- Email: [email protected]
Compact fluorescent lighting and residential natural gas consumption: Testing for interactive effects
ARTICLE IN PRESS Energy Policy 38 (2010) 1288–1296
Contents lists available at ScienceDirect
Energy Policy journal homepage: www.elsevier.com/locate/enpol
Compact fluorescent lighting and residential natural gas consumption: Testing for interactive effects Eric J. Brunner a,n, Peter S. Ford b, Mark A. McNulty c, Mark A. Thayer d a
Department of Economics, Quinnipiac University, 275 Mount Carmel Avenue, Hamden, CT 06516, USA San Diego Gas and Electric Company, 8335 Century Park Court, San Diego, CA 92123, USA c Mark McNulty and Associates, 4654 Mayapan Drive, La Mesa, CA 91941, USA d Department of Economics, San Diego State University, San Diego, CA 92182, USA b
a r t i c l e in f o
a b s t r a c t
Article history: Received 15 March 2009 Accepted 3 November 2009 Available online 9 December 2009
Replacing incandescent light bulbs with compact fluorescents (CFLs) has traditionally been seen as a cost effective means of promoting energy conservation. Recently, however, the magnitude of energy savings associated with CFLs has been called into question. Specifically, recent findings suggest an ‘‘interactive effect’’ associated with the replacement of incandescent light bulbs with CFLs in the residential sector. In this scenario, the reduced wattage of CFLs, relative to incandescent bulbs, generates less heat, which in turn, requires additional natural gas usage during the heating season. Engineering studies suggest the magnitude of the effect is significant in energy terms, which implies that the energy savings associated with CFLs may be significantly overstated. In this paper, we use billing analysis to test for the presence of interactive effects. Our analysis is based on a comprehensive dataset that includes monthly household electricity and natural gas usage, the number of CFL bulbs installed, the installation date, and a set of household characteristics. Our results suggest that CFLs do indeed save electricity. However, we do not find any support for the hypothesis that CFLs cause increased usage of natural gas. & 2009 Elsevier Ltd. All rights reserved.
Keywords: Energy efficiency Residential energy consumption
1. Introduction In the most recent Annual Energy Outlook (AEO 2008) from the US Energy Information Administration, annual electricity consumption for the US is estimated at 3717 billion kWh in 2008 (see Energy Information Administration, 2005 at www.eia.doe.gov). Further, in the AEO reference case, which uses an annualized growth rate of 1.07%, this consumption is forecast to increase by 26% to 4696 billion kWh by 2030. The residential sector represents the largest portion (37.7% or 1403 billion kWh) of current and future electricity use.1 Within the residential sector, air conditioning (17%) is the largest individual use category, with lighting a close second (15%).2 Energy efficiency, which is defined as physical, long-lasting changes to buildings and equipment that results in decreased n
Corresponding author. Tel.: + 1 203 582 3489; fax: +1 203 582 8664. E-mail addresses: [email protected] (E.J. Brunner), [email protected] (P.S. Ford), [email protected] (M.A. McNulty), [email protected] (M.A. Thayer). 1 See EPRI (2009). The commercial sector comprises 36.3% (1350 billion kWh) of current usage, whereas the industrial sector makes up approximately 25.9% (964 billion kWh). 2 Note that the Energy Information Administration US Household Electricity Report estimates that lighting is responsible for 8.8% of household electricity use (www.eia.doe.gov). 0301-4215/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.enpol.2009.11.003
energy use while maintaining a constant level of service, has been identified as an important resource for satisfying a significant portion of current and future energy demand. In addition, energy efficiency is a growth industry that may have significant spillover benefits related to global climate change and US energy independence. For example, the Environmental Protection Agency (EPA) reported that in 1999 alone, Americans bought more than 100 million Energy Star products (see Banerjee and Solomon, 2003). Through 2006, Energy Star labeled products have saved 4.8 exajoules (EJs) of primary energy and $47 billion dollars in energy bills, and have avoided 82 teragrams (Tg) carbon equivalent (see Sanchez et al., 2008). Going forward, energy efficiency is realistically expected to achieve savings of 236 billion kWh relative to the AEO 2008 reference case in 2030 (141 billion kWh in 2020). This value represents an approximate 5% reduction in projected consumption. Furthermore, according to the Energy Power Research Institute (EPRI), summer peak savings associated with energy efficiency are projected to reach approximately 7% by 2030 (see EPRI, 2009). Similarly, according to the California Energy Commission (CEC), electricity energy savings associated with energy efficiency in California is projected to reach 9% by 2016 (see California Energy Commission, 2007). One of the core energy efficiency technologies is the use of compact fluorescent lights (CFLs) to replace incandescent bulbs, which are highly inefficient sources of light because about 90% of
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
the energy used is lost as heat. The 2009 EPRI report suggests that lighting improvements, particularly the use of CFLs, represent a large savings opportunity in the residential sector, especially in the short run (i.e., next 10 years). Awareness and usage of CFLs have increased dramatically over time as production costs have declined and availability has increased (the Energy Star website lists more than 100 manufacturers). According to a recent market effects study (Cadmus Group, Inc. et al., 2009), consumer awareness of CFLs by California households has increased from 58% in 1998 to 96% in 2008. In addition, the percentage of California households purchasing at least one CFL within the previous 18 months has increased from 17% to 77% over the last decade. Seventy-nine percent of California households currently use at least one CFL inside or outside the home. Wall and Crosbie (2009) report that within their study group the mean number of bulbs/household was 21.7 and the mean percent of electricity consumption for lighting was 6.5%. They estimate that replacement of incandescent bulbs with CFLs would reduce the electricity consumption associated with lighting by 50.9%. They also suggest that savings could be much greater for the average household because their particular study group was unusually pro-environmental in their attitudes, lighting choices, and behavior (Wall and Crosbie, 2009). However, there have been some concerns raised recently regarding the use of CFLs. In particular, disposal of CFLs, given the mercury content in the bulbs, has been raised as a potential problem. A second concern about CFLs was expressed in May 2008 when the California Public Utilities Commission (CPUC) released an update of the Database for Energy Efficient Resources (DEER). This document included ‘‘interactive effects’’ associated with the replacement of incandescent light bulbs with compact fluorescent bulbs (CFL) in the residential sector. Under this hypothesis, the reduced wattage of CFLs, relative to incandescent bulbs, generates less heat which, in turn, requires additional natural gas usage during the heating season. Thus, the interactive effect serves as an offset to the electricity savings associated with CFLs. The magnitude of the effect is significant in energy terms. The DEER provides an estimate of annual interactive effects for single family homes that ranges from 44 kBtu (0.44 therms) for an 11 W CFL to 92 kBtu (0.92 therms) for a 23 W CFL. This is an important issue because an estimated 290 million CFLs were sold in the US in 2007, with 55.6 million of those sold in California (see Cadmus Group, Inc. et al., 2009).3 The magnitude of the gas impact, or interactive effect, from CFLs has created significant controversy. On one hand, the existence of residential CFL interactive effects is consistent with theoretical engineering models of energy use. On the other hand, it is an empirical question as to whether the heat differential between the incandescent and compact fluorescent bulbs is large enough to actually trip a home’s thermostat and thereby increase heating requirements. Ford (2008), using principles accepted by the American Society of Heating, Refrigeration, and Air Conditioning Engineers (ASHRAE), has estimated that converting one incandescent bulb to a CFL (28 W saving) will result in a heat loss of 36.1 BTU/h.4 This translates into 0.0036% of the heating capacity of a normal heating system (100,000 BTU/h), which Ford
3 The importance of interactive effects, which offset potential gains from energy efficiency measures, may not be limited to CFLs. For example, the efficiency gains from LED lights and new electronic appliance standards could be questioned. The latter case is especially relevant since the California Energy Commission is expected to approve new television energy efficiency standards in November 2009. These standards would require televisions sold in California to meet stringent electricity use criteria by 2011 and even tougher criteria in 2013. 4 Note that the average person at rest emits a sensible heat gain of 245 BTU/h as per 2005 ASHRAE Fundamentals page 30.4 (see Ford, 2008).
1289
suggests is insufficient to be sensed by the thermostat. Thus, ASHRAE recommends against the inclusion of internal heat gains, particularly from occupants and lighting, when sizing residential heating systems. In this paper, we take a different approach to investigate the existence and potential magnitude of CFL interactive effects in residential situations. Specifically, we use billing analysis to test for the presence of any interactive effect. The approach used herein is similar to those of Metcalf and Hassett (1999) and Sebold and Fox (1985), who use billing data to examine the returns from energy efficiency investments. Our analysis is based on a comprehensive dataset that includes monthly household electricity and natural gas usage, the number of CFL bulbs installed, the installation date, and a set of household characteristics. Our results suggest that CFLs do indeed save electricity. However, we do not find any support for the hypothesis that CFLs cause increased usage of natural gas. The paper is organized as follows. In the next section we describe the data used in our analysis. Our empirical framework and results are presented in Sections 3 and 4, respectively. Conclusions and policy implications are offered in the final section.
2. Data In late November and early December 2008, San Diego Gas and Electric (SDG&E) compiled a dataset for Low Income Energy Efficiency (LIEE) participants from 2006 and 2007. The LIEE program delivers directly installed energy efficient solutions (i.e., no cost to the recipient) for income qualified households. The income guidelines vary by household size and are established annually by the California Public Utilities Commission.5 The SDG&E dataset included: quantity and type of energy efficiency measures installed; installation date; household income; household size; living area (square footage); and air conditioner ownership (central, room, or none). We merged these data with individual-level billing data obtained from SDG&E. The billing data consist of monthly observations on electricity and natural gas usage, billing days, and climate zone for all LIEE customers in the SDG&E service area and span the time period from January 2005 to December 2008. Because the billing data start in January 2005 and end in December 2008, they include at least 12 months of pre- and post-installation electricity and natural gas usage. Our analysis focuses on the impact that replacing traditional incandescent bulbs with CFLs has on energy usage. Thus, we restrict our attention to households that had CFLs installed only during the time frame and drop households that had multiple energy efficiency measures installed. Specifically, the dataset included approximately 2800 households that had only interior lighting installed. Among these 2800 households, approximately 1600 households had only 15 and 23 W CFLs installed. Of this latter group, the vast majority, 1219 households, had only 15 Watt bulbs installed. In the empirical analysis that follows, we restrict our attention to these 1219 households. To control for weather-related energy usage, we merged the billing data with heating and cooling degree hour data, compiled by SDG&E. Heating and cooling degree hours were developed by mapping individual weather station data to the three climate zones: maritime, coastal, and transitional. Both the heating and 5 The current categories for total yearly household income before deductions are 200% of the federal poverty level or by household size are the following: household size =1–2 ($30,500), household size= 3 ($35,800), household size= 4 ($43,200), household size= 5 ($50,600), household size= 6 ($58,000), and household size greater than or equal to 7 (add $7600/person to $58,000).
