PNNL-22182
Residential Lighting End-Use
Consumption Study: Estimation
Framework and Initial Estimates
WR Gifford
1
ML Goldberg
2
PM Tanimoto
2
DR Celnicker
1
ME Poplawski
3
December 2012
Prepared for
the U.S. Department of Energy
under Contract DE-AC05-76RL01830
Prepared by
DNV KEMA Energy and Sustainability
Pacific Northwest National Laboratory
1
DNV KEMA Energy and Sustainability, Fairfax, Virginia
2
DNV KEMA Energy and Sustainability, Madison, Wisconsin
3
Pacific Northwest National Laboratory, Portland, Oregon
Executive Summary
The U.S. DOE Residential Lighting End-Use Consumption Study is an initiative of the U.S.
Department of Energy’s (DOE’s) Solid-State Lighting Program that aims to improve the understanding of
lighting energy usage in residential dwellings. The study has developed a regional estimation framework
within a national sample design that allows for the estimation of lamp usage and energy consumption 1)
nationally and by region of the United States, 2) by certain household characteristics, 3) by location
within the home, 4) by certain lamp characteristics, and 5) by certain categorical cross-classifications
(e.g., by dwelling type AND lamp type or fixture type AND control type).
The lighting estimates presented in this report leverage several recent national and regional datasets,
linking lamp usage from end-use metering studies with household characteristics and lighting inventory
profiles that are anchored to a robust regionally stratified national sample design. The lighting usage
measures were estimated using a “bottom-up” methodology, in that lamp power, hours-of-use (HOU), and
energy consumption estimates were generated at the lamp level and aggregated up to various levels of
analysis. It should be noted that the statistical model for lamp usage came from a single regional study
that has not yet been calibrated for other regions of the United States.
1
For many regions, neither a local
study nor direct reporting in a national survey was available for use in this analysis, so extrapolations
were made based on the information known from neighboring or nearby regions. The available lighting
inventory data available from the South census region were noticeably limited. Lighting inventory data
averaged across all regions were assigned to homes in locations without regionally specific data.
This study produced lighting estimates based on existing data. However, the estimation framework
was designed to make straightforward use of new data collected under similar protocols. For example, if a
state or regional organization conducted a lighting study using protocols for the collection of household
characteristics, lighting inventories, and/or the end-use metering of fixtures that would support linkages of
the collected data to the data sources being used in this study, then the new data could be easily
incorporated into the developed estimation framework. Lighting usage estimates could then be updated,
resulting in improved regional and possibly national accuracy. The estimates presented in this report
include a validation of the accuracy achieved in California using the described process for linking newly
collected data. Updates to this study will be considered if enough new data meeting the described pre
-
conditions and funding for its analysis were to become available.
Figure ES.1 through Figure ES.4 highlight the variation in estimated regional lighting usage across
the United States. Note that states with estimates aggregated from homes without regionally specific
lighting inventory data are highlighted in the figures. The estimated daily usage per lamp averaged 1.6 hr
for all lamps in the United States. Regionally, average estimated daily usage per lamp in households
varied between 1.4 and 1.6 hr. Average estimated HOU per lamp were lowest in Missouri and Virginia
(<1.5 hr per day) and highest in Massachusetts, New York, Texas, Oklahoma, Arkansas, and Louisiana
(>1.6 hr per day).
1
Calibrating this lighting usage model with end-use data collected in other regions will be the primary objective of
potential future updates to the U.S. DOE Residential Lighting End-Use Consumption Study.
iii
* Note: Lighting inventory data for this state or its neighbor was not available.
Figure ES.1. Regional Variation in Average Daily HOU per Lamp
* Note: Lighting inventory data for this state or its neighbor was not available.
Figure ES.2. Regional Variation in Average Lamp Power (W)
iv
* Note: Lighting inventory data for this state or its neighbor was not available.
Figure ES.3. Regional Variation in Average Number of Lamps per Household
* Note: Lighting inventory data for this state or its neighbor was not available.
Figure ES.4. Regional Variation in Average Annual Lighting Energy Usage per Household
v
The average lamp power in a region is largely driven by the household preference for compact
fluorescent versus incandescent lamps. The United States, as a whole, averaged 47.7 watts (W) per lamp.
