American Economic Review 2016, 106(10): 2867–2894
http://dx.doi.org/10.1257/aer.20150102
How Much Energy Do Building Energy Codes Save?
Evidence from California Houses†
By Arik Levinson*
Regulations governing the energy efficiency of new buildings have
become a cornerstone of US environmental policy. California enacted
the first such codes in 1978 and has tightened them every few years
since. I evaluate the resulting energy savings three ways: comparing
energy used by houses constructed under different standards, controlling for building and occupant characteristics; examining how
energy use varies with outdoor temperatures; and comparing energy
used by houses of different vintages in California to that same difference in other states. All three approaches yield estimated energy
savings significantly short of those projected when the regulations
were enacted. (JEL Q48, Q51, Q52)
In 1978 California began enacting some of the world’s first and most ambitious
residential building energy codes, aiming “to reduce the electricity and gas now used
in typical new buildings by at least 80 percent for new buildings constructed after
1990.”1 This ambitious goal was detailed in analyses published by the California
Energy Commission (CEC) and widely cited in the press at the time.2 Numerous
subsequent revisions to the California codes have projected further savings. The
2008 revisions were designed to reduce electricity used by s ingle-family detached
homes for heating, cooling, lighting, and water heating by 23 percent, and the 2013
revisions by another 36 percent.3 Today, some observers claim that the building
energy codes are meeting those goals.4
Calculating the energy savings from building energy codes requires knowing
how much energy would have been used in the absence of the codes, a far more
* Department of Economics, Georgetown University, 3700 O Street, Washington, DC 20057, and NBER (e-mail:
arik.levinson@georgetown.edu). I am grateful to the Georgetown Environmental Initiative for financial support,
to staff at the California Energy Commission for help accessing and understanding the Residential Appliance
Saturation Study billing data, to Jonah Birnberg, Julia Kieserman, and Derek Levinson for research assistance, and
to Max Auffhammer, Sarah Aldy, Sally Atwater, Mary Ann Bronson, Dallas Burtraw, Howard Chong, Hal Harvey,
Richard Hirsh, Grant Jacobsen, Matthew Kahn, Matthew Kotchen, Robbie Orvis, Erica Myers, and three referees
for especially helpful suggestions. The author declares that he has no relevant or material financial interests that
relate to the research described in this paper.
†
Go to http://dx.doi.org/10.1257/aer.20150102 to visit the article page for additional materials and author
disclosure statement.
1
California Energy Commission (1979, p. 55).
2
Horn et al. (1980) and “Energy Plans for New Homes,” Los Angeles Times, October 16, 1980, B3.
3
Architectural Energy Corporation (2007) and California Energy Commission (2013).
4
See, for example, Thomas Friedman citing Hal Harvey: “New houses in California now use one-fourth of
the energy they used 25 years ago,” in “Go Ahead, Vladimir, Make My Day,” New York Times, April 12, 2014,
http://www.nytimes.com/2014/04/13/opinion/sunday/friedman-go-ahead-vladimir-make-my-day.html (accessed
August 12, 2016).
2867
2868
THE AMERICAN ECONOMIC REVIEW
october 2016
50
40
MBTU
30
20
First
building
energy
codes
Natural gas
10
Electricity
0
40
19
Pre
1
9
94
–1
0
94
1
9
95
–1
0
95
1
92
74 77 82
97 00 04 08
19
19 19 19
19 20 20 20
3–
0– 75– 78–
3– 98– 01– 05–
8
7
9
19
19 19 19
19 19 20 20
9
96
–1
0
96
Year house constructed
Figure 1. Average Annual Household Energy in California, 2003 and 2009
Note: CEC’s (2003, 2009) Residential Appliance Saturation Study (RASS), single-family
detached homes without electric heat or hot water
difficult calculation than is sometimes suggested. We cannot just take engineers’
ex ante estimates of how much less energy a given building will use because that
ignores the ex post response by the building’s occupants. We cannot easily compare jurisdictions with more and less strict building codes because those jurisdictions presumably chose to enact the codes based on the energy-using characteristics
of their residents. And we cannot simply compare energy use by residents of efficient and inefficient buildings because people with larger energy needs may select
energy-efficient homes.
Figure 1 illustrates that challenge. It shows the current average annual household
electricity and natural gas used by California houses according to when they were
constructed, both measured in thousands of British thermal units (MBTU) of energy.5 Houses built recently are not using dramatically less energy than older houses
built under less strict building energy codes. Newer houses use a third less natural
gas but 50 percent more electricity. The comparison is not fair, of course, because
houses built more recently are larger, have more occupants, and are in less temperate
parts of the state, and because both patterns start before the first building codes in
1978. Controlling for those home features, time trends, and the selection of people
with high energy demand into recently-built homes, is the objective here.
The stakes are far higher than whether one state’s building energy codes have
worked as promised. Energy efficiency policies like California’s are held up as a
model for other states and countries and have become central to US climate policy.
These include appliance standards, building weatherization incentives, and other
5
One kilowatt hour of electricity is the equivalent of 3.412 MBTU; one therm of natural gas is 100 MBTU
(www.eia.gov).
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2869
e fficiency-related regulations. More than o ne-third of the greenhouse gas emissions
reductions targeted by Massachusetts in its Global Warming Solutions Act of 2008
were projected to come from energy efficiency improvements to buildings and appliances. Nearly 20 percent of the reductions in California’s 2006 Global Warming
Solutions Act were projected to come from new energy efficiency standards for
buildings and utilities.6 And half of the projected carbon reductions from the US
Environmental Protection Agency’s Clean Power Plan for existing electric power
plants are expected to come from demand-side energy efficiency improvements,
including building energy codes. The United States is putting much of its climate
effort into this one policy.
Americans now have four decades of experience with energy efficiency standards. But most attempts to measure the energy savings fail to address the empirical
challenges of behavioral responses, policy endogeneity, and selection by occupants. I address those problems in three ways. First, I use the CEC’s (2003, 2009)
Residential Appliance Saturation Study (RASS) to estimate annual energy use as a
function of resident demographics, building characteristics, and year of construction.
If building codes save energy, otherwise similar houses built more recently under
stricter standards should use less energy. Though this first approach still ignores the
potential selection of tenants into buildings, I control for a richer set of occupant and
house characteristics than has been possible before.
The second approach focuses on the sensitivity of energy use to outdoor temperature changes. I match the monthly utility bills from the California RASS data
with monthly temperatures in the households’ zip codes. If building codes work
as promised, during months that are hotter than usual, electricity use for air conditioning should increase less steeply in buildings constructed under more stringent
standards. And similarly, in months that are colder than usual, natural gas use for
heating should increase less steeply.