ARTICLE IN PRESS 1290
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
Table 1 Electricity summary statistics for full sample and high-use low-income households (1) Full sample
(2) Pre-CFL installation
(3) Post-CFL installation
Variable
Mean
St. dev.
Mean
St. dev.
Mean
St. dev.
All low income households Time varying characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
35.12 0.60 2.99 3.53 2.46 0.20 0.26 0.28 0.26
21.96 0.49 3.19 2.81 2.20 0.40 0.44 0.45 0.44
35.06 0.00 0.00 3.65 2.22 0.48 0.35 0.17 0.00
22.06 0.00 0.00 2.69 2.19 0.50 0.48 0.38 0.00
35.17 1.00 5.09 3.45 2.61 0.01 0.21 0.35 0.44
21.89 1.00 2.57 2.89 2.19 0.10 0.41 0.48 0.50
Household characteristics Square footage of home Household size Household income ($)
896 3.46 22,844
306 1.90 9069
896 3.46 22,844
306 1.90 9069
896 3.46 22,844
306 1.90 9069
High-use low income households Monthly characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
65.90 0.61 3.67 3.65 2.69 0.19 0.25 0.28 0.28
24.66 0.49 2.88 2.88 2.34 0.39 0.43 0.45 0.45
67.30 0.00 0.00 3.80 2.46 0.48 0.33 0.19 0.00
25.17 0.00 0.00 2.76 2.36 0.50 0.47 0.39 0.00
64.97 1.00 5.12 3.55 2.84 0.01 0.21 0.34 0.45
24.28 1.00 2.35 2.95 2.31 0.09 0.40 0.47 0.50
Household characteristics Square footage of home Household size Household income ($)
1095 4.17 25,995
408 2.00 10,726
1095 4.17 25,995
408 2.00 10,726
1095 4.17 25,995
408 2.00 10,726
cooling degree hours are based on a 651 degree set point assumption. Each participant’s home was mapped to one of the three climate zones using a zip code reference table provided by SDG&E. As an alternative, we also tried mapping to the three climate zones using the California Energy Commission’s climate zone to zip code list assuming that climate zone 7 was maritime, zone 10 was coastal, and zone 14 was transitional. In the empirical work that follows, we use the heating and cooling degree hours from SDG&E’s mapping.6 Table 1 presents the electricity usage variable definitions and summary statistics. Electricity usage is measured in thousands of Btus (kBtu). Summary statistics are presented for two specific sample groups: (1) all low-income households (upper panel) and (2) low-income households that have average electricity consumption that equals or exceeds 48.8 kBtu/day (14.3 kWh/day; lower panel). This latter group has average electricity consumption of approximately 65.9 kBtu/day (19.3 kWh/day), which is consistent with overall average electricity usage in the San Diego Gas and Electric service territory (see McNulty et al., 2006). For each group, Column 1 contains electricity usage summary statistics for the entire time period of analysis. In Columns 2 and 3 we provide the same information for the pre-installation and post-installation periods, respectively. Note that household characteristics do not vary from the pre- to the post-period. As expected, low-income households in the full sample (upper panel) have smaller homes and incomes and use less electricity per day than representative San Diego households. However,
6 The results reported later in the paper are not sensitive to which weather mapping we use. Specifically, the core results reported in Tables 3–6 are robust to using the alternative heating and cooling degree hours mapping.
family size exceeds the comparable value in San Diego (2.80 people/household in McNulty et al., 2006). The high-use sample was selected to more closely match the SDG&E service territory customer base. Therefore, the average electricity use is closely aligned with the surrounding population. However, home size, income, and household size are still quite different.7 In terms of the variable of interest, post-installation electricity usage per day declines from the pre-installation period for the high-use group, in spite of the increase in average cooling degree hours, an indicator of air conditioning demand. The full sample shows no significant change in electricity use between study periods. Table 2 presents gas usage summary statistics for both the full sample of LIEE customers (upper panel) and LIEE customers that are in the high-use group (lower panel). High use is defined as an average of 68 kBtu per day (0.68 therms per day) or more, which translates into an overall sample average of 106 kBtu/day (1.06 therms/day). This latter value is consistent with the average natural gas use in the SDG&E service territory (see McNulty et al., 2006). As in Table 1, Column 1 contains gas usage summary statistics for the entire time period of analysis, while Columns 2 and 3 provide the same information for the pre-installation and post-installation periods, respectively. Note that the gas samples differ from the electricity samples in two ways. First, for the full sample, there are fewer low-income
7 The following values were taken from the McNulty, Murdoch, and Thayer study: average daily electricity use= 65.0 kBtu (19.3 kWh), average daily gas use= 1070 kBtu (1.07 therms), average home size= 1725 ft2, and average household size= 2.80 persons. In addition, because only low-income households qualify for the program being studied herein, we know average income is below the comparable average service territory average.
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
1291
Table 2 Natural gas summary statistics for full sample and high-use low-income households (1) Full sample
(2) Pre-CFL installation
(3) Post CFL installation
Variable
Mean
St. dev.
Mean
St. dev.
Mean
St. dev.
All low income households Time varying characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
73.99 0.63 3.15 3.50 2.38 0.20 0.26 0.28 0.26
58.16 0.48 3.27 2.79 2.14 0.40 0.44 0.45 0.44
74.65 0.00 0.00 3.59 2.08 0.52 0.33 0.15 0.00
57.84 0.00 0.00 2.64 2.08 0.50 0.47 0.36 0.00
73.61 1.00 5.19 3.44 2.55 0.01 0.22 0.35 0.41
58.34 1.00 2.94 2.87 2.15 0.11 0.42 0.48 0.49
Household characteristics Square footage of home Household size Household income ($)
931 3.64 23,226
318 1.95 9549
931 3.64 $23,226
318 1.95 9549
931 3.64 23,226
318 1.95 9549
High-use low income households Time varying characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
106.15 0.64 3.28 3.52 2.41 0.20 0.26 0.28 0.26
61.22 0.48 3.43 2.79 2.15 0.40 0.44 0.45 0.44
109.02 0.00 0.00 3.59 2.12 0.53 0.33 0.14 0.00
59.71 0.00 0.00 2.63 2.09 0.50 0.47 0.35 0.00
104.88 1.00 5.29 3.48 2.57 0.01 0.23 0.36 0.41
62.00 1.00 3.07 2.87 2.16 0.11 0.42 0.48 0.49
Household characteristics Square footage of home Household size Household income ($)
1044 4.18 $24,798
346 2.05 $10,742
1044 4.18 $24,798
346 2.05 $10,742
1044 4.18 $24,798
346 2.05 10,742
gas customers than electricity customers, due to the fact that some homes do not use natural gas. This is especially apparent on comparison of the upper panel of Table 1 to the upper panel of Table 2. Specifically, the full low-income sample for the gas usage model consists of 870 households, whereas the full low-income sample for the electricity usage model consists of 1219 households. Second, the high-use gas group is larger (lower panel, Table 3) than the electricity high-use group (lower panel, Table 1). This occurs because the gas usage threshold is somewhat lower to qualify for high-use designation. Thus, our high-use gas group is about twice as large as the corresponding electricity group. From the perspective of this study, the most interesting result shown in Table 2 is that gas usage does not seem to be affected by the installation of CFLs, as pre-installation and post-installation values are nearly identical for each sample. As with the electricity samples (Table 1), both the full sample of low-income households (upper panel) and the high-use sample have smaller homes and incomes but larger household size than representative San Diego households.
3. Empirical specification To examine the impact of replacing traditional incandescent bulbs with CFLs on electricity usage, we estimate the following model: Eit ¼ ai þ b1 CFLit þ b2 HDHit þ b3 CDHit þ Yeart g1 þMontht g2 þ ðai HDHit Þd1 þ ðai CDHit Þd2 þ eit
ð1Þ
where Eit is the electricity usage, measured in kBtu per day, for household i in time period t; ai is a vector of household fixed
effects; CFLit is the number of compact florescent light bulbs installed in household i in period t (CFLit =0 in the pre-installation period); HDHit the number of heating degree hours; CDHit the number of cooling degree hours; Yeart is a vector of year fixed effects; Montht is a vector of month fixed effects; ai HDHit denotes the interaction of the number of heating degree hours for household i in period t with the household-specific fixed effects; ai CDHit denotes the interaction of the number of cooling degree hours with the household-specific fixed effects; and eit is a random disturbance term. The coefficient of primary interest is b1. For the electricity usage model, b1 should be negative because replacing traditional incandescent bulbs with CFLs should reduce electricity usage. In terms of interpretation, b1 measures the average daily electricity savings associated with the installation of one additional CFL. Note that the inclusion of individual-specific fixed effects in Eq. (1) implies that we are using only ‘‘within household’’ variation in energy consumption to identify the model. Thus, any factor that influences energy consumption but does not vary over time (e.g. square feet of living space, number of occupants in household, etc.) is captured by the individual fixed effects. Consequently, our model controls for any observable or unobservable factors that influence energy consumption but do not vary over time. Furthermore, the inclusion of the interaction terms ai HDHit and ai CDHit adjusts for each household’s weathersensitive energy usage (i.e., individual-specific weather-related energy consumption patterns). While Eq. (1) provides a useful way of modeling electricity savings due to the installment of CFLs, it may be less useful for modeling the interactive effects that the installation of CFLs may have on gas consumption. Specifically, the interactive effect that
ARTICLE IN PRESS 1292
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
Table 3 Estimated model results—electricity usage/day Specification
Full sample
High use
Number of CFLs
0.127nn (0.0637) 0.392nnn (0.115) 0.438nnn (0.111) 0.125 (0.324) 0.927nn (0.436) 1.537nnn (0.575) 2.385nnn (0.201) 4.015nnn (0.273) 4.999nnn (0.353) 4.493nnn (0.449) 3.424nnn (0.554) 1.956nnn (0.630) 0.303 (0.678) 0.328 (0.656) 2.613nnn (0.558) 3.597nnn (0.467) 1.904nnn (0.334)
0.423nn (0.178) 2.708nnn (0.258) 0.148 (0.307) 0.613 (1.120) 1.840 (1.442) 2.143 (2.080) 3.692nnn (0.673) 6.510nnn (1.014) 8.419nnn (1.330) 7.536nnn (1.669) 5.801nnn (2.008) 1.468 (2.290) 2.372 (2.417) 0.970 (2.293) 4.766nn (1.935) 7.206nnn (1.611) 3.333nnn (1.133)
38592 1219 0.85
7310 240 0.72
Cooling degree hours (1000s) Heating degree hours (1000s) 2006 2007 2008 February March April May June July August September October November December Observations Number of households R2
Notes: Dependent variable is in kBtu/day. Column 1 presents basline electricity usage estimates for full sample of LIEE households with 15 W CFL installation. Column 2 presents electricity usage estimates for sample of high-usage LIEE households with 15 W CFL installation, where high usage is defined as electricity usage above 48.8 kBtu/day (14.3 kWh/day), which corresponds to an average usage of 65.9 kBtu/day (19.3 kWh/day). All models include individual-specific heating degree hour slopes and individual-specific cooling degree hour slopes. Robust clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnnsignificant at 1%.