Figure ES.2 shows a clear regional association with average lamp power, with the Midwest showing the
highest and the Northeast the lowest average. Several states in the Northeast averaged less than 43 W per
lamp, led by New York with 40.5 W per lamp. Illinois had the highest average, at 53.5 W per lamp,
followed by several other Midwestern states averaging at least 53 W per lamp. Note that the lamp power
assigned to many states in the South census region is simply the U.S. average, given the lack of available
lighting inventory data for any of the states in this region.
Figure ES.3 shows the regional variation in the average number of lamps per household. In large part,
varying home sizes drives this variation. For example, California and New York contain a higher
concentration of multi-family households than Wyoming, where, on average, larger single-family
residences are more typically found. Regional variation in the number of lamps per lighting space type
can also impact these household estimates. For example, the estimated number of lamps per Living Room
varies by almost a factor of two across the United States, from 4.1 in Massachusetts to 7.9 in Illinois. The
accuracy of this impact is dependent on the availability and statistical quality of regional lighting
inventory data. Variations in estimates for the South census region states with lighting inventories
assigned according to the U.S. average are likely driven more by home size than states with regionally
representative lighting inventory data.
Figure ES.4 shows average annual lighting energy consumption per residence – reflecting average
HOU, lamp power, and number of lamps. Massachusetts, New York, and California had the lowest
household lighting energy consumption, each averaging fewer than 1,500 kWh per home per year. Idaho,
Montana, Utah, Wyoming, Missouri, and Arizona consumed the most household lighting energy, each
averaging over 2,100 kWh per home per year. Overall, the United States averaged just over 1,700 kWh
per home per year for lighting.
Figure ES.5 shows average usage estimates for select room types in U.S. households. Usage varies
significantly. Exterior lamps average close to 3 hr of use per day while hallway lamps average less than
1 hr of use per day. Lamps in bedrooms, bathrooms, living rooms, and kitchens consume the most energy,
on average, of all spaces within a home.
vi
Figure ES.5. Average Daily Lamp Usage and Energy Consumption, by Lighting Space Type
1
1
Note that estimates are for all instances of a lighting space type, not per instance.
vii
Acknowledgments
The authors would like to thank the following individuals, who contributed to this study in various
ways, including: review of the method, results and report; assistance obtaining data used in the analysis;
insight into data used in the analysis; and guidance in structuring the framework such that it could be
extended to incorporate additional data and improve estimation accuracy:
• Dan Chwastyk – Navigant
• Ed Cureg – U.S. Energy Information Administration
• Kelly Gordon – Pacific Northwest National Laboratory
• Marc Ledbetter – Pacific Northwest National Laboratory
• Thomas Leckey – U.S. Energy Information Administration
• Hiroaki Minato – U.S. Energy Information Administration
• Michael Myer – Pacific Northwest National Laboratory
• Eileen O’Brien – U.S. Energy Information Administration
• Lisa Wilson-Wright – Nexus Market Research
The authors would also like to thank the following organizations that sponsored collection of certain
data used in the study and granted their permission to use this data in the development of the estimation
framework and baseline estimates:
• Ameren IU (Illinois)
• Ameren UE (Missouri)
• California Public Utilities Commission (CPUC)
• ComEd (Illinois)
• Connecticut Energy Efficiency Board
• Consumers Energy (Michigan)
• Dayton Power and Light
• Maryland Public Service Commission (EmPower Program)
• Massachusetts ENERGY STAR Lighting Program Administrators
• National Grid Rhode Island
• New York State Energy Research and Development Authority (NYSERDA)
• Salt River Project
• Wisconsin Public Service Commission
ix
CV
Acronyms and Abbreviations
AHS American Housing Survey
ANCOVA analysis of covariance
CA RLMS California Residential Lighting Metering Study
CFL compact fluorescent light
CPUC California Public Utilities Commission
coefficient of variation
DOE U.S. Department of Energy
EIA U.S. Energy Information Administration
HOU hours of use
HUD U.S. Department of Housing and Urban Development
IOU Investor-Owned Utility
LMC Lighting Market Characterization
NMR Nexus Market Research Group, Inc.