Finally, the third approach compares California with the rest of the United States.
California established the nation’s first and most stringent building energy efficiency
standards and today still appears at the top of rankings of state energy efficiency regulations.7 If the building codes have been as effective as suggested, new houses in
California should use less energy than older houses, and that gap should be larger in
California than in other states. For this third approach I need data on non-California
houses, and so I turn to a nationwide data source, the US Energy Information
Administration’s (1993–2009) Residential Energy Consumption Survey (RECS).
With two energy sources—electricity and natural gas—and three empirical strategies, I generate six basic findings. For electricity, after controlling for building
and occupant characteristics, houses constructed after about 1990 are using 10 to
15 percent less electricity than those built before California’s building codes were
enacted in 1978. But that difference between newer and older houses is no different
in California than in other US states with less strict building codes, and the newer
houses in California do not increase their electricity use less when the weather is
hot. For natural gas, after controlling for building and occupant characteristics,
6
Massachusetts Office of Energy and Environmental Affairs (2010, p. ES-6); and California Air Resources
Board (2008, p. 17).
7
American Council for an Energy Efficient Economy (ACEEE), www.aceee.org.
2870
THE AMERICAN ECONOMIC REVIEW
october 2016
houses built after 1990 are using up to 25 percent less energy than those built before
1978, and the newer houses do increase gas use by less when outside temperatures
fall. But again, that difference between houses built at different times is no different
in California than in the rest of the country. Although the findings vary by energy
source and empirical approach, all six estimates fall significantly short of the savings anticipated when the building codes were established.
Because results like these can be controversial, let me be clear about what this
paper does not say. Nothing here should be interpreted as a general critique of
building energy codes. The codes are part of construction standards that prohibit
homebuilders from cutting corners on hidden costs like fire safety, electric wiring,
plumbing, and also insulation and appliance quality. Plus, new and old buildings
might use similar amounts of energy today because residents of efficient houses
respond to the lowered cost of lighting, air conditioning, and heating by using
more: the s o-called “rebound” effect. In that case, the codes may make homeowners
warmer in winter and cooler in summer but not save as much energy as promised.
Nor should anything here be taken as evidence that energy efficiency technologies
themselves don’t work as promised, only that mandated new building codes alone, in
the absence of other public policies, haven’t delivered as much savings as projected.
This paper studies only the codes that apply to new houses. Many other policies
affect home energy use, including codes applicable to remodels, retrofit subsidies,
appliance and lighting standards, and energy prices. New and old houses might use
similar amounts of energy today because owners of older homes have taken steps
to increase their energy efficiency, either on their own or motivated by subsidies or
building codes applicable to remodels. But if those retrofits do explain the fact that
energy use by new and old houses differs less than expected, we shouldn’t credit
policy with the sum of projected savings from retrofits and new building codes. That
would double count the savings. In the analyses below I do control for house characteristics related to retrofits. Those controls do not explain the fact that the estimated
energy savings from new building codes fall short of projections.
Before describing the research to date on this question, it is worth looking at the
policy and projected savings in detail.
I. Projected Savings and the Evidence So Far
California enacted its first building energy codes in 1978 and has updated them
13 times since.8 To describe those codes, in Table 1 I have reproduced one page
from a detailed cost-benefit analysis done by CEC in 1980. The report contains 48
separate analyses, 1 each for detached, attached, and multifamily homes in each of
16 parts of the state. The example in Table 1 is for single-family detached homes in
Sacramento. Column 1 reports the expenditures on various e nergy-related home construction features without the new California building codes: the “business-as-usual”
costs. Column 2 reports the expenditures associated with the building codes. I’ve
added column 3, the difference, demonstrating that the codes added $8,000 to the
cost of constructing a new house, about 10 percent of the median 1980 California
8
See the California Energy Commission, “Building Energy Efficiency Program,” www.energy.ca.gov/title24/.
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2871
Table 1—Projected Costs and Savings from 1980 California Energy Codes:
Single-Family Homes, Sacramento
Business
as usual
(1)
Insulation
Window glazing
Overhang
Shading
Caulking, sealing, etc.
Thermostat
Heating system
Cooling system
Duct insulation
Total building envelope
Water heater
Lighting
Total initial cost
Total energy (1,000 BTU)
Energy savings
Regulation
(2)
Difference
(3)
—
$879
—
—
—
82
1,360
1,129
—
$2,831
2,108
468
360
551
138
1,360
965
61
$2,831
1,229
468
360
551
56
—
−164
61
$3,450
$8,842
$5,392
284
97
2,736
333
2,452
236
$3,831
$11,911
$8,080
187,209
43,025
−144,184
−77%
Note: The median California home price in 1980 was $80,000.
Source: Horn et al. (1980)
home price. In return for that investment, the home was expected to save nearly
80 percent of the energy that a pre-code home would have consumed. And the
Sacramento house described in Table 1 is no outlier: across the 16 regions studied
by that 1980 CEC analysis, the average energy savings projected was 75 percent.
Have energy codes like those described in Table 1 lived up to their e nergy-saving
promise? Given the importance of this question and the challenges to answering it, a
wide variety of strategies have been taken. Most assessments of the energy savings
from building codes rely on engineering analyses. CEC estimated that its residential
building codes saved 7,039 gigawatt hours of electricity in 2012, or 7.8 percent of
total residential demand.9 This calculation presumes the building codes are enforced,
the savings predicted by engineers are realized, and there is no behavioral response.
But there is reason to doubt all three assumptions. Jaffe and Stavins (1995) show
that actual levels of insulation in homes did not increase as required by building
energy codes. Metcalf and Hassett (1999) show that when insulation is installed, the
realized savings fall short of engineers’ predictions. And in theory, when building
codes reduce heating and air conditioning costs, the rebound effect may result in
people using those systems more (Gillingham, Rapson, and Wagner 2014).
As an alternative to engineers’ predictions, some researchers have regressed aggregate local energy consumption on energy prices, weather, population demographics,
and some proxy for energy efficiency policies. Haeri and Stewart (2013) use lagged
expenditures on utility energy efficiency programs as the measure of policy and conclude that the $7 billion California utilities spent reduced electricity consumption by
9
CEC (2014, Table 2 (2012 total) and Table 25 (savings)).