CFLs have on gas consumption arises due to the change in ambient air temperature associated with replacing a traditional incandescent bulb with a compact fluorescent bulb.8 The resulting change in ambient air temperature will depend on several factors, such as the wattage of the traditional incandescent bulb that is being replaced, the flow of air through a home, and perhaps most importantly, volume (size) of the space within which the light bulbs are housed. To see this, note that if we consider a room that does not leak any heat, we can express the change in room temperature associated with use of a traditional incandescent bulb of a given wattage as
DT ¼
Watts t Vdc
ð2Þ
where DT denotes the change in temperature; Watts is the wattage of a traditional incandescent bulb; t the time measured in 8 Note that there is also a potential interactive effect for electricity. Specifically, replacing incandescent lights with CFLs may result in a lower cooling season temperature and hence a household may utilize its air conditioning system less often. The dataset we have assembled for this study is not sufficiently comprehensive to estimate this effect.
seconds; V the volume of the room measured in m3 (meters cubed); d the density of air measured in kg/m3 (kilograms per meter cubed); and c the specific heat capacity of air measured in kJ/(kg 1C) (kilojoules per kilograms degree Celsius). Thus, for any given values of d and c, the change in room temperature varies inversely with the size of the room and the larger the room, the smaller the change in ambient air temperature. Consequently, the impact that replacing traditional incandescent bulbs with CFLs has on ambient air temperature should be larger in smaller rooms and smaller in larger rooms. This suggests that the relationship between gas usage and installation of CFLs is not linear as specified in Eq. (1). Rather, the impact CFLs have on gas usage depends critically on the density of CFLs, measured for example, as CFLs per square feet. The discussion above suggests that an ideal way to measure the potential impact CFLs have on gas usage would be to construct some measure of the number of CFLs per volume (height width length) of heating area. Unfortunately, to our knowledge, no such data exist. However, we do have information on the square footage of homes, a variable that should be highly correlated with the volume of heating area. Thus, to examine the impact of replacing traditional incandescent bulbs with CFLs on gas usage, we estimate models of the following form: Git ¼ ai þ b1 CFLit =SQFTi þ b2 HDHit þ b3 CDHit þ Yeart g1 þ Montht g2 þ ðai HDHit Þd1 þ ðai CDHit Þd2 þ eit ð3Þ where Git is the gas usage, measured in kBtu per day, for household i in time period t, SQFTi denotes the square footage of housing unit i, and all other variables are as defined in Eq. (1). In Eq. (3), b1 measures the average effect on gas consumption of installing one additional CFL per square foot of living space. If interaction effects are present, b1 should be positive and significantly different from zero. Furthermore, note that the marginal effect of installing one additional CFL has on gas usage can be expressed as
@ Usage 1 ¼ b1 @ CFL SQFT
ð4Þ
which depends on the square footage of a home—the larger the home, the smaller the marginal effect of an additional CFL on gas usage.
4. Empirical results 4.1. Electricity usage results Results based on the estimation of Eq. (1) are presented in Table 3. In all regressions, the dependent variable equals kBtu per day. In the first column, we present electricity usage estimates for the full sample of LIEE households that have had only 15 W CFLs installed. In the second column, we present electricity usage estimates for the high-use sample of LIEE households that have had only 15 W CFLs installed. All models also include individualspecific heating degree hour slopes and individual-specific cooling degree hour slopes (see Eq. (1) above). For brevity, the approximately 3650 (full sample) or 720 (high-use sample) coefficients are not presented, but are available on request. All standard errors reported in Table 3 and subsequent tables are clustered at the individual level to allow for within-individual autocorrelation of the disturbance term. The indicators n, nn, and nnn imply the variable is significant at the 10% level, 5% level, or 1% level, respectively.
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
There are several noteworthy results in Table 3. First, a high proportion of the variation in electricity usage is explained by the independent variables as R2 ranges from 0.72 (high-use sample) to 0.85 (full sample). Second, the month and year dummy variables, which are interpreted as variations from the relevant omitted category (2005 for year, January for month), perform as expected. Specifically, there is some evidence that electricity usage is increasing over time, especially for the full sample. In addition, the monthly effects suggest that, all else being constant, electricity use/day in January is similar to that in August–September period (July–September for high-use sample), whereas during all the other months, daily electricity usage is significantly below this level. Third, heating and cooling degree hours perform as expected (e.g., more cooling degree hours imply more cooling demand, which is positively related to electricity use), even when the model includes individual-specific heating and cooling degree hour slopes. It should be noted that if the individual-specific heating and cooling degree hour slopes are omitted from the model, then the heating and cooling degree hours variables continue to perform as expected. The final important result pertains to our focus variable, the number of compact fluorescent (CFL) bulbs installed in the home. As is indicated in Table 3, the coefficient on the number of CFLs is negatively related to daily electricity usage and is significant at the 5% level in each of the estimated models. The calculation of the impact of an additional CFL is straightforward. The coefficient multiplied by 365 yields the annual savings from an additional bulb. For the full sample of low-income households in Table 3, the savings value attributable to another CFL is approximately 46.4 kBtu/year (13.6 kWh/year). This value is very close to those reported in a recent study by West Hill Energy and Computing, Inc. et al. (2007).9 The results for the full sample can also be compared to those of our high-use sample in Table 3. The estimated coefficient on number of CFLs for this latter group suggests significantly larger savings than those for the full sample. Specifically, the electricity savings attributable to another CFL for the high-use group are approximately 155 kBtu/year (45.3 kWh/year). If a high-use household installed the average number of CFLs (5), then annual electricity savings would be 773 kBtu (226.5 kWh). Dividing this value by annual usage of 24,565 kBtu (from Table 1 average daily usage of 67.30 kBtu times 365) produces annual electricity savings as a percentage of total usage of 3.2%. This percentage reduction in overall electricity usage is very close to the estimated value in Wall and Crosbie (2009), which was 3.31% (50.9% reduction in the 6.5% share of total electricity use). The magnitude of electricity savings also suggests that if each household were to replace five incandescent bulbs with CFLs, California would be approximately one-third the way to achieving the projected 9% electricity energy savings by 2016 (see California Energy Commission, 2007). 4.2. Gas usage results Having established that the replacement of traditional incandescent bulbs with CFLs results in relatively large and statistically significant reductions in electricity usage, we now turn to examine the impact of CFLs on gas usage. Table 4 presents gas usage results based on the estimation of Eq. (3). In all 9 In this study, the estimated coefficient on CFLs produced annual savings of 37.5 kBtu/year (11 kWh/year) with a confidence interval of 20.5–54.6 kBtu/year (6–16 kWh/year). The authors stated that their regression results were likely biased downward due to CFL attrition and purchase outside the program, so they recommended the high end of the confidence interval (see West Hill Energy and Computing and Inc. et al., 2007)
1293
Table 4 Estimated model results—gas usage/day Specification
Full sample
High use
Number of CFLs/ft2
0.184 (160.7) 3.034nnn (0.165) 5.442nnn (0.247) 3.525nnn (0.957) 2.485n (1.298) 2.415 (1.547) 0.453 (0.815) 4.570nnn (0.868) 10.06nnn (1.082) 8.884nnn (1.232) 6.378nnn (1.508) 8.178nnn (1.712) 10.24nnn (1.842) 13.01nnn (1.777) 13.52nnn (1.617) 11.81nnn (1.368) 9.027nnn (0.992)
148.1 (278.6) 3.386nnn (0.287) 4.822nnn (0.417) 4.994nnn (1.659) 3.384 (2.219) 3.635 (2.446) 0.716 (1.461) 7.349nnn (1.463) 15.41nnn (1.866) 13.82nnn (2.105) 9.658nnn (2.585) 11.47nnn (2.922) 14.02nnn (3.152) 18.24nnn (3.023) 19.63nnn (2.779) 16.96nnn (2.365) 12.54nnn (1.714)
28190 870 0.859
14581 435 0.797
Cooling degree hours (1000s) Heating degree hours (1000s) 2006 2007 2008 February March April May June July August September October November December Observations Number of households R2
Notes: Dependent variable is in kBtu/day. Column 1 presents baseline gas usage estimates for full sample of LIEE households with 15 W CFL installation. Column 2 presents gas usage estimates for sample of high-use LIEE households with 15 W CFL installation, where high usage is defined as gas usage above 68 kBtu /day (0.68 therms/day), which corresponds to an average usage of 106 kBtu/day (1.06 therms/day). All models include individual-specific heating degree hour slopes and individual-specific cooling degree hour slopes. Robust clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnn significant at 1%.
regressions, the dependent variable is natural gas usage in kBtu per day. Similar to Table 3, in the first column we present gas usage estimates for the full sample of LIEE households that had only 15 W CFLs installed. In the second column, we present estimates for the high-use sample of LIEE households. All models (once again) include individual-specific heating and cooling degree hour slopes. Similar to the electricity usage results, the results reported in Table 4 indicate that a high proportion of the variation in gas usage is explained by the model as indicated by an R2 of 0.86 for the full sample and an R2 of 0.80 in the high-use sample. The month and year dummy variables, which are interpreted as variations from the relevant omitted category (2005 for year, January for month), perform as expected. The annual trends indicate that gas consumption in 2006–2008 is significantly smaller than the usage recorded in 2005, even though there has been no substantive change in heating degree hours. This may reflect a price effect. The months with relatively high gas usage are January and February. Minimum gas usage occurs in the September–November period. The heating and cooling degree hours variables perform as expected, suggesting that individuals
ARTICLE IN PRESS 1294
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
Table 5 Robustness checks: full sample of LIEE customers Panel
Specification
Baseline
Individual-specific heating degree hour slopes only
Dropping month of installment
Dropping homes with 1700 ft2 of living area or more
(1)
Number of CFLs/ft2
0.184 (160.7)
45.8 (152.1)
34.7 (162.5)
26.1 (163.3)
(2)
Number of CFLs/ft2
74.1 (164.6) 20.9 (25.1)
24.5 (154.1) 20.2 (24.3)
102.3 (165.6) 19.5 (28.6)
50.4 (167.7) 21.6 (25.4)
CFLs/ft2 heating degree hours
(3)
Number of CFLs
0.191 (0.174)
0.229 (0.165)
0.151 (0.18)
0.258 (0.187)
(4)
Number of CFLs
0.113 (0.18) 0.021 (0.0342)
0.154 (0.173) 0.0203 (0.0331)
0.0953 (0.182) 0.0153 (0.0382)
0.172 (0.198) 0.0232 (0.0371)
CFLs heating degree hours
Notes: Table presents various specification checks for the full sample of LIEE customers with 15 W CFL installation. Dependent variable in all specifications is in therms/day. Panels 1 and 2 contain estimates when the number of CFLs/ft2 is the key explanatory variable. Panels 3 and 4 contain estimates when the number of CFLs is the key explanatory variable. Robust, clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnnsignificant at 1%.