PNNL Pacific Northwest National Laboratory
RECS (EIA) Residential Energy Consumption Survey
xi
Contents
Executive Summary .............................................................................................................................. iii
Acknowledgments................................................................................................................................. ix
Acronyms and Abbreviations................................................................................................................ xi
1.0 Introduction................................................................................................................................... 1.1
2.0 Data Sources ................................................................................................................................. 2.1
2.1 California Residential Lighting Metering Study.................................................................. 2.1
2.2 Residential Energy Consumption Survey ............................................................................ 2.4
2.3 American Housing Survey ................................................................................................... 2.5
2.4 Nexus Market Research Group Multi-State CFL Modeling Study...................................... 2.5
2.5 Consideration of Additional Data Sources........................................................................... 2.7
3.0 Methodology................................................................................................................................. 3.1
3.1 Estimation Framework Characteristics ................................................................................ 3.1
3.1.1 Household Characteristic Variables .......................................................................... 3.1
3.1.2 Lighting Inventory Variables .................................................................................... 3.3
3.1.3 Regional Variables .................................................................................................... 3.4
3.2 Input Dataset Preparation ..................................................................................................... 3.5
3.2.1 2009 RECS Microdata .............................................................................................. 3.5
3.2.2 2009 AHS Microdata ................................................................................................ 3.5
3.2.3 2009-2010 Multi-State CFL Study Microdata .......................................................... 3.6
3.3 Combining Datasets ............................................................................................................. 3.8
3.3.1 Extending RECS Housing Units with AHS Data...................................................... 3.8
3.3.2 Extending RECS Housing Units with Multi-State CFL Study Data......................... 3.10
3.3.3 Estimation Framework Summary.............................................................................. 3.10
3.4 Lighting Estimates................................................................................................................ 3.13
3.4.1 Lamp Usage and Energy Consumption..................................................................... 3.14
3.4.2 Seasonal Variation..................................................................................................... 3.16
3.4.3 Number of Fixtures and Lamps................................................................................. 3.16
4.0 Initial Estimation Highlights......................................................................................................... 4.1
4.1 Total Energy Consumption .................................................................................................. 4.1
4.2 Lamp-Level Attributes ......................................................................................................... 4.3
5.0 Accuracy and Validity of the Estimates ....................................................................................... 5.1
5.1 Comparison with CA RLMS Estimates ............................................................................... 5.1
5.2 Statistical Precision of Estimates ......................................................................................... 5.2
5.3 Sources of Bias and Variability Introduced in the Estimates............................................... 5.3
5.4 Opportunities for Improving Estimates................................................................................ 5.4
xiii
Figures
ES.1 Regional Variation in Average Daily HOU per Lamp................................................................ iv
ES.2 Regional Variation in Average Lamp Power (W)....................................................................... iv
ES.3 Regional Variation in Average Number of Lamps per Household............................................. v
ES.4 Regional Variation in Average Annual Lighting Energy Usage per Household ........................ v
ES.5 Average Daily Lamp Usage and Energy Consumption, by Lighting Space Type...................... vii
2.1 The DENT Instruments LIGHTING Logger Used to Collect End-Use Metering Data ............. 2.2
4.1 National Estimates of Average Daily HOU per Lamp, by Lighting Space Type and
Month .......................................................................................................................................... 4.9
xiv
Tables
2.1 ANCOVA Model Variables Used in the CA RLMS .................................................................. 2.3
2.2 2009 RECS-Reportable Domains................................................................................................ 2.4
2.3 Multi-State CFL Study Clients.................................................................................................... 2.6
2.4 Multi-State CFL Regional Studies by Year, with Number of Household Characteristics
and Lighting Inventory Records.................................................................................................. 2.6
3.1 Housing Unit Characteristic Variables in the Estimation Framework........................................ 3.2
3.2 Occupant Demographic Variables in the Estimation Framework............................................... 3.2
3.3 Lighting Space Types in the Estimation Framework.................................................................. 3.3
3.4 Lamp Characteristics in the Estimation Framework ................................................................... 3.3
3.5 Fixture Characteristics in the Estimation Framework................................................................. 3.4