2872
THE AMERICAN ECONOMIC REVIEW
october 2016
6.5 percent, at an average cost of $0.03 per kilowatt hour. Horowitz (2007) groups
US states into quartiles based on the US Energy Information Administration’s
reported cumulative energy savings from d emand-side management programs and
finds that states with the strongest commitments to energy efficiency saw a 9.1 percent increase in residential electricity use relative to states with weaker commitments. These types of studies typically ignore the potential endogeneity of the key
policy variables. Utilities expecting faster growth in electricity demand or with
conservation-minded constituents may invest more in energy efficiency programs.
One clever version of this regression-based approach that does address policy
endogeneity is Aroonruengsawat, Auffhammer, and Sanstad (2012). They regress
per capita residential electricity consumption in US states on energy prices, weather,
and the share of housing stock built since each state’s initial implementation of
energy building codes. Because implementation is endogenous, the authors use
lagged heating and cooling degree-days as an instrument, on the theory that particularly harsh winters or hot summers spurred states to enact building energy codes.
They find that states with a higher fraction of housing stock built after building
codes were enacted use less energy per capita, and that those savings amount to 2 to
5 percent of nationwide residential energy use.
An altogether different strategy decomposes changes in energy demand into
those components due to exogenous trends and attributes the remainder to efficiency. Studies differ in what changes they control for as being unrelated to energy
efficiency. Metcalf (2008) shows that US energy consumption per dollar of gross
domestic product declined by 47 percent from 1970 to 2003, about one-quarter of
which can be explained by changing personal consumption expenditures, value
added by businesses, and vehicle miles traveled. The remaining three-quarters he
ascribes to energy efficiency, though he is using a broad definition of efficiency that
includes other demographic changes and reduced consumption.
Hojjati and Wade (2012) control for shifts in the size and mix of housing types,
the regional distribution of households, and weather. Even after accounting for
those trends, from 1980 to 2005, US household energy consumption per square foot
decreased by 38 percent, which the authors ascribe to prima facie “evidence of the
efficacy of … e nergy-efficiency … standards and programs” (Hojjati and Wade,
p. 304). But that 38 percent decrease is still at least partly attributable to trends their
decomposition omits. For example, during that period, the average household size
declined by 7 percent, driving down energy consumption per household but driving
up energy consumption per person.
The best approaches use household-level data and focus on particular programs.
Davis, Fuchs, and Gertler (2014) evaluate a Mexican program that gave consumers subsidies to replace old appliances with newer, more e nergy-efficient models.
Households that replaced their refrigerators did use less electricity, but only by
one-quarter of the amount predicted. And households that replaced air conditioners
used more electricity after the replacement than before. Grimes et al. (forthcoming)
study a New Zealand program that retrofitted 12,000 homes with insulation and
clean heat sources. They find that homes retrofitted with insulation used 1 percent
less energy, but the homes retrofitted with clean heat used more energy.
The ideal approach would randomly assign residents to energy-efficient and inefficient homes. Though that is impractical, Fowlie, Greenstone, and Wolfram (2015)
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2873
try the next-best strategy. They encourage a randomly selected subset of eligible
homeowners to take up Michigan’s 2009 Weatherization Assistance Program, then
use that random treatment as an instrumental variable, comparing energy use by
households that did and did not receive the encouragement. They find that weatherized homes do use less energy but that the savings are only about o ne-third of what
energy auditors predicted for those very same homes.10
One final approach, and the one most similar to the one I take here, is to examine
energy use by homes constructed before and after a particular change in a building code. Jacobsen and Kotchen (2013) compare the utility bills in 2 004–2006 for
homes in Gainesville, Florida, built just before and after the city tightened its building energy codes in 2002. They find that homes built after the change use 4 percent
less electricity and 6 percent less natural gas.11 One concern their paper cannot
address is that the new and old homes may differ in ways that are correlated with
energy consumption. Their data contain no information about the number or characteristics of the homes’ occupants, for example, and their strategy cannot distinguish
building age from year of construction. All the homes subject to the new building
codes were recently constructed, conflating building vintage with building age.12
Before detailing the three approaches I take here, below I describe the data and
discuss two issues: the confounding effects of building vintage, age, and survey
year, and trends in electrification of heat and hot water.
II. Data and Two Often-Overlooked Issues
For this project, I use two main sources of data. The RASS contains detailed
information about the buildings, occupants, and energy consumption of more than
22,000 California households in 2003 and another 26,000 in 2009. The data come
from two-stage mailed surveys to representative samples of California households,
with follow-up telephone and in-person interviews for nonrespondents, and with
an online version available in 2009. I focus on detached s ingle-family houses with
complete information about key building and occupant characteristics where I could
match the household to annual energy billing data provided by CEC. That leaves
about 7,200 houses in 2003 and 6,800 in 2009.13
For the second empirical approach—examining the sensitivity of monthly energy
use to outdoor temperatures—I go back to the RASS data and match households by
zip code to nearby weather station data maintained by the National Oceanographic
and Atmospheric Administration (NOAA). The third empirical strategy compares
California with other US states. For that, I turn to the RECS, a nationally representative survey of household characteristics and energy use conducted every three to
four years by the US Department of Energy.
Also see Allcott and Greenstone (2015).
Below I discuss Kotchen’s (2015) follow up in which the electricity savings have disappeared but the gas
savings have grown.
12
Costa and Kahn (2010) take a similar approach using a cross section of homes in one California county, Kahn,
Kok, and Quigley (2014) explore a nationally representative cross section of commercial buildings, and Kahn, Kok,
and Liu (2016) examine hotels in particular.
13
See online Appendix Table A1 for details about the sample construction. More information about the RASS
can be found at www.energy.ca.gov/appliances/rass. I obtained access to the utility billing data for RASS households by an open records request to the California Energy Commission.
