. In the physical choice task, which involves unobserved attributes, the model predicts share with an average of 6% error, compared to 10% error for a random model. These metrics suggest that other attributes present only in the physical task, such as brand or packaging, are important for predicting share. However, holding these elements constant, the model appears to do well capturing the drivers of choice behavior and predicting share. 4. CONCLUSIONS We examined reasons for limited adoption of compact fluorescent bulbs using a choice-based conjoint experiment to quantify the effect of product and consumer attributes on consumer choice in conditions where annual operating cost estimates are disclosed vs. withheld. Our results suggest that consumer choices are significantly affected by most bulb characteristics tested, including color, brightness, lifetime, power, type, and price. When each attribute is varied within the ranges observed in the market, lifetime has the largest influence on consumer choices, followed by price, power, type, brightness, and color. CFL toxicity and environmental/political attitudes are the consumer-specific factors that have significant influence on preferences, while other personal attributes did not have significant effects on consumer choices. This result suggests that policy makers can set priorities in implementing energy efficiency policies. A policy intervention that directly affects real or perceived bulb features, such as disclosing operating cost information, providing a subsidy, or improving lifetime or energy use is likely to be more effective in promoting the adoption of efficient light bulbs than educational programs or campaigns on topics like health issues, climate change, or technical details of light bulbs. Also, our results suggest that consumer-specific characteristics are not as significant in predicting consumer choices as bulb characteristics. We found that providing operating cost information induces consumers to prefer bulbs with longer lifetime and lower energy consumption. They were willing to pay $0.12 more for a 1,000-hour increase of lifetime and $0.33 less for a 10W increase of power when the cost information is given than when it is not. Implicit discount rates decreased from over 500% to 100% when respondents were provided annual operating cost information. This suggests that consumers weigh future savings more strongly when the information is given. Even when the information is available, the estimated implicit discount rate for individual lamp choices of around 100% is larger than most values used for other technology types in the NEMS model. Our findings can be meaningfully used to update such models. The combination of these two findings put the new FTC labeling rule on a strong footing. While this study focuses on the impacts of disclosing the cost information on the new label, the FTC could further investigate potential improvement of the new label through additional choice experiments or other marketing tools. Future studies can examine why the discount rates are so high for lighting and whether alternative models such as hyperbolic discounting or models that account for satisficing behavior can explain consumer choices better than traditional economic discounting. The proposed approach can be used to estimate the effectiveness of other potential policy interventions by computing estimated changes in share of choices resulting from the introduction of changes in certain bulb 11

12 characteristics affected by those interventions. Future studies can examine why the discount rates are so high for lighting and whether alternative models such as hyperbolic discounting or models that account for satisficing behavior can explain consumer choices better than traditional economic discounting. Bibliography Bickel, S., T. Swope and D. Lauf (2009). Energy Star CFL Market Profile, The United States Department of Energy. Brown, M. A., M. D. Levine, W. Short and J. G. Koomey (2000). Scenarios for a clean energy future, Oak Ridge National Laboratory. Canseco, J. (2009). California Compact Fluorescent Lamp (CFL) Market Overview, KEMA, Inc. Cordano, M., S. A. Welcomer and R. F. Scherer (2003). "An analysis of the predictive validity of the New Ecological Paradigm scale." The Journal of Environmental Education 34(3): Ding, M., R. Grewal and J. Liechty (2005). "Incentive-aligned conjoint analysis." Journal of Marketing Research 42(1): Dunlap, R. E., K. D. Van Liere, A. G. Mertig and R. E. Jones (2000). "New trends in measuring environmental attitudes: measuring endorsement of the new ecological paradigm: a revised NEP scale." Journal of social issues 56(3): FTC (2010). Appliance Labeling Rule. 16 CFR Part 305. F. T. Commision. Green, L., J. Myerson and E. McFadden (1997). "Rate of temporal discounting decreases with amount of reward." Memory & Cognition 25(5): Hausman, J. (1979). "Individual discount rates and the purchase and utilization of energy-using durables." The Bell Journal of Economics 10(1): Kuhfeld, W. F. (1997). Efficient experimental designs using computerized searches. Train, K. (2003). Discrete choice methods with simulation, Cambridge University Press. Vredin Johansson, M., T. Heldt and P. Johansson (2006). "The effects of attitudes and personality traits on mode choice." Transportation Research Part A: Policy and Practice 40(6):