3.6 Lighting Inventory Aggregations in the Estimation Framework ................................................ 3.4
3.7 Recoding of Lighting Space Types for the 2009 Four-State Study Datasets.............................. 3.7
3.8 Fixture Type Distribution, Before and After Imputation in Michigan – Consumers
Study Territory ............................................................................................................................ 3.8
3.9 Average Lamp Power by Lamp Type, Before and After Imputation in 2009 Four-State
Study Territories.......................................................................................................................... 3.8
3.10 Top Five Lighting Space Configurations for a RECS ID 10284 ................................................ 3.9
3.11 Summary of Data Sources by State............................................................................................. 3.11
3.12 Estimated lighting measures and geographic aggregation levels................................................ 3.13
3.13 Categorical Aggregation Levels.................................................................................................. 3.13
3.14 Assumed Day Type Distribution for Estimates of Annual Usage .............................................. 3.15
4.1 Household Average Daily Energy Consumption, Number of Lamps, Daily HOU per
Lamp, and Lamp Power, by RECS Domain ............................................................................... 4.2
4.2 Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy
Consumption, by Dwelling Type and RECS Domain................................................................. 4.4
4.3 Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy
Consumption, by Lamp Type and RECS Domain ...................................................................... 4.5
4.4 Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy
Consumption, by Lamp Type and Lighting Space Type ............................................................ 4.6
4.5 Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy
Consumption, by Fixture Type and Lighting Space Type .......................................................... 4.7
4.6 Household Average Dimmed Number of Lamps, Daily HOU per Lamp, Daily HOU for
all Lamps, and Nominal Lamp Power, by Fixture Type and Lighting Space Type.................... 4.8
5.1 Daily HOU by Lamp Type for California................................................................................... 5.2
5.2 Daily CFL HOU by Dwelling Type for California ..................................................................... 5.2
5.3 Daily CFL HOU by Space Type for California .......................................................................... 5.2
xv
1.0 Introd u c tio n
The U.S. Residential Lighting End-Use Consumption Study aimed to develop reliable estimates of
residential lamp usage and energy consumption at both national and regional levels. Multiple approaches
for pursuing this goal were investigated, exploring tradeoffs in accuracy, flexibility, and required time –
or associated cost. The chosen methodology prioritized flexibility, meaning here the ease of incorporating
new data that might become available in the future. As a result, this effort is best described as the
application of lamp hours-of-use (HOU) models to a newly developed regional estimation framework that
represents the U.S. housing stock. The estimation framework is simply a constructed set of sample
housing units. Each sample housing unit in the estimation framework is described by its household
characteristics (including both housing unit and occupant demographic data) and lighting inventory. This
estimation framework is capable of producing regional and national estimates of lighting usage and
energy consumption for the entire United States, and incorporating new regional data (that meet defined
pre-conditions) for calibrating the HOU models to improve estimation accuracy. This report describes the
development of the estimation framework and the application of the HOU models to the framework,
presents a limited set of lighting estimates, and discusses the accuracy and validity of the presented
estimates. A companion Microsoft Excel spreadsheet contains the full set of estimates produced by this
study, including average number of fixtures, number of lamps, daily HOU per lamp by month, lamp
power, daily energy consumption, and annual energy consumption. The spreadsheet is organized to allow
the estimates to be easily filtered to various levels of aggregation, and by various household and lamp
characteristics and categorical cross-classifications.
Several national and regional studies that occurred between 2008 and 2010 and collected household
characteristics, lighting inventory profiles, and/or lighting end-use metering data were evaluated for use in
developing the HOU models and estimation framework. This study heavily leverages the recent
California Residential Lighting Metering Study (CA RLMS) and U.S. Energy Information Administration
(EIA) Residential Energy Consumption Survey (RECS) datasets. The estimation framework is rooted in
the 2009 RECS housing sample, and the analysis of covariance (ANCOVA) HOU models developed for
the 2008-2009 CA RLMS were used to estimate lighting usage for each lamp type (e.g., incandescent,
compact fluorescent light [CFL], or other type). These and other datasets used in this study are described
in more detail in Section 2.0. Section 3.0 explains the creation of the estimation framework, a combined
dataset containing all the input variables required by the ANCOVA HOU models, and the challenges in
constructing housing unit samples with household characteristics and lighting inventory data that are as
regionally specific as possible. The methods used to apply the models to the estimation framework,
generate lamp-level usage and energy consumption estimates, and aggregate those and other estimates to
various levels are also discussed here. Section 4.0 presents a limited set of lamp usage and related energy
consumption estimates. These estimates were selected to demonstrate the ability of the estimation
framework to generate estimates at regional levels of aggregation and with categorical cross-
classifications. Section 5.0 discusses how the standard error is calculated for all estimates and the data
quality flag in the companion spreadsheet. The section concludes with examples of how the lighting usage
model might be calibrated with end-use data collected in other regions, which represents the primary
objective of potential future updates to the U.S. Residential Lighting End-Use Consumption Study.