10
11
2874
october 2016
THE AMERICAN ECONOMIC REVIEW
Table 2—Selected Characteristics of California Single-Family Homes
RASS
RECS 1993–2009
AHS (CA)
2003
(1)
2009
(2)
California
(3)
US
(4)
23.76
(12.54)
53.14
(27.90)
1.84
(0.81)
3.26
(0.86)
0.28
26.70
(13.63)
43.09
(23.74)
1.93
(0.84)
3.32
(0.86)
0.23
27.44
(18.01)
48.31
(31.86)
2.19
(1.15)
6.27
(1.64)
44.74
(26.81)
53.75
(61.47)
2.57
(1.37)
6.68
(1.79)
0.16
0.15
16.3
(14.5)
2.90
(1.50)
97.00
(64.45)
0.25
(0.66)
0.44
(0.75)
0.56
18.7
(14.7)
2.84
(1.49)
92.04
(60.43)
0.22
(0.65)
0.54
(0.79)
0.60
Disabled resident
0.094
0.110
0.156
Household head black
0.032
0.030
0.048
Household head Latino
[“Hispanic” in AHS]
0.122
0.134
0.216
Own home
0.92
0.93
0.80
0.88
0.78
Central AC
0.48
0.57
0.42
0.60
0.58
Room AC
0.10
0.11
Refrigerators
1.29
(0.50)
1.38
(0.55)
1.29
(0.49)
1.28
(0.50)
Observations
7,201
6,844
1,904
15,868
Variables
Annual electricity (MBTU)
Annual gas (MBTU)
Square feet (1,000s)
Bedrooms
[Total rooms in RECS]
Electric cooking
Remodeled
Years at address
Number of residents
Household income
[Thousand $2010]
Residents aged 0–5
Residents aged 65+
Household head graduated college
3.23
(1.74)
68.25
(40.57)
2.89
(1.46)
62.22
(37.10)
0.34
(0.65)
0.32
(0.64)
2011
(5)
3.32
(0.91)
0.28
13.8
(13.4)
3.04
(1.62)
102.5
(130.7)
0.40
(0.69)
0.48
0.13
14,692
Notes: RASS 2003 and 2009, single-family homes without electric heat or hot water. California homes in the
Residential Energy Consumption Survey (RECS) 1993–2009. American Housing Survey (AHS) 2011, detached
single-family homes in California. For gas, there were 6,391 and 5,967 observations in the 2003 and 2009 RASS,
respectively.
Table 2 provides some descriptive statistics for the RASS and RECS data. For
comparison, column 5 reports California values from the American Housing Survey
for those variables it has in common with the other two surveys. The AHS is conducted by the Census Bureau and may therefore suffer less from nonresponse bias.
In general, the values are similar, though the RASS seems to have higher incomes,
smaller household sizes, fewer minorities, and more homeownership. And even if
the RASS is not entirely representative, that would only bias the results here if
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2875
response rates differ systematically by both home vintage and energy use, i.e., if
residents of older houses are more likely to respond if they use less energy, while
residents of newer homes respond if they use more energy. But that seems unlikely.
A larger problem involves the relationship between home vintage and building age.
A. The Confounding Effects of Building Age, Vintage, and Survey Year
Even the best of the existing studies fail to account for one important determinant of energy efficiency: building age. The distinction is between building age,
or how old the building is at the time of the survey, and building vintage, or when
the building was constructed. A building’s vintage determines the stringency of the
regulations the builder faced, but in a cross section of data, building age and vintage
cannot be separately measured. All ten-year-old homes surveyed in 2009 were built
in 1999. If homes become draftier with age, researchers may spuriously attribute
newer homes’ lower energy use to stricter building codes. But repeated cross sections of data will contain ten-year-old homes built in different years under different
standards. That is one of the advantages of the empirical strategies that I take: I use
surveys constructed from repeated cross sections.
Recall the example from Gainesville, Florida. That study found energy use in 2004–
2006 to be 4 to 6 percent lower for homes built just after Florida’s 2002 building code
change.14 But that result confounds the age of the building with its vintage of construction.15 In particular, the post-2002 houses are all new and have new occupants,
and there might be many reasons why those homes and occupants use less energy. The
appliances are new, the air filters clean, the windows and doors seal well, and the occupants may still be acquiring some energy-using devices. In an earlier draft of this study,
I speculated that “something about the newness of the [Gainesville] homes led them
to use less energy, not the building codes when they were constructed.”16 Prompted
in part by that speculation, Kotchen (2015) revisits those Gainesville homes ten
years later and finds that the 4 percent drop in electricity use for homes built after the
2002 building code change has disappeared, but the 6 percent drop in natural gas use
has doubled. It appears that newly built homes differ in energy consumption because
they are new, whether or not they were constructed to meet stricter building codes.
Whatever the explanation, by using repeated cross sections of California houses, I
can distinguish between building age and vintage in several ways. First, CEC issued
its first energy efficiency building standards in 1978, and any newness effect will
have faded by now for houses built before and after those regulations took effect.
Second, I can control for the number of years the residents have lived at the address,
separating the effect of a new owner from the effect of a new building.
But one tricky remaining problem involves simultaneously controlling for the
year of the survey. In different years, homeowners could well consume different
amounts of energy. Appliances change and new models become available, energy
prices change, weather differs, and people adopt different patterns of energy use.
See Jacobsen and Kotchen (2013, Figure 3).
This confusion between age and vintage was central to a debate about immigrants’ wages back in the 1980s.
Chiswick (1978) and others interpreted cross-sectional differences in immigrants’ wages to be an age effect, but
Borjas (1985) showed it was a cohort effect. Recent cohorts of immigrants have had fewer skills.
16
Levinson (2014b, p. 7).
14
15
2876
october 2016
THE AMERICAN ECONOMIC REVIEW
50
Constructed 2000s
1990s
40
1980s
1970s
1960s
1950s
30
1940s
MBTU
’01
Pre ’40
’05 ’09
’97
’90
’87 RECS
20
’93
10
0
0
20
40
60
80
Building age
Figure 2. Household Electricity Use, Synthetic Cohorts by Construction Decade
Note: Single-family detached homes without electric heat or hot water, 1993–2009
But the three variables—vintage, age, and survey year—are linearly related. A
regression of energy use on home characteristics cannot include all three covariates.
Figure 2 illustrates the problem using seven cross sections of national RECS data
going back to 1987. From those data I constructed synthetic cohorts, grouping houses
by building age and decade of construction. The figure plots electricity use by building age, separately for each vintage of construction, controlling for no other characteristics. Each line in Figure 2 represents a different vintage of houses, by decade of
construction. Each dot on a line represents a different RECS, from 1987 through 2009.
Three features stand out. First, more recently constructed houses (higher lines)
use more electricity. As noted, that may be explained by other home characteristics
and is what this paper is in large part an attempt to explain. Second, the age-energy
profiles are upward-sloping for every vintage. That could be the result of homes
aging, or it could be the general time trend of increasing electricity consumption.
Third, the age-energy profile is steepest for the very newest houses, those constructed
in the 1990s and surveyed in the 1993 or 1997 RECS and those constructed in the
1980s and surveyed in the 1987 or 1990 RECS. This is unlikely to be the result of
general trends in electricity consumption because it affects new buildings differently
from old buildings.17 This third distinction is most likely a newness effect that could
easily be mistaken for the efficacy of the most recent building code.18
17
The slopes of the energy-age profiles are statistically indistinguishable for all but the newest buildings.