in our low income sample modify their gas usage in response to variations in weather. Turning to the variable of interest, the estimated coefficient on the number of CFLs per square feet is positive, as predicted if interactive effects are present, but never statistically different from zero at any commonly used significance level. For the results based on the full sample, the point estimate on the number of CFLs per square feet implies that each additional CFL per square foot increases natural gas usage by approximately 0.2 kBtu/day (0.002 therms/day). Evaluated at the mean square footage of homes in the sample, this estimate implies an interactive effect of only 0.07 kBtu (0.0007 therms) per year.10 Furthermore, the magnitude of this effect declines with square footage, so that if the square footage of a representative San Diego home were to be used (1725 ft2) in the calculation, the magnitude of the interactive effect would be approximately half as large. For the high-use sample, the estimated coefficient on the number of CFLs is significantly larger in magnitude but, once again, statistically insignificant. Evaluated at the mean square footage of homes in the sample, the estimate coefficient implies an interactive effect of approximately 52 kBtu (0.52 therms) per year. If the square footage of a representative San Diego home were to be used in the calculation, the magnitude of the interactive effect would fall to approximately 31 kBtu (0.31 therms) per year. 4.3. Robustness checks The results reported in Table 4 suggest that interactive effects for CFLs are not present in a residential setting. Because this result is counter to the DEER Team position, we considered several other models of gas usage, and various alternative specifications, to determine whether or not the results reported in Table 4 are robust. Specifically, we examined whether our results were robust to the following alternative specification checks: (1) using only individual-specific heating degree hour slopes rather than both individual-specific heating and cooling degree hour slopes; (2) eliminating the month in which the installment of CFLs occurred; (3) eliminating homes with a living area of 1700 ft2 or more; (4) including the number of CFLs per square feet interacted with the number of heating degree hours as an explanatory 10
0.07 ft2 =(0.184 365)/931 ft2 for full sample.
variable; (5) replacing the number of CFLs per square feet with simply the number of CFLs as in the electricity usage models; and (6) including the number of CFLs interacted with the number of heating degree hours as the variable of interest. We consider models where we include only individual-specific heating degree hour slopes to examine whether our results are sensitive to how we model individual-specific weather-related energy consumption patterns. We drop the month in which the CFLs were installed to reduce any measurement error associated with uncertainty concerning the exact date of installation. We drop homes with a living area of 1700 ft2 or more (top 2.5% of homes in our sample) to test whether interactive effects are more likely to exist in smaller homes. Finally, we estimate models that include an interaction term between the number of CFLs per square feet (or the number of CFLs) and the number of heating degree hours to examine whether the interactive effect CFLs may have on gas usage varies with the number of heating degree days. The results for these robustness checks for the full sample of LIEE customers are presented in Table 5. In the interest of brevity, we report only the estimated coefficients on the number of CFLs per square feet (or number of CFLs) and the interaction term between the number of CFLs per square feet (or number of CFLs) and the number of heating degree hours. We note however, that all models include the same control variables listed in Table 4 and, with the exception of Column 2, all models include individualspecific heating degree hour and individual-specific cooling degree hour slopes. The first two panels of Table 5 report estimates when the key variable of interest is the number of CFLs per square foot of living area. Panels 3 and 4 present estimates when the key variable of interest is the number of CFLs. For comparison purposes, Column 1 of Panel 1 reports the estimated coefficient on the number of CFLs given in Column 1 of Table 4. A brief inspection of Table 5 reveals that none of the estimated coefficients reported in the table is statistically different from zero. The estimated coefficient on the number of CFLs per square feet becomes larger in magnitude when we drop the month during which the CFLs were installed, although it remains statistically insignificant at any conventional significance level. In contrast, the estimated coefficient on the number of CFLs per square feet turns negative when we utilize only individualspecific heating degree hour slopes or drop homes with 1700 ft2 or more of living area. When we replace the number of CFLs per
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
1295
Table 6 Robustness checks: high-use sample of LIEE customers Panel
Specification
Baseline
Individual-specific heating degree hour Slopes Only
Dropping month of installment
Dropping homes with 1700 ft2 of living area or more
(1)
Number of CFLs/ft2
148.1 (278.6)
71.7 (267.5)
211.6 (279.9)
119.6 (283.8)
(2)
Number of CFLs/ft2
170.4 (284.5) 6.37 (45.2)
108.7 (269.1) 10.8 (43.8)
212.9 (276.7) 0.381 (53.4)
133.6 (292.0) 4.01 (45.8)
CFLs/ft2 heating degree hours
(3)
Number of CFLs
0.209 (0.292)
0.272 (0.279)
0.131 (0.298)
0.245 (0.317)
(4)
Number of CFLs
0.179 (0.315) 0.00801 (0.0571)
0.229 (0.303) 0.0115 (0.0552)
0.141 (0.314) 0.00267 (0.0646)
0.252 (0.342) 0.00176 (0.0622)
CFLs heating degree hours
Notes: Table presents various specification checks for the high-use sample of LIEE customers with 15 W CFL installation. Dependent variable in all specifications is in therms/day. Panels 1 and 2 contain estimates when the number of CFLs/ft2 is the key explanatory variable. Panels 3 and 4 contain estimates when the number of CFLs is the key explanatory variable. Robust, clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnnsignificant at 1%.
square feet with the number of CFLs, the estimated coefficient on the number of CFLs is consistently negative (incorrect sign) and never close to being statistically significant. Table 6 presents our robustness checks for the sample of highuse LIEE customers. The results reported in Table 6 tend to mirror the results reported in Table 5. Specifically, similar to Table 5, none of the estimated coefficients reported in the Table 6 is statistically different from zero. Thus, across a wide array of specifications, our results consistently suggest that interactive effects for CFLs are not present in a residential setting.
5. Concluding remarks The primary results of our research effort are twofold. First, there is strong statistical evidence that replacing incandescent lights with compact fluorescent lights in a residential setting generates significant electricity savings. The coefficient that relates the installation of CFLs to electricity usage is both negative and significantly different from zero at the 5% level. The magnitude of the effect (46.4 kBtu/year or 13.6 kWh/year) for the full sample of low-income households is consistent with a recent study authored by West Hill Energy and Computing and Inc. et al. (2007). In addition, the magnitude of the effect (155 kBtu/year of 45.3 kWh/year) for our high-use sample, which is designed to be more closely representative of SDG&E service territory households, is consistent with (although slightly greater than) estimated savings in the DEER documentation and with a recent study by Wall and Crosbie (2009). The second significant result is that there is no statistical evidence of any interactive effect of CFLs on the usage of natural gas. In the full sample of low-income households, the estimated interactive effect is clearly indistinguishable from zero, both in the statistical sense and in the magnitude of the coefficient. In the high-use sample, the estimated effect is also not significantly different from zero at any conventional significance level. In terms of the magnitude of the effect (ignoring statistical significance), the estimated coefficient converts to an interactive effect that ranges from 0.07 kBtu (0.0007 therms) per year for the full sample to 52 kBtu (0.52 therms) per year for the high-use group. This effect diminishes with square footage of living area so that the range for an average sized home (approximately 1725 ft2) is
0.038 kBtu for the full sample to 31 kBtu per year for the high-use group. Our finding that interactive effects of CFLs on natural gas usage are indistinguishable from zero holds across a wide array of specifications. That is, the results are invariant to various possible specifications of the relationship between CFLs and gas usage. The primary policy implications of these findings for energy efficiency programs are the following. First, our results strongly suggest that programs designed to replace incandescent light bulbs with CFLs can generate significant electricity savings in the residential sector. Second, the lack of any statistically measurable interactive effect suggests that CFLs have little to no effect on natural gas consumption. This, in turn, suggests that the replacement of incandescent bulbs with CFLs does not lead to a significant offset in the energy savings as suggested by the Database for Energy Efficient Resources (DEER). The results of this study are subject to two caveats. First, the sample sizes are smaller than we prefer. For example, the high-use sample of electricity usage is only 240 households. Second, this study considers only low-income households, and they might not be representative of typical households in the SDG&E service territory that would participate in the energy efficiency programs. Therefore, it would be interesting to test for the existence and potential magnitude of interactive effects using either a larger sample of low-income households and/or a general sample of the overall population in order to replicate the results of this study.
References Banerjee, A., Solomon, B.D., 2003. Eco-labeling for energy efficiency and sustainability: a meta-evaluation of U.S. programs. Energy Policy 31, 109–123. Cadmus Group, Inc., KEMA, Itron, Inc., Nexus Market Research, and Goett Consulting. 2009. Compact fluorescent lamps market effects: interim report—draft. Prepared for the California Public Utilities Commission, January. California Energy Commission. 2007. Statewide energy efficiency potential estimates and target for California utilities: draft staff report. CEC-200-2007019-SD, August. Electric Power Research Institute (EPRI). 2009. Assessment of achievable potential from energy efficiency and demand response programs in the U.S. 2010–2030. EPRI Technical Report, January. Energy Information Administration. 2005. U.S. household electricity report, www. eia.doe.gov, July. Ford, P., 2008. Interactive effects of CFL lighting in residential applications. Engineering Opinion to Sempra Utilities October.
ARTICLE IN PRESS 1296
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
McNulty, M., J. Murdoch, and M. Thayer, The home energy comparison tool. Prepared for San Diego Gas and Electric Company, Final Report, June 2006. Metcalf, G.E., Hassett, K.A., 1999. Measuring the energy savings from home improvement investments: evidence from monthly billing data. The Review of Economics and Statistics 81 (3), 516–528. Sanchez, M.C., Brown, R.E., Webber, C., Homan, G.K., 2008. Savings estimates for the United States Environmental Protection Agency’s ENERGY STAR Voluntary Labeling Program. Energy Policy 36, 2098–2108.
Sebold, F.D., Fox, E., 1985. Realized savings from residential conservation activity. The Energy Journal 6 (2), 73–88. Wall, R., Crosbie, T., 2009. Potential for reducing electricity demand for lighting in households: an exploratory socio-technical study. Energy Policy 37 (3), 1021–1031. West Hill Energy and Computing, Inc., et al., Impact evaluation of the 2005 California Low Income Energy Efficiency Program. Final Report, Prepared for Southern California Edison Company, December 2007.