All estimates presented in this study are bottom-up, in that they are derived from the lamp and fixture
level within rooms of a housing unit sample, up to the desired level of analysis. Energy consumption is
computed (as the product of lamp power and lamp HOU) at the lowest level and then aggregated up using
1.1
sampling weights. Top-down estimates of energy consumption, on the other hand, are the products of
weighted averages of lamp power and HOU. In general, top-down and bottom-up estimates will not
match, because the average of products usually differs from the product of averages. Bottom-up estimates
are typically more accurate, because the paired relationship between lamp power and HOU is preserved.
For example, suppose one desired to compute the energy consumption of a group of three lamps:
Lamp 1: 100 W, 1.0 HOU per day
Lamp 2: 20 W, 1.5 HOU per day
Lamp 3: 30 W, 2.0 HOU per day
Bottom-Up Energy Consumption = (100 W × 1.0 HOU) + (20 W × 1.5 HOU) + (30 W × 2.0 HOU)
= 190 Watt-hr
Top-Down Energy Consumption = 3 Lamps × Average Lamp Power × Average HOU
= 3 Lamps × (100 + 20 + 30 W) / 3 × (1.0 + 1.5 + 2.0 HOU) / 3
= 3 Lamps × 50 Watts/Lamp × 1.5 HOU
= 225 Watt-hr
Although estimation accuracy is enhanced by the bottom-up approach facilitated by the estimation
framework, it is still limited by the viability of the statistical HOU model, which comes from a single
regional study that has not yet been calibrated for other regions of the United States. However, the CA
RLMS dataset used to create this model comes from perhaps the most comprehensive and statistically
rigorous lighting inventory and end-use metering study to date. The CA RLMS dataset contains complete
inventories for all lamp types collected in more than 1,200 California households and end-use metering
data for a random sampling of up to seven fixtures (each containing one or more lamps) per home, for a
period of several months. Although the actual bias in the estimates presented in this study is unknown, the
statistical precision of the estimates can be quantified.
1.2
2.0 Data Sources
This study applies lamp HOU models to an estimation framework to generate regional and national
estimates of lighting usage and energy consumption. Many datasets were identified and explored for use
in constructing both the HOU models and estimation framework. The CA RLMS showed that the
development of an accurate model requires many variables, spanning both household characteristic and
lighting inventory data, and more than can be found in any one national residential stock assessment,
(e.g., the RECS). Each ANCOVA model requires a dataset containing household characteristics, lighting
inventory data, and end-use metering data. The collection of such data is both time-consuming and
expensive, which largely accounts for the limited number and size of studies that meet these criteria. An
estimation framework was therefore constructed, consisting of a representative sample of U.S. residences,
each of which is described by household characteristic and lighting inventory variables that are used in
the HOU model.
The RECS and other survey results show that household characteristics vary by region, which both
justifies the estimation of lamp usage and energy consumption at regional levels of aggregation, and
suggests the need to acquire data for HOU variables not in the RECS dataset at ideally the same regional
levels. Although merging several smaller studies allows for the creation of more robust models and
greater regional specificity, the validity of such an approach and the accuracy of results derived from the
combined dataset is highly dependent on how consistent each data type is across the studies – which is a
function of how well the data collection protocols match. During the analysis of the various datasets
identified as candidates for use in this study, it was determined that re-categorizing household
characteristic and lighting inventory data was onerous, but possible. Conversely, it was decided that
ensuring end-use metering data from different datasets were of similar accuracy and contained similar
bias was much more difficult, and likely not possible to any degree of certainty. As a result, a strategic
decision was made to construct the estimation framework from the fewest, largest sets of available data,
and reuse the HOU model developed during the CA RLMS without modification.
The following sections describe key characteristics of CA RLMS and the other datasets used in this
study that collected identical or re-categorized versions of the CA RLMS model input variables. Not all
studies collected data for the same variables, but all studies contain some common variables. These
linking variables are essential, as they facilitate the assignment of data from one dataset to housing
samples in another. These linkages are described in detail in Section 3.