Figure A1 in the o nline Appendix has a version of Figure 2 that plots the residuals after regressing electricity use
on building and occupant characteristics. The plots of the synthetic cohorts’ residuals are all still upward sloping
and steepest for the newest houses, but they are all centered around zero. That is, homes built more recently do not
use any more or less electricity.
18
Note that each of the synthetic cohorts turns down in the final survey year. That phenomenon is observable in
national data, as the steadily growing pattern of household electricity consumption seems to have plateaued starting
around 2007.
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2877
The most common solution to the collinearity of age, cohort, and survey year
involves using theory to assume that one of those three variables is nonlinear.19 I
could assume a concave functional form for the effect of building age and assume
that the general year effects apply to all buildings equally. But each of those assumptions has different implications for the measured vintage effects, which proxy for the
changing building codes at the heart of this paper. I do not want those assumptions
or their effects to determine the outcome.
Instead, I control for general time effects with year-of-survey indicators, for vintage of construction with vintage indicators, and for length of occupancy by current
residents, but I do not control for building age. That means that the vintage indicators, which report residential energy consumption for houses constructed at different
times controlling for other characteristics, combine the vintage and age effects. In
this way I bias the results in favor of finding that more newly constructed houses
(which are also newer) use less energy. But because the newness effect fades over
time, that bias will be strongest in the most recent years and will stand out in the
patterns of vintage coefficients.
A building’s vintage also matters when accounting for a second overlooked issue:
the rising and falling trend in electrification of heat and hot water. If ignored, it could
seem as though homes built since California’s 1978 building codes use less electricity and more natural gas because of those codes, rather than because of trends in
home construction.
B. Electrification of Heat and Hot Water
The proportion of houses with electric space heat and hot water in California
(as opposed to gas heat or water), peaked in the late 1970s when 15 percent of
houses had electric heat, hot water, or both. After that, the trend reversed so that
very few houses built recently have electric hot water and almost none have electric
heat.20 Why? In the 1950s a consortium of utilities and appliance manufacturers
launched “Live Better Electrically” campaigns, granting allowances to homebuilders to construct a ll-electric homes throughout the United States.21 But by the 1980s,
the program had ended, along with popularity of a ll-electric homes. The pattern
has obvious implications for electricity use by building vintage. In what follows, I
drop from the sample the 7 percent of s ingle-family houses with electric heat or hot
water.22
All of this will be clearer with some results in hand, and so with those preliminary
caveats out of the way, the next three sections discuss each of the three empirical
approaches to assessing the energy savings from California’s building energy codes.
19
In the context of wages, Deaton (1985), Foster (1990), and Borjas (1994, 2013) have all discussed the difficulties of simultaneously controlling for age, cohort, and year.
20
In online Appendix Figure A2 I plot the shares of homes with electric heat and hot water, by year of construction. Both peak for houses built between 1978 and 1982, at 7 percent for heat and 13 percent for hot water.
21
Not-yet-Governor Ronald Reagan promoted the program on his show “General Electric Theater.” See Diane
Wedner, “The All-Consuming Bills of an All-Electric Home,” Los Angeles Times, August 13, 2001, http://articles.
latimes.com/2001/aug/13/news/mn-33663 (accessed August 12, 2016).
22
In the online Appendix, I report alternative specifications that include homes with electric heat or hot water.
The results are nearly identical.
2878
THE AMERICAN ECONOMIC REVIEW
october 2016
III. Strategy 1: Controlling for Building and Occupant Characteristics
The first approach is straightforward. I regress the log of annual household energy
use on occupant and building characteristics and a set of indicators for each of the
different construction vintages. If building codes have been effective, we should
expect houses constructed after California’s 1978 standards to be using less energy
today than houses built before the codes were enacted, controlling for other observable features of the houses and their occupants.
The basic specification is
θ
j ConstructEraj i + εi,
(1) ln (Ei ) = Xi β + ∑
j
where Ei is household i’s energy use, in MBTU, and Xi is a vector of house and
occupant characteristics, including survey year and region fixed effects.23 The
dependent variable is in logs because that specification has the lowest residual sum
of squares after a B
ox-Cox transformation, but linear versions yield qualitatively
similar conclusions. The goal of estimating (1) is to examine whether the construction era coefficients (the θs) are smaller or more negative for later eras than for
earlier eras, which would suggest that all else equal, houses constructed under more
stringent building codes use less energy today.
I start with electricity because patterns of electricity use are often cited as evidence for the success of California’s standards (Rosenfeld and Poskanzer 2009)
and because electricity generation has become a focus of energy and environmental
policy now that greenhouse gas emissions are a concern. Table 3 presents the results
for electricity using the RASS survey and a full set of control variables in column 1.
Larger houses with electric ovens and more and wealthier occupants in hotter places
use more electricity.24 Importantly, residents who have lived at the house longer use
more electricity. Since length of occupancy is correlated with house age and vintage, any analysis that fails to control for occupancy could falsely attribute energy
savings to newer buildings that simply have newer occupants. And homeowners use
less than otherwise similar renters, which makes sense given that renters typically
aren’t responsible for choosing their appliances or insulation, and sometimes aren’t
responsible for the utility bills (Levinson and Niemann 2004; Myers 2015). Year
and region indicators absorb any variation over time or across parts of California,
including differences in climate and energy price schedules.
The important coefficients are those on the construction era dummies. To
help understand them, rather than display them in Table 3 I have plotted them in
Figure 3. The first line in Figure 3 (dark dots) is from a version of equation (1) that
excludes all controls except for the construction era dummies. The pattern of those
23
The construction era indicators correspond to the 12 vintages identified in the RASS, dropping the one for
homes built before 1940. The year and geographic fixed effects absorb differences in weather and energy price
schedules, among other characteristics that vary across survey years or regions of the state. (Price differences
would be relevant only if they differed systematically by vintage of house construction.) The idea is to compare
energy use by otherwise similar houses in the same place during the same year that just happen to have been built
at different times.
24
A degree-day is the difference between the average of the daily maximum and minimum temperatures and
65°F. A heating degree-day occurs when that average temperature is less than 65°, and a cooling degree-day when
it is greater than 65°.