Contents lists available at ScienceDirect
Energy Policy journal homepage: www.elsevier.com/locate/enpol
Compact fluorescent lighting and residential natural gas consumption: Testing for interactive effects Eric J. Brunner a,n, Peter S. Ford b, Mark A. McNulty c, Mark A. Thayer d a
Department of Economics, Quinnipiac University, 275 Mount Carmel Avenue, Hamden, CT 06516, USA San Diego Gas and Electric Company, 8335 Century Park Court, San Diego, CA 92123, USA c Mark McNulty and Associates, 4654 Mayapan Drive, La Mesa, CA 91941, USA d Department of Economics, San Diego State University, San Diego, CA 92182, USA b
a r t i c l e in f o
a b s t r a c t
Article history: Received 15 March 2009 Accepted 3 November 2009 Available online 9 December 2009
Replacing incandescent light bulbs with compact fluorescents (CFLs) has traditionally been seen as a cost effective means of promoting energy conservation. Recently, however, the magnitude of energy savings associated with CFLs has been called into question. Specifically, recent findings suggest an ‘‘interactive effect’’ associated with the replacement of incandescent light bulbs with CFLs in the residential sector. In this scenario, the reduced wattage of CFLs, relative to incandescent bulbs, generates less heat, which in turn, requires additional natural gas usage during the heating season. Engineering studies suggest the magnitude of the effect is significant in energy terms, which implies that the energy savings associated with CFLs may be significantly overstated. In this paper, we use billing analysis to test for the presence of interactive effects. Our analysis is based on a comprehensive dataset that includes monthly household electricity and natural gas usage, the number of CFL bulbs installed, the installation date, and a set of household characteristics. Our results suggest that CFLs do indeed save electricity. However, we do not find any support for the hypothesis that CFLs cause increased usage of natural gas. & 2009 Elsevier Ltd. All rights reserved.
Keywords: Energy efficiency Residential energy consumption
1. Introduction In the most recent Annual Energy Outlook (AEO 2008) from the US Energy Information Administration, annual electricity consumption for the US is estimated at 3717 billion kWh in 2008 (see Energy Information Administration, 2005 at www.eia.doe.gov). Further, in the AEO reference case, which uses an annualized growth rate of 1.07%, this consumption is forecast to increase by 26% to 4696 billion kWh by 2030. The residential sector represents the largest portion (37.7% or 1403 billion kWh) of current and future electricity use.1 Within the residential sector, air conditioning (17%) is the largest individual use category, with lighting a close second (15%).2 Energy efficiency, which is defined as physical, long-lasting changes to buildings and equipment that results in decreased n
Corresponding author. Tel.: + 1 203 582 3489; fax: +1 203 582 8664. E-mail addresses: [email protected] (E.J. Brunner), [email protected] (P.S. Ford), [email protected] (M.A. McNulty), [email protected] (M.A. Thayer). 1 See EPRI (2009). The commercial sector comprises 36.3% (1350 billion kWh) of current usage, whereas the industrial sector makes up approximately 25.9% (964 billion kWh). 2 Note that the Energy Information Administration US Household Electricity Report estimates that lighting is responsible for 8.8% of household electricity use (www.eia.doe.gov). 0301-4215/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.enpol.2009.11.003
energy use while maintaining a constant level of service, has been identified as an important resource for satisfying a significant portion of current and future energy demand. In addition, energy efficiency is a growth industry that may have significant spillover benefits related to global climate change and US energy independence. For example, the Environmental Protection Agency (EPA) reported that in 1999 alone, Americans bought more than 100 million Energy Star products (see Banerjee and Solomon, 2003). Through 2006, Energy Star labeled products have saved 4.8 exajoules (EJs) of primary energy and $47 billion dollars in energy bills, and have avoided 82 teragrams (Tg) carbon equivalent (see Sanchez et al., 2008). Going forward, energy efficiency is realistically expected to achieve savings of 236 billion kWh relative to the AEO 2008 reference case in 2030 (141 billion kWh in 2020). This value represents an approximate 5% reduction in projected consumption. Furthermore, according to the Energy Power Research Institute (EPRI), summer peak savings associated with energy efficiency are projected to reach approximately 7% by 2030 (see EPRI, 2009). Similarly, according to the California Energy Commission (CEC), electricity energy savings associated with energy efficiency in California is projected to reach 9% by 2016 (see California Energy Commission, 2007). One of the core energy efficiency technologies is the use of compact fluorescent lights (CFLs) to replace incandescent bulbs, which are highly inefficient sources of light because about 90% of
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
the energy used is lost as heat. The 2009 EPRI report suggests that lighting improvements, particularly the use of CFLs, represent a large savings opportunity in the residential sector, especially in the short run (i.e., next 10 years). Awareness and usage of CFLs have increased dramatically over time as production costs have declined and availability has increased (the Energy Star website lists more than 100 manufacturers). According to a recent market effects study (Cadmus Group, Inc. et al., 2009), consumer awareness of CFLs by California households has increased from 58% in 1998 to 96% in 2008. In addition, the percentage of California households purchasing at least one CFL within the previous 18 months has increased from 17% to 77% over the last decade. Seventy-nine percent of California households currently use at least one CFL inside or outside the home. Wall and Crosbie (2009) report that within their study group the mean number of bulbs/household was 21.7 and the mean percent of electricity consumption for lighting was 6.5%. They estimate that replacement of incandescent bulbs with CFLs would reduce the electricity consumption associated with lighting by 50.9%. They also suggest that savings could be much greater for the average household because their particular study group was unusually pro-environmental in their attitudes, lighting choices, and behavior (Wall and Crosbie, 2009). However, there have been some concerns raised recently regarding the use of CFLs. In particular, disposal of CFLs, given the mercury content in the bulbs, has been raised as a potential problem. A second concern about CFLs was expressed in May 2008 when the California Public Utilities Commission (CPUC) released an update of the Database for Energy Efficient Resources (DEER). This document included ‘‘interactive effects’’ associated with the replacement of incandescent light bulbs with compact fluorescent bulbs (CFL) in the residential sector. Under this hypothesis, the reduced wattage of CFLs, relative to incandescent bulbs, generates less heat which, in turn, requires additional natural gas usage during the heating season. Thus, the interactive effect serves as an offset to the electricity savings associated with CFLs. The magnitude of the effect is significant in energy terms. The DEER provides an estimate of annual interactive effects for single family homes that ranges from 44 kBtu (0.44 therms) for an 11 W CFL to 92 kBtu (0.92 therms) for a 23 W CFL. This is an important issue because an estimated 290 million CFLs were sold in the US in 2007, with 55.6 million of those sold in California (see Cadmus Group, Inc. et al., 2009).3 The magnitude of the gas impact, or interactive effect, from CFLs has created significant controversy. On one hand, the existence of residential CFL interactive effects is consistent with theoretical engineering models of energy use. On the other hand, it is an empirical question as to whether the heat differential between the incandescent and compact fluorescent bulbs is large enough to actually trip a home’s thermostat and thereby increase heating requirements. Ford (2008), using principles accepted by the American Society of Heating, Refrigeration, and Air Conditioning Engineers (ASHRAE), has estimated that converting one incandescent bulb to a CFL (28 W saving) will result in a heat loss of 36.1 BTU/h.4 This translates into 0.0036% of the heating capacity of a normal heating system (100,000 BTU/h), which Ford
3 The importance of interactive effects, which offset potential gains from energy efficiency measures, may not be limited to CFLs. For example, the efficiency gains from LED lights and new electronic appliance standards could be questioned. The latter case is especially relevant since the California Energy Commission is expected to approve new television energy efficiency standards in November 2009. These standards would require televisions sold in California to meet stringent electricity use criteria by 2011 and even tougher criteria in 2013. 4 Note that the average person at rest emits a sensible heat gain of 245 BTU/h as per 2005 ASHRAE Fundamentals page 30.4 (see Ford, 2008).
1289
suggests is insufficient to be sensed by the thermostat. Thus, ASHRAE recommends against the inclusion of internal heat gains, particularly from occupants and lighting, when sizing residential heating systems. In this paper, we take a different approach to investigate the existence and potential magnitude of CFL interactive effects in residential situations. Specifically, we use billing analysis to test for the presence of any interactive effect. The approach used herein is similar to those of Metcalf and Hassett (1999) and Sebold and Fox (1985), who use billing data to examine the returns from energy efficiency investments. Our analysis is based on a comprehensive dataset that includes monthly household electricity and natural gas usage, the number of CFL bulbs installed, the installation date, and a set of household characteristics. Our results suggest that CFLs do indeed save electricity. However, we do not find any support for the hypothesis that CFLs cause increased usage of natural gas. The paper is organized as follows. In the next section we describe the data used in our analysis. Our empirical framework and results are presented in Sections 3 and 4, respectively. Conclusions and policy implications are offered in the final section.
2. Data In late November and early December 2008, San Diego Gas and Electric (SDG&E) compiled a dataset for Low Income Energy Efficiency (LIEE) participants from 2006 and 2007. The LIEE program delivers directly installed energy efficient solutions (i.e., no cost to the recipient) for income qualified households. The income guidelines vary by household size and are established annually by the California Public Utilities Commission.5 The SDG&E dataset included: quantity and type of energy efficiency measures installed; installation date; household income; household size; living area (square footage); and air conditioner ownership (central, room, or none). We merged these data with individual-level billing data obtained from SDG&E. The billing data consist of monthly observations on electricity and natural gas usage, billing days, and climate zone for all LIEE customers in the SDG&E service area and span the time period from January 2005 to December 2008. Because the billing data start in January 2005 and end in December 2008, they include at least 12 months of pre- and post-installation electricity and natural gas usage. Our analysis focuses on the impact that replacing traditional incandescent bulbs with CFLs has on energy usage. Thus, we restrict our attention to households that had CFLs installed only during the time frame and drop households that had multiple energy efficiency measures installed. Specifically, the dataset included approximately 2800 households that had only interior lighting installed. Among these 2800 households, approximately 1600 households had only 15 and 23 W CFLs installed. Of this latter group, the vast majority, 1219 households, had only 15 Watt bulbs installed. In the empirical analysis that follows, we restrict our attention to these 1219 households. To control for weather-related energy usage, we merged the billing data with heating and cooling degree hour data, compiled by SDG&E. Heating and cooling degree hours were developed by mapping individual weather station data to the three climate zones: maritime, coastal, and transitional. Both the heating and 5 The current categories for total yearly household income before deductions are 200% of the federal poverty level or by household size are the following: household size =1–2 ($30,500), household size= 3 ($35,800), household size= 4 ($43,200), household size= 5 ($50,600), household size= 6 ($58,000), and household size greater than or equal to 7 (add $7600/person to $58,000).
ARTICLE IN PRESS 1290
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
Table 1 Electricity summary statistics for full sample and high-use low-income households (1) Full sample
(2) Pre-CFL installation
(3) Post-CFL installation
Variable
Mean
St. dev.