2.1 California Residential Lighting Metering Study
The CA RLMS was conducted over 2008-2009 by KEMA for the California Public Utilities
Commission (CPUC). Household characteristics and lighting inventories were collected onsite from a
random sample of more than 1,200 residences throughout the state. The inventories included detailed
information on all lamps and lighting fixtures in the residence (e.g., fixture type, socket type, control type,
lamp type, lamp power, location). In addition, end-use metering data were collected for a random sample
of up to seven lighting fixtures (each containing one or more lamps) per residence using the DENT
Instruments LIGHTINGlogger
TM
(Figure 2.1), resulting in datasets for more than 8,000 lighting fixtures.
The large sample size, scope (i.e., coverage of residence types, room types, and lighting inventory), and
uniform collection protocol make this easily the best single dataset for relating end-use metering data to
household characteristic and lighting inventory data.
2.1
The CA RLMS developed ANCOVA statistical models to produce HOU estimates for a given lamp
type (i.e., incandescent, CFL, other). These models were derived from the full inventory of lamps in all
1,200+ metered homes. The ANCOVA models produce HOU estimates for all lamps in a given fixture,
using characteristics of the fixture and its associated household. The estimated HOU for a particular
fixture varies depending on the combination of characteristics that make up the model inputs. An identical
fixture will have different predicted HOU depending on, for example, whether the household is a single-
family home, multi-family home, or mobile home. The development of these models resulted in a set of
key variables determined to be the most indicative of a household’s lamp use. In some cases multiple
variables captured similar effects, thereby providing the opportunity to reduce the number of variables
used in the model. For example, either Bedrooms or Bathrooms could be used as a successful proxy for
home size (square footage) or income level. The full list of ANCOVA HOU model variables is shown in
Table 2.1. More information about the ANCOVA HOU model for CFLs can be found in the CA RLMS
report.
1
The CA RLMS dataset was initially analyzed to produce estimates of annual HOU for CFLs as part of
the CPUC study. The State of California’s Database for Energy Efficient Resources funded additional
analysis of HOU for the remaining (i.e., incandescent, other) lamp types. HOU models for all lamp types
were made available to this study when permission was granted to use the CA RLMS data.
Figure 2.1. The DENT Instruments LIGHTING Logger Used to Collect End-Use Metering Data
1
Upstream Lighting Program Evaluation Report, Volume 1, Section 8.5.
http://www.calmac.org/publications/FinalUpstreamLightingEvaluationReport_Vol1_CALMAC_3.pdf
2.2
Table 2.1. ANCOVA Model Variables Used in the CA RLMS
Variable
Description
Valid Responses
Bedrooms
Total number of bedrooms in the home
1
2 to 3
4+
Bathrooms
Total number of bathrooms in the home
1
2
3+
Composition
Presence of kids (0-17 years in age) in the
Kids
household
No Kids
Education Level
Highest education level of the household
Less than High School
respondent
High School Graduate
College
Post-Graduate
Lighting Space Type
Type of room or location, inside or outside of
the home, where the lamp is located
Bedroom
Bathroom
Dining Room
Living Room
Kitchen
Office
Hall Entrance
Garage
Exterior
Other
Fixture Type
Mounting location of the fixture housing the
lamp
Ceiling
Other
Number of Sockets
Number of CFLs
Total number of sockets in the home, whether
occupied by a lamp or not
Total number of installed CFLs in the home
Numeric
1 to 2
3 to 4
5+
CFL Saturation
Total number of medium screw-base CFLs in
Numeric
the home (whether installed or in storage)
divided by the total number of medium screw-
base lamps in the home (whether installed or in
storage)
Investor-Owned Utility
(IOU)
IOU that serves the household
Pacific Gas & Electric
Southern California Edison
San Diego Gas & Electric
2.3
2.2 Residential Energy Consumption Survey
The EIA administers the RECS, which collects household characteristics and usage patterns from a
nationally representative sample of housing units using specially trained interviewers. This information is
combined with data from energy suppliers to these homes to estimate energy costs and usage for heating,
cooling, appliances, and other end uses. The RECS was conducted yearly from 1978-1982, every third
year from 1984-1993, and every fourth year thereafter. Various sets of RECS microdata are made
publically available over time on the EIA website.