VOL. 106 NO. 10
2879
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
Table 3—Annual Energy Use by California Households in the RASS Survey
Electricity
Dependent variable: ln(annual MBTUs)
Cooling degree-days (100s)
Heating degree-days (100s)
ln (square feet)
Bedrooms
Electric cooking
Remodeled
ln (years at address)
ln (number of residents)
ln (household income)
Household head graduated college
Disabled resident
Own home
Refrigerators
Room AC
Central AC
Central AC × sq. feet (1,000s)
Age of central AC
Full controls
(1)
Age of AC
(2)
0.021
(0.003)
−0.001
(0.004)
0.266
(0.016)
0.025
(0.007)
0.036
(0.007)
0.027
(0.010)
0.026
(0.005)
0.243
(0.009)
0.095
(0.008)
−0.051
(0.007)
0.143
(0.009)
−0.055
(0.017)
0.191
(0.007)
0.058
(0.011)
0.141
(0.029)
0.030
(0.009)
0.021
(0.003)
−0.001
(0.004)
0.266
(0.016)
0.025
(0.007)
0.036
(0.007)
0.026
(0.010)
0.027
(0.005)
0.243
(0.010)
0.095
(0.008)
−0.051
(0.007)
0.142
(0.009)
−0.055
(0.017)
0.191
(0.007)
0.058
(0.011)
0.153
(0.031)
0.030
(0.009)
−0.0012
(0.0005)
Home heater age
Year home built
Observations
R2
Natural gas
Full controls
(3)
−0.016
(0.006)
0.029
(0.007)
0.377
(0.021)
0.016
(0.008)
−0.037
(0.007)
−0.010
(0.013)
0.011
(0.004)
0.131
(0.013)
0.064
(0.008)
−0.034
(0.011)
0.080
(0.017)
−0.078
(0.025)
Age of heater
(4)
−0.016
(0.006)
0.028
(0.007)
0.387
(0.025)
0.009
(0.007)
−0.032
(0.008)
−0.011
(0.013)
0.010
(0.004)
0.138
(0.013)
0.058
(0.007)
−0.038
(0.012)
0.083
(0.016)
−0.087
(0.024)
0.0017
(0.0005)
Coefficients displayed in Figures 3 and 4
14,045
0.437
14,045
0.437
12,358
0.266
11,644
0.267
Notes: RASS 2003 and 2009, standard errors clustered by county. Regressions also include 13 climate zone indicators, kids, seniors, black, Latino, and an indicator for the 2009 survey year. Excluded construction category is homes
“Built pre-1940.” Full set of coefficients in online Appendix Tables A2 and A3.
c oefficients mimics the raw data plotted in Figure 1. The lower line (hollow symbols) plots coefficients from the regression in column 1 of Table 3. Controlling for
other characteristics, new houses do not appear to consume statistically significantly
less electricity than houses built earlier. Only the very newest houses built after 2005
have coefficients statistically significantly lower than houses built before 1978. And
2880
THE AMERICAN ECONOMIC REVIEW
october 2016
0.6
No controls
Full controls
ln(MBTU)
0.4
0.2
0.0
−0.2
40
-19
Pre
9
94
–1
40
19
9
95
–1
50
19
9
4 7 2
2
7 0 4 8
97 97 98
99
99 00 00 00
–1 –1 –1
–1
–1 –2 –2 –2
70 75 78
83
93 98 01 05
19 19 19
19
19 19 20 20
96
–1
60
19
Year house constructed
Figure 3. Residential Electricity Use in California, Controlling for Characteristics
Note: RASS 2003 and 2009, single-family detached California homes without electric heat
or hot water
as we have seen from Figure 2 and the two Gainesville papers, very new homes
use less electricity for reasons likely unrelated to building codes: new appliances,
well-sealed windows, etc.
Because the dependent variable in column 1 is in logs, the construction era coefficients can be interpreted as percentage differences relative to houses built before
1940. If we ignore the newest houses, houses built after 1990 appear to use 10 to
15 percent less electricity than houses built in the 1970s. But that difference is not
statistically significant and is far smaller than projected at the time the building
codes were enacted.25
Why aren’t otherwise similar houses built recently saving more electricity? One
possible explanation is that homeowners may have retrofitted the p re-1978 houses,
upgrading the insulation, windows, and appliances. The RASS data lack detailed
information about whether such energy upgrades were done, but they do indicate
whether the house was ever remodeled in any way and the age of its central a ir conditioner. Both measures work in the opposite direction. The “remodeled” coefficient
in Table 3 is positive. Houses that have been remodeled use more electricity, all else
equal. And the a ir conditioner age, added in column 2, has a negative coefficient,
suggesting that people who buy new a ir conditioners buy larger ones or use them
more.
Columns 3 and 4 of Table 3 estimate versions of equation (1) with natural gas
as the dependent variable. As before, larger houses with more occupants who are
25
Online Appendix Table A2 contains full set of coefficients and some additional specifications, including one
where I include the 7 percent of homes with electric heat and hot water. The construction era coefficients in those
specifications show even less variation.
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2881
0.0
ln(MBTU)
−0.1
−0.2
−0.3
No controls
Full controls
−0.4
40
-19
Pre
49
19
0–
4
19
59
19
0–
5
19
4 7 2
7 0 4 8
2
97 97 98
99 00 00 00
99
–1 –1 –1
–1 –2 –2 –2
–1
70 975 978
93 998 001 005
83
9
9
9
1 1 1
1 1 2 2
1
69
19
0–
6
19
Year house constructed
Figure 4. Residential Natural Gas Use in California, Controlling for
Characteristics
Note: RASS 2003 and 2009, single-family detached California homes without electric heat
or hot water
wealthier and have lived there longer use more gas. Of course, houses use less gas if
they have electric stoves, and more gas in cold weather rather than hot. And houses
with older heaters use more gas—about 1.7 percent for every ten years of heater age.
Figure 4 plots the construction era coefficients from column 3 of Table 3. As in
Figure 3, the top line (dark dots) plots the construction era coefficients when those
are the only control variables, and the pattern mimics that seen in Figure 1. The
lower line (hollow symbols) plots the coefficients from Table 3. The figure depicts
a steady decline in natural gas consumption. Houses built in each successive era
after the 1950s are using less natural gas today. And controlling for other house and
occupant characteristics increases that decline in gas consumption.
One interpretation of Figure 4 is that the building codes enacted first in 1978 and
strengthened every few years thereafter have worked. But that interpretation would
come with two caveats. First, although the confidence intervals are large, the trend
appears to have begun before 1978. Something reduced gas consumption by otherwise similar houses built in the 1970s relative to those built earlier, and the building
codes enacted starting in 1978 cannot be credited with that part of the trend. Second,
even the largest p ost-1978 declines in gas use do not exceed 20 or 25 percent, short
of the reduction anticipated when the building codes were enacted.