Mean
St. dev.
Mean
St. dev.
All low income households Time varying characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
35.12 0.60 2.99 3.53 2.46 0.20 0.26 0.28 0.26
21.96 0.49 3.19 2.81 2.20 0.40 0.44 0.45 0.44
35.06 0.00 0.00 3.65 2.22 0.48 0.35 0.17 0.00
22.06 0.00 0.00 2.69 2.19 0.50 0.48 0.38 0.00
35.17 1.00 5.09 3.45 2.61 0.01 0.21 0.35 0.44
21.89 1.00 2.57 2.89 2.19 0.10 0.41 0.48 0.50
Household characteristics Square footage of home Household size Household income ($)
896 3.46 22,844
306 1.90 9069
896 3.46 22,844
306 1.90 9069
896 3.46 22,844
306 1.90 9069
High-use low income households Monthly characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
65.90 0.61 3.67 3.65 2.69 0.19 0.25 0.28 0.28
24.66 0.49 2.88 2.88 2.34 0.39 0.43 0.45 0.45
67.30 0.00 0.00 3.80 2.46 0.48 0.33 0.19 0.00
25.17 0.00 0.00 2.76 2.36 0.50 0.47 0.39 0.00
64.97 1.00 5.12 3.55 2.84 0.01 0.21 0.34 0.45
24.28 1.00 2.35 2.95 2.31 0.09 0.40 0.47 0.50
Household characteristics Square footage of home Household size Household income ($)
1095 4.17 25,995
408 2.00 10,726
1095 4.17 25,995
408 2.00 10,726
1095 4.17 25,995
408 2.00 10,726
cooling degree hours are based on a 651 degree set point assumption. Each participant’s home was mapped to one of the three climate zones using a zip code reference table provided by SDG&E. As an alternative, we also tried mapping to the three climate zones using the California Energy Commission’s climate zone to zip code list assuming that climate zone 7 was maritime, zone 10 was coastal, and zone 14 was transitional. In the empirical work that follows, we use the heating and cooling degree hours from SDG&E’s mapping.6 Table 1 presents the electricity usage variable definitions and summary statistics. Electricity usage is measured in thousands of Btus (kBtu). Summary statistics are presented for two specific sample groups: (1) all low-income households (upper panel) and (2) low-income households that have average electricity consumption that equals or exceeds 48.8 kBtu/day (14.3 kWh/day; lower panel). This latter group has average electricity consumption of approximately 65.9 kBtu/day (19.3 kWh/day), which is consistent with overall average electricity usage in the San Diego Gas and Electric service territory (see McNulty et al., 2006). For each group, Column 1 contains electricity usage summary statistics for the entire time period of analysis. In Columns 2 and 3 we provide the same information for the pre-installation and post-installation periods, respectively. Note that household characteristics do not vary from the pre- to the post-period. As expected, low-income households in the full sample (upper panel) have smaller homes and incomes and use less electricity per day than representative San Diego households. However,
6 The results reported later in the paper are not sensitive to which weather mapping we use. Specifically, the core results reported in Tables 3–6 are robust to using the alternative heating and cooling degree hours mapping.
family size exceeds the comparable value in San Diego (2.80 people/household in McNulty et al., 2006). The high-use sample was selected to more closely match the SDG&E service territory customer base. Therefore, the average electricity use is closely aligned with the surrounding population. However, home size, income, and household size are still quite different.7 In terms of the variable of interest, post-installation electricity usage per day declines from the pre-installation period for the high-use group, in spite of the increase in average cooling degree hours, an indicator of air conditioning demand. The full sample shows no significant change in electricity use between study periods. Table 2 presents gas usage summary statistics for both the full sample of LIEE customers (upper panel) and LIEE customers that are in the high-use group (lower panel). High use is defined as an average of 68 kBtu per day (0.68 therms per day) or more, which translates into an overall sample average of 106 kBtu/day (1.06 therms/day). This latter value is consistent with the average natural gas use in the SDG&E service territory (see McNulty et al., 2006). As in Table 1, Column 1 contains gas usage summary statistics for the entire time period of analysis, while Columns 2 and 3 provide the same information for the pre-installation and post-installation periods, respectively. Note that the gas samples differ from the electricity samples in two ways. First, for the full sample, there are fewer low-income
7 The following values were taken from the McNulty, Murdoch, and Thayer study: average daily electricity use= 65.0 kBtu (19.3 kWh), average daily gas use= 1070 kBtu (1.07 therms), average home size= 1725 ft2, and average household size= 2.80 persons. In addition, because only low-income households qualify for the program being studied herein, we know average income is below the comparable average service territory average.
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
1291
Table 2 Natural gas summary statistics for full sample and high-use low-income households (1) Full sample
(2) Pre-CFL installation
(3) Post CFL installation
Variable
Mean
St. dev.
Mean
St. dev.
Mean
St. dev.
All low income households Time varying characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
73.99 0.63 3.15 3.50 2.38 0.20 0.26 0.28 0.26
58.16 0.48 3.27 2.79 2.14 0.40 0.44 0.45 0.44
74.65 0.00 0.00 3.59 2.08 0.52 0.33 0.15 0.00
57.84 0.00 0.00 2.64 2.08 0.50 0.47 0.36 0.00
73.61 1.00 5.19 3.44 2.55 0.01 0.22 0.35 0.41
58.34 1.00 2.94 2.87 2.15 0.11 0.42 0.48 0.49
Household characteristics Square footage of home Household size Household income ($)
931 3.64 23,226
318 1.95 9549
931 3.64 $23,226
318 1.95 9549
931 3.64 23,226
318 1.95 9549
High-use low income households Time varying characteristics Usage (kBtu/day) CFLs installed (1 =yes) Number of CFLs Heating degree hours (1000s) Cooling degree hours (1000s) 2005 2006 2007 2008
106.15 0.64 3.28 3.52 2.41 0.20 0.26 0.28 0.26
61.22 0.48 3.43 2.79 2.15 0.40 0.44 0.45 0.44
109.02 0.00 0.00 3.59 2.12 0.53 0.33 0.14 0.00
59.71 0.00 0.00 2.63 2.09 0.50 0.47 0.35 0.00
104.88 1.00 5.29 3.48 2.57 0.01 0.23 0.36 0.41
62.00 1.00 3.07 2.87 2.16 0.11 0.42 0.48 0.49
Household characteristics Square footage of home Household size Household income ($)
1044 4.18 $24,798
346 2.05 $10,742
1044 4.18 $24,798
346 2.05 $10,742
1044 4.18 $24,798
346 2.05 10,742
gas customers than electricity customers, due to the fact that some homes do not use natural gas. This is especially apparent on comparison of the upper panel of Table 1 to the upper panel of Table 2. Specifically, the full low-income sample for the gas usage model consists of 870 households, whereas the full low-income sample for the electricity usage model consists of 1219 households. Second, the high-use gas group is larger (lower panel, Table 3) than the electricity high-use group (lower panel, Table 1). This occurs because the gas usage threshold is somewhat lower to qualify for high-use designation. Thus, our high-use gas group is about twice as large as the corresponding electricity group. From the perspective of this study, the most interesting result shown in Table 2 is that gas usage does not seem to be affected by the installation of CFLs, as pre-installation and post-installation values are nearly identical for each sample. As with the electricity samples (Table 1), both the full sample of low-income households (upper panel) and the high-use sample have smaller homes and incomes but larger household size than representative San Diego households.
3. Empirical specification To examine the impact of replacing traditional incandescent bulbs with CFLs on electricity usage, we estimate the following model: Eit ¼ ai þ b1 CFLit þ b2 HDHit þ b3 CDHit þ Yeart g1 þMontht g2 þ ðai HDHit Þd1 þ ðai CDHit Þd2 þ eit
ð1Þ
where Eit is the electricity usage, measured in kBtu per day, for household i in time period t; ai is a vector of household fixed
effects; CFLit is the number of compact florescent light bulbs installed in household i in period t (CFLit =0 in the pre-installation period); HDHit the number of heating degree hours; CDHit the number of cooling degree hours; Yeart is a vector of year fixed effects; Montht is a vector of month fixed effects; ai HDHit denotes the interaction of the number of heating degree hours for household i in period t with the household-specific fixed effects; ai CDHit denotes the interaction of the number of cooling degree hours with the household-specific fixed effects; and eit is a random disturbance term. The coefficient of primary interest is b1. For the electricity usage model, b1 should be negative because replacing traditional incandescent bulbs with CFLs should reduce electricity usage. In terms of interpretation, b1 measures the average daily electricity savings associated with the installation of one additional CFL. Note that the inclusion of individual-specific fixed effects in Eq. (1) implies that we are using only ‘‘within household’’ variation in energy consumption to identify the model. Thus, any factor that influences energy consumption but does not vary over time (e.g. square feet of living space, number of occupants in household, etc.) is captured by the individual fixed effects. Consequently, our model controls for any observable or unobservable factors that influence energy consumption but do not vary over time. Furthermore, the inclusion of the interaction terms ai HDHit and ai CDHit adjusts for each household’s weathersensitive energy usage (i.e., individual-specific weather-related energy consumption patterns). While Eq. (1) provides a useful way of modeling electricity savings due to the installment of CFLs, it may be less useful for modeling the interactive effects that the installation of CFLs may have on gas consumption. Specifically, the interactive effect that
ARTICLE IN PRESS 1292
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
Table 3 Estimated model results—electricity usage/day Specification
Full sample
High use
Number of CFLs
0.127nn (0.0637) 0.392nnn (0.115) 0.438nnn (0.111) 0.125 (0.324) 0.927nn (0.436) 1.537nnn (0.575) 2.385nnn (0.201) 4.015nnn (0.273) 4.999nnn (0.353) 4.493nnn (0.449) 3.424nnn (0.554) 1.956nnn (0.630) 0.303 (0.678) 0.328 (0.656) 2.613nnn (0.558) 3.597nnn (0.467) 1.904nnn (0.334)
0.423nn (0.178) 2.708nnn (0.258) 0.148 (0.307) 0.613 (1.120) 1.840 (1.442) 2.143 (2.080) 3.692nnn (0.673) 6.510nnn (1.014) 8.419nnn (1.330) 7.536nnn (1.669) 5.801nnn (2.008) 1.468 (2.290) 2.372 (2.417) 0.970 (2.293) 4.766nn (1.935) 7.206nnn (1.611) 3.333nnn (1.133)
38592 1219 0.85
7310 240 0.72
Cooling degree hours (1000s) Heating degree hours (1000s) 2006 2007 2008 February March April May June July August September October November December Observations Number of households R2
Notes: Dependent variable is in kBtu/day. Column 1 presents basline electricity usage estimates for full sample of LIEE households with 15 W CFL installation. Column 2 presents electricity usage estimates for sample of high-usage LIEE households with 15 W CFL installation, where high usage is defined as electricity usage above 48.8 kBtu/day (14.3 kWh/day), which corresponds to an average usage of 65.9 kBtu/day (19.3 kWh/day). All models include individual-specific heating degree hour slopes and individual-specific cooling degree hour slopes. Robust clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnnsignificant at 1%.