The estimation framework for this study is fundamentally based on the sample design for the 2009
RECS. The National Opinion Research Center collected onsite data for the 2009 RECS from February
through August 2010. Although the previous 2005 RECS collected data from 4,382 households, the 2009
survey collected data from 12,083 households in housing units statistically selected to represent the 113.6
million housing units that are occupied as a primary residence. The large sample size and scope of the
2009 RECS make it an effort not likely to be repeated in the foreseeable future.
The sample size expansion enabled the creation of reportable domains at smaller geographic levels
than the nine census divisions and four largest states provided by previous RECS microdata. The 2009
RECS defined 27 reportable domains (Table 2.2). A total of 16 individual states were designated as self-
representing domains. The remaining states and the District of Columbia were divided into groups by
geographic alignment within census divisions, with one exception. The Mountain division was further
divided into two subdivisions: Mountain North, representing CO, ID, MT, UT, and WY and Mountain
South, representing AZ, NM, and NV. The 27 RECS-reportable domains may not be available in future
RECS as the sample size may fall back to pre-2009 levels.
Table 2.2. 2009 RECS-Reportable Domains
Domain
Census
Census
Number
States in Domain
Division or Subdivision
Region
1
Connecticut, Maine, New Hampshire, Rhode Island, Vermont
New England
Northeast
2
Massachusetts
New England
Northeast
3
New York
Mid-Atlantic
Northeast
4
New Jersey
Mid-Atlantic
Northeast
5
Pennsylvania
Mid-Atlantic
Northeast
6
Illinois
East North Central
Midwest
7
Indiana, Ohio
East North Central
Midwest
8
Michigan
East North Central
Midwest
9
Wisconsin
East North Central
Midwest
10
Iowa, Minnesota, N. Dakota, S. Dakota
West North Central
Midwest
11
Kansas, Nebraska
West North Central
Midwest
12
Missouri
West North Central
Midwest
13
Virginia
South Atlantic
South
14
Delaware, District of Columbia, Maryland, W. Virginia
South Atlantic
South
15
Georgia
South Atlantic
South
16
N. Carolina, S. Carolina
South Atlantic
South
17
Florida
South Atlantic
South
18
Alabama, Kentucky, Mississippi
East South Central
South
19
Tennessee
East South Central
South
20
Arkansas, Louisiana, Oklahoma
West South Central
South
21
Texas
West South Central
South
22
Colorado
Mountain North
West
2.4
Table 2.2. (contd)
Domain
Census
Census
Number
States in Domain
Division or Subdivision
Region
23
Idaho, Montana, Utah, Wyoming
Mountain North
West
24
Arizona
Mountain South
West
25
Nevada, New Mexico
Mountain South
West
26
California
Pacific
West
27
Alaska, Hawaii, Oregon, Washington
Pacific
West
The 2009 RECS household characteristics microdata were released in November 2011; square
footage microdata were not made available until October 2012.
1
2.3 American Housing Survey
The American Housing Survey (AHS) is “the most comprehensive national housing survey in the
United States.”
2
The U.S. Census Bureau conducts the survey biannually for the Department of Housing
and Urban Development (HUD). A variety of household characteristics, including room-type distribution,
demographics, appliances, and amenity data, are collected using a computer-based questionnaire. The
national survey, which began in 1973, has sampled the same units since 1985; it also samples new
construction to ensure continuity and timeliness of the data. AHS microdata are available on the HUD
website.
3
The estimation framework developed for this study uses only the 2009 AHS national dataset; the
metropolitan survey data were not used.
4
The 2009 AHS national dataset included about 60,000 housing
units. The households in this sample were interviewed between April and September 2009. Each housing
unit in the sample represents about 2,000 other units in the national survey. The weighting was designed
to minimize sampling error and utilize independent estimates of occupied and vacant housing units.
2.4 Nexus Market Research Group Multi-State CFL Modeling Study
During 2009-2010, Nexus Market Research Group, Inc. (NMR) led two rounds of regional studies for
multiple clients focused on gathering household CFL inventory and saturation data. These studies –
collectively referred to here as simply the multi-state CFL study – were commissioned by 16 entities
(Table 2.3) including electric utilities, energy service organizations, public service commissions, and state
agencies. Numerous evaluation contractors collected data and performed analysis in coordination with
NMR. Microdata from the multi-state CFL study are not publically available.
The sponsors of each regional study determined its geographic coverage area. All regional studies
included a survey of household characteristics. A subsample of those households participated in an in-
depth, onsite survey of the lighting inventory information within each room of the home. A summary of
the separate regional components of the multi-state CFL study is provided in Table 2.4.