Not surprisingly, versions of this analysis for total household energy use look like
a combination of Figures 3 and 4, depending on the units used to combine electricity
and natural gas. If I sum the two according to their cost to the homeowners, energy
expenditures are about 10 percent lower for houses built in the 1980s and 1990s,
relative to those built before 1978, controlling for other building and occupant
characteristics. If I sum the two by carbon content, houses built more recently use
2882
THE AMERICAN ECONOMIC REVIEW
october 2016
15 percent less carbon. And, if I sum the two by heat content (BTUs), the newer
houses use almost 20 percent less energy. All of this suggests that there may be some
pollution and cost savings from the building codes, though they fall short of the
savings projected when the regulations were enacted and come mostly from natural
gas rather than electricity.26
The simple analysis so far does ignore one potential problem: selection by residents into e nergy-efficient homes. If people who really like a ir conditioning and
heating choose to live in energy-efficient newer houses, while people who prefer
open windows or are often away from home choose less efficient older houses, that
might explain the lack of observed differences in energy use.27 The next two strategies attempt to address the potential selection issue. The idea is to look at energy
consumption by house vintage, as in strategy 1, but to differentiate the effect along
two dimensions that are unlikely to be associated with residents’ preferences for
energy: unseasonably hot or cold temperatures, and comparisons between California
and other states.
IV. Strategy 2: Temperature Effects
Most of the building code costs described in Table 1 involve weatherization. Half
of the $8,000 cost for Sacramento houses was for insulation and window glazing,
and another one-sixth was for other building envelope and space cooling or heating
features. If the codes save energy as projected, energy use should increase less on
hot or cold days for buildings constructed under more stringent standards. I cannot
entirely eliminate the possibility that homeowners who are likely to use more a ir
conditioning when the weather gets hot, or more heat when the weather is cold, are
more likely to select newer homes because they are e nergy-efficient. But if that hot
or cold weather is unexpected or unseasonable for a region, that may mitigate the
selection problem. This is a type of difference-in-differences approach, where the
first difference is between houses of different vintages, and the second is between
houses experiencing extreme or mild temperatures. Strategy 1 examined the first difference—between houses of different vintages. Here I examine how that difference
changes when the outdoor temperature varies.
This strategy builds on Chong (2012), who matches tax assessment data with
electric utility billing records in Riverside, California. He finds that homes use more
electricity during hot months and, surprisingly, that post-1978 homes have an even
larger hot-weather electricity increase. Chong controls for home size in square feet,
but because he uses census block data, he has little other information about specific
homes’ characteristics or residents’ demographics. New homes have more occupants with higher incomes, and that could in theory explain his finding. He also does
not examine home heating in cold weather or study unseasonable temperatures.
26
Natural gas has lower prices and carbon content per BTU than electricity, and most of the post-1978 reductions come from natural gas, which is why the cost and pollution savings are lower than the energy savings. The
combined results can be seen in Appendix Figure A3 and Table A4.
27
Note however, that this selection story conflicts with a major justification for the building codes in the first
place: that housing markets underprovide energy efficiency because home purchasers cannot see or will not pay for
the up-front costs. If homebuyers don’t value energy efficiency, it’s hard to imagine that unobserved selection by
homeowners can explain the gap between projected and observed savings.
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2883
Panel A. Electricity
1,000 BTUs/month
3,000
2,000
1,000
Constructed after 1980
Constructed before 1980
0
1,000 BTUs/month
Panel B. Natural gas
8,000
6,000
4,000
2,000
0
Jan.
Feb.
Mar.
Apr.
May
Jun.
Jul.
Aug.
Sep.
Oct.
Nov.
Dec.
Figure 5. Seasons and Energy
Note: RASS 2003 and 2009, single-family detached California homes without electric heat
or hot water
To replicate Chong, but with a full set of building and occupant characteristics,
I turn to the monthly billing data in the RASS and match those with the number
of heating and cooling d egree-days at nearby weather stations each month.28 For
each RASS house, I know the latitude and longitude of the population-weighted
centroid of that house’s zip code, from the Census Bureau. I draw a 3 0-kilometer
circle around that centroid and calculate a weighted average of the reported
weather variables for all the NOAA weather stations inside that circle and where
the weights are the inverse of the distance from the weather station to the zip code
centroid.
Figure 5 plots average energy use by month for houses built before and after
1980. The top panel shows electricity. Houses constructed after 1980 use more electricity in every month, and during summer months electricity use appears to grow
relatively more for new houses than for old. But again, new houses are larger, in
hotter places, etc., and in what follows I control for those characteristics. The bottom panel plots natural gas, for which there appears to be little difference between
the new and old houses, in either the monthly averages or the increase during winter
months.
28
Billing dates do not correspond exactly to months because billing dates are not typically the first or last day
of the month. To match utility bills to monthly temperature data I “calendarized” the billing data by assigning usage
proportionally to the months spanned by each utility bill. The 2003 RASS data I received from CEC had already
been calendarized this way; I followed its method for 2009.
2884
THE AMERICAN ECONOMIC REVIEW
october 2016
To examine whether energy use increases during hot or cold weather less for
newer houses, controlling for other house characteristics, I estimate versions of
(2) ln (Eim
) = Xi β + αCD Di m + ∑
θ
j ConstructEraji
j
+ ∑
π
j (CDDim
× ConstructEraji ) + γm
+ εim .
j
Subscript irefers to houses and mrefers to months. Again, X
i includes year and
county fixed effects.29 For the regressions where natural gas is the energy measure
on the left-hand side, I replace cooling degree-days (CDDi m) on the right-hand side
of equation (2) with heating d egree-days (HDDi m).30 As in the first approach, the
semi-log specification in (2) has the best Box-Cox fit, but linear versions of (2) have
qualitatively similar conclusions.
This is basically the same regression as in equation (1), except that instead of one
annual observation on energy and weather for each house, I use monthly observations spanning at least a year, the temperature measurements are the monthly CDDs
and HDDs in the zip code in which the house is located, and I interact those monthly
temperature measurements with the construction era indicators. The coefficients πj
on the interaction between CDDs and construction era estimate whether during hot
months, houses of vintage juse more electricity.