CFLs have on gas consumption arises due to the change in ambient air temperature associated with replacing a traditional incandescent bulb with a compact fluorescent bulb.8 The resulting change in ambient air temperature will depend on several factors, such as the wattage of the traditional incandescent bulb that is being replaced, the flow of air through a home, and perhaps most importantly, volume (size) of the space within which the light bulbs are housed. To see this, note that if we consider a room that does not leak any heat, we can express the change in room temperature associated with use of a traditional incandescent bulb of a given wattage as
DT ¼
Watts t Vdc
ð2Þ
where DT denotes the change in temperature; Watts is the wattage of a traditional incandescent bulb; t the time measured in 8 Note that there is also a potential interactive effect for electricity. Specifically, replacing incandescent lights with CFLs may result in a lower cooling season temperature and hence a household may utilize its air conditioning system less often. The dataset we have assembled for this study is not sufficiently comprehensive to estimate this effect.
seconds; V the volume of the room measured in m3 (meters cubed); d the density of air measured in kg/m3 (kilograms per meter cubed); and c the specific heat capacity of air measured in kJ/(kg 1C) (kilojoules per kilograms degree Celsius). Thus, for any given values of d and c, the change in room temperature varies inversely with the size of the room and the larger the room, the smaller the change in ambient air temperature. Consequently, the impact that replacing traditional incandescent bulbs with CFLs has on ambient air temperature should be larger in smaller rooms and smaller in larger rooms. This suggests that the relationship between gas usage and installation of CFLs is not linear as specified in Eq. (1). Rather, the impact CFLs have on gas usage depends critically on the density of CFLs, measured for example, as CFLs per square feet. The discussion above suggests that an ideal way to measure the potential impact CFLs have on gas usage would be to construct some measure of the number of CFLs per volume (height width length) of heating area. Unfortunately, to our knowledge, no such data exist. However, we do have information on the square footage of homes, a variable that should be highly correlated with the volume of heating area. Thus, to examine the impact of replacing traditional incandescent bulbs with CFLs on gas usage, we estimate models of the following form: Git ¼ ai þ b1 CFLit =SQFTi þ b2 HDHit þ b3 CDHit þ Yeart g1 þ Montht g2 þ ðai HDHit Þd1 þ ðai CDHit Þd2 þ eit ð3Þ where Git is the gas usage, measured in kBtu per day, for household i in time period t, SQFTi denotes the square footage of housing unit i, and all other variables are as defined in Eq. (1). In Eq. (3), b1 measures the average effect on gas consumption of installing one additional CFL per square foot of living space. If interaction effects are present, b1 should be positive and significantly different from zero. Furthermore, note that the marginal effect of installing one additional CFL has on gas usage can be expressed as
@ Usage 1 ¼ b1 @ CFL SQFT
ð4Þ
which depends on the square footage of a home—the larger the home, the smaller the marginal effect of an additional CFL on gas usage.
4. Empirical results 4.1. Electricity usage results Results based on the estimation of Eq. (1) are presented in Table 3. In all regressions, the dependent variable equals kBtu per day. In the first column, we present electricity usage estimates for the full sample of LIEE households that have had only 15 W CFLs installed. In the second column, we present electricity usage estimates for the high-use sample of LIEE households that have had only 15 W CFLs installed. All models also include individualspecific heating degree hour slopes and individual-specific cooling degree hour slopes (see Eq. (1) above). For brevity, the approximately 3650 (full sample) or 720 (high-use sample) coefficients are not presented, but are available on request. All standard errors reported in Table 3 and subsequent tables are clustered at the individual level to allow for within-individual autocorrelation of the disturbance term. The indicators n, nn, and nnn imply the variable is significant at the 10% level, 5% level, or 1% level, respectively.
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
There are several noteworthy results in Table 3. First, a high proportion of the variation in electricity usage is explained by the independent variables as R2 ranges from 0.72 (high-use sample) to 0.85 (full sample). Second, the month and year dummy variables, which are interpreted as variations from the relevant omitted category (2005 for year, January for month), perform as expected. Specifically, there is some evidence that electricity usage is increasing over time, especially for the full sample. In addition, the monthly effects suggest that, all else being constant, electricity use/day in January is similar to that in August–September period (July–September for high-use sample), whereas during all the other months, daily electricity usage is significantly below this level. Third, heating and cooling degree hours perform as expected (e.g., more cooling degree hours imply more cooling demand, which is positively related to electricity use), even when the model includes individual-specific heating and cooling degree hour slopes. It should be noted that if the individual-specific heating and cooling degree hour slopes are omitted from the model, then the heating and cooling degree hours variables continue to perform as expected. The final important result pertains to our focus variable, the number of compact fluorescent (CFL) bulbs installed in the home. As is indicated in Table 3, the coefficient on the number of CFLs is negatively related to daily electricity usage and is significant at the 5% level in each of the estimated models. The calculation of the impact of an additional CFL is straightforward. The coefficient multiplied by 365 yields the annual savings from an additional bulb. For the full sample of low-income households in Table 3, the savings value attributable to another CFL is approximately 46.4 kBtu/year (13.6 kWh/year). This value is very close to those reported in a recent study by West Hill Energy and Computing, Inc. et al. (2007).9 The results for the full sample can also be compared to those of our high-use sample in Table 3. The estimated coefficient on number of CFLs for this latter group suggests significantly larger savings than those for the full sample. Specifically, the electricity savings attributable to another CFL for the high-use group are approximately 155 kBtu/year (45.3 kWh/year). If a high-use household installed the average number of CFLs (5), then annual electricity savings would be 773 kBtu (226.5 kWh). Dividing this value by annual usage of 24,565 kBtu (from Table 1 average daily usage of 67.30 kBtu times 365) produces annual electricity savings as a percentage of total usage of 3.2%. This percentage reduction in overall electricity usage is very close to the estimated value in Wall and Crosbie (2009), which was 3.31% (50.9% reduction in the 6.5% share of total electricity use). The magnitude of electricity savings also suggests that if each household were to replace five incandescent bulbs with CFLs, California would be approximately one-third the way to achieving the projected 9% electricity energy savings by 2016 (see California Energy Commission, 2007). 4.2. Gas usage results Having established that the replacement of traditional incandescent bulbs with CFLs results in relatively large and statistically significant reductions in electricity usage, we now turn to examine the impact of CFLs on gas usage. Table 4 presents gas usage results based on the estimation of Eq. (3). In all 9 In this study, the estimated coefficient on CFLs produced annual savings of 37.5 kBtu/year (11 kWh/year) with a confidence interval of 20.5–54.6 kBtu/year (6–16 kWh/year). The authors stated that their regression results were likely biased downward due to CFL attrition and purchase outside the program, so they recommended the high end of the confidence interval (see West Hill Energy and Computing and Inc. et al., 2007)
1293
Table 4 Estimated model results—gas usage/day Specification
Full sample
High use
Number of CFLs/ft2
0.184 (160.7) 3.034nnn (0.165) 5.442nnn (0.247) 3.525nnn (0.957) 2.485n (1.298) 2.415 (1.547) 0.453 (0.815) 4.570nnn (0.868) 10.06nnn (1.082) 8.884nnn (1.232) 6.378nnn (1.508) 8.178nnn (1.712) 10.24nnn (1.842) 13.01nnn (1.777) 13.52nnn (1.617) 11.81nnn (1.368) 9.027nnn (0.992)
148.1 (278.6) 3.386nnn (0.287) 4.822nnn (0.417) 4.994nnn (1.659) 3.384 (2.219) 3.635 (2.446) 0.716 (1.461) 7.349nnn (1.463) 15.41nnn (1.866) 13.82nnn (2.105) 9.658nnn (2.585) 11.47nnn (2.922) 14.02nnn (3.152) 18.24nnn (3.023) 19.63nnn (2.779) 16.96nnn (2.365) 12.54nnn (1.714)
28190 870 0.859
14581 435 0.797
Cooling degree hours (1000s) Heating degree hours (1000s) 2006 2007 2008 February March April May June July August September October November December Observations Number of households R2
Notes: Dependent variable is in kBtu/day. Column 1 presents baseline gas usage estimates for full sample of LIEE households with 15 W CFL installation. Column 2 presents gas usage estimates for sample of high-use LIEE households with 15 W CFL installation, where high usage is defined as gas usage above 68 kBtu /day (0.68 therms/day), which corresponds to an average usage of 106 kBtu/day (1.06 therms/day). All models include individual-specific heating degree hour slopes and individual-specific cooling degree hour slopes. Robust clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnn significant at 1%.
regressions, the dependent variable is natural gas usage in kBtu per day. Similar to Table 3, in the first column we present gas usage estimates for the full sample of LIEE households that had only 15 W CFLs installed. In the second column, we present estimates for the high-use sample of LIEE households. All models (once again) include individual-specific heating and cooling degree hour slopes. Similar to the electricity usage results, the results reported in Table 4 indicate that a high proportion of the variation in gas usage is explained by the model as indicated by an R2 of 0.86 for the full sample and an R2 of 0.80 in the high-use sample. The month and year dummy variables, which are interpreted as variations from the relevant omitted category (2005 for year, January for month), perform as expected. The annual trends indicate that gas consumption in 2006–2008 is significantly smaller than the usage recorded in 2005, even though there has been no substantive change in heating degree hours. This may reflect a price effect. The months with relatively high gas usage are January and February. Minimum gas usage occurs in the September–November period. The heating and cooling degree hours variables perform as expected, suggesting that individuals
ARTICLE IN PRESS 1294
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
Table 5 Robustness checks: full sample of LIEE customers Panel
Specification
Baseline
Individual-specific heating degree hour slopes only
Dropping month of installment
Dropping homes with 1700 ft2 of living area or more
(1)
Number of CFLs/ft2
0.184 (160.7)
45.8 (152.1)
34.7 (162.5)
26.1 (163.3)
(2)
Number of CFLs/ft2
74.1 (164.6) 20.9 (25.1)
24.5 (154.1) 20.2 (24.3)
102.3 (165.6) 19.5 (28.6)
50.4 (167.7) 21.6 (25.4)
CFLs/ft2 heating degree hours
(3)
Number of CFLs
0.191 (0.174)
0.229 (0.165)
0.151 (0.18)
0.258 (0.187)
(4)
Number of CFLs
0.113 (0.18) 0.021 (0.0342)
0.154 (0.173) 0.0203 (0.0331)
0.0953 (0.182) 0.0153 (0.0382)
0.172 (0.198) 0.0232 (0.0371)
CFLs heating degree hours
Notes: Table presents various specification checks for the full sample of LIEE customers with 15 W CFL installation. Dependent variable in all specifications is in therms/day. Panels 1 and 2 contain estimates when the number of CFLs/ft2 is the key explanatory variable. Panels 3 and 4 contain estimates when the number of CFLs is the key explanatory variable. Robust, clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnnsignificant at 1%.