Table 4 shows estimates of equation (2) for electricity. To conserve space, I report
only the interactive terms (the πjs) and a few other key coefficients. (Full results
are in online Appendix Table A5.) Column 1 of Table 4 contains estimates of equation (2) with the full set of home characteristics and two measures of monthly temperature: the actual number of CDDs in the house’s zip code for the month, and
the average number of CDDs in that zip code during that month for the past ten
years. This average can be thought of as the climate a homeowner can expect in
any particular month, compared with the climate actually experienced, in the first
row. In other words, if August is typically a hot month in Sacramento, that effect is
absorbed by the average monthly CDD in row 2. But if August 2009 is particularly
hot in Sacramento, that effect is identified by the CDD coefficient in row 1 and its
interactions with the construction era indicators.
As before, rather than display all of the interaction coefficients in the table, I have
plotted them. The top panel of Figure 6 plots the interaction coefficients (the πj s)
from column 1 of Table 4. The height of the markers can be thought of as the weather
sensitivity of electricity use for houses built in different eras. And those coefficients
generally increase, rather than decrease, with building vintage. On average, each
extra 10 CDDs in a month adds about 2 percent to electricity use above that of the
pre-1940s houses (the omitted category), and an additional 0.5 percent or more to
houses built post-1990. Electricity use in newer houses rises faster when the temperature increases, even controlling for observable house and occupant characteristics, consistent with Chong (2012).
As with equation (1), the year, month, and county fixed effects here absorb differences in price schedules.
Versions of equation (2) that include both HDD and CDD yield results with larger confidence intervals but
nearly identical coefficients.
29
30
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2885
Table 4—Monthly Electricity Use and Monthly Cooling Degree-Days (CDD)
in California
Only homes with central AC
Dependent variable:
ln(monthly electricity)
CDD per month (10s)
Full controls
(1)
Age of AC
(2)
AC age × CDD
(3)
0.0005
(0.0030)
0.0208
(0.0031)
0.0014
(0.0032)
0.0185
(0.0040)
−0.0023
(0.0005)
0.0003
(0.0044)
0.0185
(0.0040)
−0.0017
(0.0007)
Avg. monthly CDD in zip code
Age of AC
CDD × year home built
Coefficients displayed in Figures 6 and 7
Observations
R2
265,599
0.368
135,840
0.323
135,840
0.323
Notes: Standard errors clustered by county. Other controls: ln(square feet), bedrooms, electric cooking, remodeled, ln(years at address), ln(residents), ln(household income), residents
aged 0–5, residents aged 65–99, household head graduated college, disabled resident, household head black, household head Latino, own home, refrigerators, central AC, room AC, and
fixed effects for survey year, vintage of construction, county, and month. Column 3 includes
interactions between age of air conditioner and CDDs. Full set of coefficients in online
Appendix Table A5.
ln(BTU) per 10 CDDs
Panel A. Electricity
0.015
0.01
0.005
0
−0.005
−0.01
ln(BTU) per 10 HDDs
Panel B. Natural gas
0.01
0.005
0
−0.005
40
-19
Pre
1
9
94
–1
0
94
1
9
95
–1
0
95
1
92
74 77 82
97 00 04 08
19
19 19 19
19 20 20 20
3–
0– 75– 78–
3– 98– 01– 05–
8
7
9
19
19 19 19
19 19 20 20
9
96
–1
0
96
Year house constructed
Figure 6. Energy Use per Degree-Day
Note: RASS 2003 and 2009, single-family detached homes
Why might electricity use increase with temperature more in newer houses than
in older ones, controlling for other house and occupant characteristics? One possible
omitted variable is tree shade. Older houses may be surrounded by taller trees that
2886
THE AMERICAN ECONOMIC REVIEW
october 2016
reduce air conditioning demand by providing shade. Maher (2013) estimates that
homes in Florida that have had nearby trees removed use 3 percent more electricity
the following year, and Pandit and Laband (2010) estimate that homes with average
shading use 4 percent less summertime electricity than homes with no tree shade.
Another possibility is remodeling. Perhaps the older houses have new,
energy-efficient
air conditioners. The regressions in Table 4 contain the same
“remodel” indicator as in the annual specifications in Table 3, and as before it has a
positive and statistically significant coefficient. But remodeling is broadly defined
in the RASS, and in this weather-sensitivity analysis in Table 4 I can distinguish the
vintage of the air conditioner from the vintage of the house, using only the 52 percent of houses with central air conditioning.
Columns 2 and 3 of Table 4 add a variable for the age of the air conditioner,
reported in the RASS data in six vintage categories, ranging from less than one year
to more than 35 years. I converted that to a continuous measure by taking midpoints
of those ranges. As with the annual data in Table 3, here in column 2 of Table 4 each
year of a ir conditioner age reduces average electricity use. Older a ir conditioners
must be either smaller or used less.
In column 3 of Table 4 I interact the measure of CDD per month with the air
conditioner vintage categories:
θ
j ConstructEraj i
(2a) ln (Ei m) = Xi β + αCDDi m + ∑
j
+ ∑
π
j (CDDi m × ConstructEraj i)
j
+ ∑
ω
j (AC agej i) + ∑
γ
j (CDDi m × AC agej i) + γm
+ εim .
j
j
Those coefficients (the γ s) are displayed in the top panel of Figure 7. Electricity
use does not increase less during hot weather for houses with new air conditioners. If anything, it increases more, though the confidence intervals are large. More
importantly, controlling for air conditioner age does not change the house vintage
interactions (the πj s). Upgraded a ir conditioners in older houses do not explain why
those older houses do not use more electricity in hot months.
Turning from electricity to natural gas, in Table 5 I estimate versions of equation (2) for natural gas use and heating degree-days. Column 1 shows the results
controlling for other house and occupant characteristics, and the bottom panel of
Figure 6 plots the interaction coefficients. The trend seems to be declining, with
houses built after 1978 using less natural gas per HDD than those built in the 1950s
through the 1970s. But that decline is statistically insignificant and small. Houses
in the RASS sample experienced an average of 200 January HDDs, with a standard
deviation of 90. If we look at just the point estimates in Figure 6, ignoring their statistical significance, newer houses use about 5 percent less gas than pre-1978 houses
in response to a one standard deviation increase in HDDs.
Perhaps here remodeling explains the small energy savings in newer houses. The
remodeled indicator (reported in online Appendix Table A6) is small and statistically insignificant. In column 2 of Table 5 I include the age of the house’s main
VOL. 106 NO. 10
LEVINSON: HOW MUCH ENERGY DO BUILDING ENERGY CODES SAVE?
2887
ln(BTU) per 10 CDDs
Panel A. Electricity
0.015
0.01
0.005
0
−0.005
−0.01
>30
14–30
9–13
4–8
1–3
Purchase answer to see full
attachment