in our low income sample modify their gas usage in response to variations in weather. Turning to the variable of interest, the estimated coefficient on the number of CFLs per square feet is positive, as predicted if interactive effects are present, but never statistically different from zero at any commonly used significance level. For the results based on the full sample, the point estimate on the number of CFLs per square feet implies that each additional CFL per square foot increases natural gas usage by approximately 0.2 kBtu/day (0.002 therms/day). Evaluated at the mean square footage of homes in the sample, this estimate implies an interactive effect of only 0.07 kBtu (0.0007 therms) per year.10 Furthermore, the magnitude of this effect declines with square footage, so that if the square footage of a representative San Diego home were to be used (1725 ft2) in the calculation, the magnitude of the interactive effect would be approximately half as large. For the high-use sample, the estimated coefficient on the number of CFLs is significantly larger in magnitude but, once again, statistically insignificant. Evaluated at the mean square footage of homes in the sample, the estimate coefficient implies an interactive effect of approximately 52 kBtu (0.52 therms) per year. If the square footage of a representative San Diego home were to be used in the calculation, the magnitude of the interactive effect would fall to approximately 31 kBtu (0.31 therms) per year. 4.3. Robustness checks The results reported in Table 4 suggest that interactive effects for CFLs are not present in a residential setting. Because this result is counter to the DEER Team position, we considered several other models of gas usage, and various alternative specifications, to determine whether or not the results reported in Table 4 are robust. Specifically, we examined whether our results were robust to the following alternative specification checks: (1) using only individual-specific heating degree hour slopes rather than both individual-specific heating and cooling degree hour slopes; (2) eliminating the month in which the installment of CFLs occurred; (3) eliminating homes with a living area of 1700 ft2 or more; (4) including the number of CFLs per square feet interacted with the number of heating degree hours as an explanatory 10
0.07 ft2 =(0.184 365)/931 ft2 for full sample.
variable; (5) replacing the number of CFLs per square feet with simply the number of CFLs as in the electricity usage models; and (6) including the number of CFLs interacted with the number of heating degree hours as the variable of interest. We consider models where we include only individual-specific heating degree hour slopes to examine whether our results are sensitive to how we model individual-specific weather-related energy consumption patterns. We drop the month in which the CFLs were installed to reduce any measurement error associated with uncertainty concerning the exact date of installation. We drop homes with a living area of 1700 ft2 or more (top 2.5% of homes in our sample) to test whether interactive effects are more likely to exist in smaller homes. Finally, we estimate models that include an interaction term between the number of CFLs per square feet (or the number of CFLs) and the number of heating degree hours to examine whether the interactive effect CFLs may have on gas usage varies with the number of heating degree days. The results for these robustness checks for the full sample of LIEE customers are presented in Table 5. In the interest of brevity, we report only the estimated coefficients on the number of CFLs per square feet (or number of CFLs) and the interaction term between the number of CFLs per square feet (or number of CFLs) and the number of heating degree hours. We note however, that all models include the same control variables listed in Table 4 and, with the exception of Column 2, all models include individualspecific heating degree hour and individual-specific cooling degree hour slopes. The first two panels of Table 5 report estimates when the key variable of interest is the number of CFLs per square foot of living area. Panels 3 and 4 present estimates when the key variable of interest is the number of CFLs. For comparison purposes, Column 1 of Panel 1 reports the estimated coefficient on the number of CFLs given in Column 1 of Table 4. A brief inspection of Table 5 reveals that none of the estimated coefficients reported in the table is statistically different from zero. The estimated coefficient on the number of CFLs per square feet becomes larger in magnitude when we drop the month during which the CFLs were installed, although it remains statistically insignificant at any conventional significance level. In contrast, the estimated coefficient on the number of CFLs per square feet turns negative when we utilize only individualspecific heating degree hour slopes or drop homes with 1700 ft2 or more of living area. When we replace the number of CFLs per
ARTICLE IN PRESS E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
1295
Table 6 Robustness checks: high-use sample of LIEE customers Panel
Specification
Baseline
Individual-specific heating degree hour Slopes Only
Dropping month of installment
Dropping homes with 1700 ft2 of living area or more
(1)
Number of CFLs/ft2
148.1 (278.6)
71.7 (267.5)
211.6 (279.9)
119.6 (283.8)
(2)
Number of CFLs/ft2
170.4 (284.5) 6.37 (45.2)
108.7 (269.1) 10.8 (43.8)
212.9 (276.7) 0.381 (53.4)
133.6 (292.0) 4.01 (45.8)
CFLs/ft2 heating degree hours
(3)
Number of CFLs
0.209 (0.292)
0.272 (0.279)
0.131 (0.298)
0.245 (0.317)
(4)
Number of CFLs
0.179 (0.315) 0.00801 (0.0571)
0.229 (0.303) 0.0115 (0.0552)
0.141 (0.314) 0.00267 (0.0646)
0.252 (0.342) 0.00176 (0.0622)
CFLs heating degree hours
Notes: Table presents various specification checks for the high-use sample of LIEE customers with 15 W CFL installation. Dependent variable in all specifications is in therms/day. Panels 1 and 2 contain estimates when the number of CFLs/ft2 is the key explanatory variable. Panels 3 and 4 contain estimates when the number of CFLs is the key explanatory variable. Robust, clustered standard errors in parentheses. nSignificant at 10%; nnsignificant at 5%; nnnsignificant at 1%.
square feet with the number of CFLs, the estimated coefficient on the number of CFLs is consistently negative (incorrect sign) and never close to being statistically significant. Table 6 presents our robustness checks for the sample of highuse LIEE customers. The results reported in Table 6 tend to mirror the results reported in Table 5. Specifically, similar to Table 5, none of the estimated coefficients reported in the Table 6 is statistically different from zero. Thus, across a wide array of specifications, our results consistently suggest that interactive effects for CFLs are not present in a residential setting.
5. Concluding remarks The primary results of our research effort are twofold. First, there is strong statistical evidence that replacing incandescent lights with compact fluorescent lights in a residential setting generates significant electricity savings. The coefficient that relates the installation of CFLs to electricity usage is both negative and significantly different from zero at the 5% level. The magnitude of the effect (46.4 kBtu/year or 13.6 kWh/year) for the full sample of low-income households is consistent with a recent study authored by West Hill Energy and Computing and Inc. et al. (2007). In addition, the magnitude of the effect (155 kBtu/year of 45.3 kWh/year) for our high-use sample, which is designed to be more closely representative of SDG&E service territory households, is consistent with (although slightly greater than) estimated savings in the DEER documentation and with a recent study by Wall and Crosbie (2009). The second significant result is that there is no statistical evidence of any interactive effect of CFLs on the usage of natural gas. In the full sample of low-income households, the estimated interactive effect is clearly indistinguishable from zero, both in the statistical sense and in the magnitude of the coefficient. In the high-use sample, the estimated effect is also not significantly different from zero at any conventional significance level. In terms of the magnitude of the effect (ignoring statistical significance), the estimated coefficient converts to an interactive effect that ranges from 0.07 kBtu (0.0007 therms) per year for the full sample to 52 kBtu (0.52 therms) per year for the high-use group. This effect diminishes with square footage of living area so that the range for an average sized home (approximately 1725 ft2) is
0.038 kBtu for the full sample to 31 kBtu per year for the high-use group. Our finding that interactive effects of CFLs on natural gas usage are indistinguishable from zero holds across a wide array of specifications. That is, the results are invariant to various possible specifications of the relationship between CFLs and gas usage. The primary policy implications of these findings for energy efficiency programs are the following. First, our results strongly suggest that programs designed to replace incandescent light bulbs with CFLs can generate significant electricity savings in the residential sector. Second, the lack of any statistically measurable interactive effect suggests that CFLs have little to no effect on natural gas consumption. This, in turn, suggests that the replacement of incandescent bulbs with CFLs does not lead to a significant offset in the energy savings as suggested by the Database for Energy Efficient Resources (DEER). The results of this study are subject to two caveats. First, the sample sizes are smaller than we prefer. For example, the high-use sample of electricity usage is only 240 households. Second, this study considers only low-income households, and they might not be representative of typical households in the SDG&E service territory that would participate in the energy efficiency programs. Therefore, it would be interesting to test for the existence and potential magnitude of interactive effects using either a larger sample of low-income households and/or a general sample of the overall population in order to replicate the results of this study.
References Banerjee, A., Solomon, B.D., 2003. Eco-labeling for energy efficiency and sustainability: a meta-evaluation of U.S. programs. Energy Policy 31, 109–123. Cadmus Group, Inc., KEMA, Itron, Inc., Nexus Market Research, and Goett Consulting. 2009. Compact fluorescent lamps market effects: interim report—draft. Prepared for the California Public Utilities Commission, January. California Energy Commission. 2007. Statewide energy efficiency potential estimates and target for California utilities: draft staff report. CEC-200-2007019-SD, August. Electric Power Research Institute (EPRI). 2009. Assessment of achievable potential from energy efficiency and demand response programs in the U.S. 2010–2030. EPRI Technical Report, January. Energy Information Administration. 2005. U.S. household electricity report, www. eia.doe.gov, July. Ford, P., 2008. Interactive effects of CFL lighting in residential applications. Engineering Opinion to Sempra Utilities October.
ARTICLE IN PRESS 1296
E.J. Brunner et al. / Energy Policy 38 (2010) 1288–1296
McNulty, M., J. Murdoch, and M. Thayer, The home energy comparison tool. Prepared for San Diego Gas and Electric Company, Final Report, June 2006. Metcalf, G.E., Hassett, K.A., 1999. Measuring the energy savings from home improvement investments: evidence from monthly billing data. The Review of Economics and Statistics 81 (3), 516–528. Sanchez, M.C., Brown, R.E., Webber, C., Homan, G.K., 2008. Savings estimates for the United States Environmental Protection Agency’s ENERGY STAR Voluntary Labeling Program. Energy Policy 36, 2098–2108.
Sebold, F.D., Fox, E., 1985. Realized savings from residential conservation activity. The Energy Journal 6 (2), 73–88. Wall, R., Crosbie, T., 2009. Potential for reducing electricity demand for lighting in households: an exploratory socio-technical study. Energy Policy 37 (3), 1021–1031. West Hill Energy and Computing, Inc., et al., Impact evaluation of the 2005 California Low Income Energy Efficiency Program. Final Report, Prepared for Southern California Edison Company, December 2007.