1
http://www.eia.gov/consumption/residential/data/2009/
2
http://www.census.gov/housing/ahs/
3
http://www.huduser.org/portal/datasets/ahs.html
4
The 2011 AHS dataset was not available during the time this study was being completed. Data collection for the
2011 survey did not begin until August 2011, and the microdata is not expected to be available until late 2012.
2.5
Table 2.3. Multi-State CFL Study Clients
2009 Clients
2010 Clients
California Public Utilities Commission
Ameren Illinois Utilities
Connecticut Energy Efficiency Board
Ameren Union Electric (Missouri)
Consumers Energy (Michigan)
ComEd (Illinois)
Massachusetts ENERGY STAR Lighting Program
Dayton Power and Light
Administrators (Cape Light Compact, NSTAR, National
Grid, Northeast Utilities – WMECO, and Unitil)
New York State Energy Research and Development
Maryland Public Service Commission
Authority
(EmPower Program)
Xcel Energy (Colorado)
Massachusetts program administrators (Cape
Light Compact, NSTAR, National Grid,
Northeast Utilities [WMECO portion], and
Unitil)
Wisconsin Public Service Commission
National Grid Rhode Island
New York State Energy Research and
Development Authority
Salt River Project
Table 2.4. Multi-State CFL Regional Studies by Year, with Number of Household Characteristics and
Lighting Inventory Records
Number of
Number of
Study
Household
Inventory
Code
Study Name
Year
Records
Records
AZ
Arizona – Salt River Project
2010
100
100
CA
California
2009
699
77
CPUC
California RLMS
2010
1,200
1,200
CO
Colorado – Xcel Energy
2009
NA
NA
CT
Connecticut
2009
500
95
GA
Georgia
2009
579
63
ILa
Illinois – Ameren Illinois Utilities
2010
503
92
ILc
Illinois – ComEd
2010
500
100
IN
Indiana
2009
678
88
2010
402
55
KS
Kansas
2009
525
71
2010
465
95
MDa
Maryland
2009
57
57
MD
Maryland – Allegheny, Baltimore Gas & Electric, PEPCO,
2010
79
79
Southern Maryland Electric Cooperative
MA
Massachusetts
2009
100
N/A
2010
169
137
MI
Michigan – Consumers Energy
2009
657
86
2010
300
99
MO
Missouri – Ameren Union Electric
2010
44
87
NYC
New York – NYC
2009
502
100
2010
100
100
2.6
Table 2.4. (contd)
Number of
Number of
Study
Household
Inventory
Code
Study Name
Year
Records
Records
NYS
New York – Upstate
2009
1,001
203
2010
200
200
OH
Ohio – Except for Cincinnati
2009
501
98
OHd
Ohio – Dayton Power & Light
2010
351
72
PA
Pennsylvania
2009
653
60
SD
South Dakota – Part Pennington County
2010
93
93
TX
Texas – Houston
2009
503
99
2010
201
100
DC
Washington DC
2009
500
97
WI
Wisconsin
2009
578
82
NA = not available
The estimation framework developed for this study uses 26 of the 31 datasets collected as part of the
multi-state CFL study. The clients who funded each component study were solicited for permission to use
the microdata in their territory explicitly for this study. Only one client (Xcel Energy) was unable to
release data from their territory (Colorado). The available data spanned 18 different geographic areas and
included household characteristics and demographics from over 11,500 telephone surveys and, from a
subset of more than 2,600 residences, onsite lighting inventories. Although the multi-state CFL study did
not have a national sample design, the collection territories did cover some part of every U.S. census
division except the East South Central division.
2.5 Consideration of Additional Data Sources
Other data sources besides those reviewed in Sections 2.1 through 2.4 were considered for use,
including the 2007-2008 CFL Lighting Markdown Evaluation in New England published in 2009
1
and the
2006-2007 Northwest ENERGY STAR Homes Energy Analysis.
2
The data from these studies were
ultimately not pursued due to some HOU model input fields not having been collected, and the high
expected level of effort required to recode the data. Although use of the multi-state CFL study data
required the solicitation of permission from a large set of funding clients, it was ultimately pursued
because of the data processing efficiencies expected from some level of consistency in data collected and
measurement protocols, and its broad geographic coverage area.
1
http://www.env.state.ma.us/dpu/docs/electric/09-64/12409nstrd2ae.pdf
2
http://neea.org/docs/reports/northwestenergystarhomesenergyanalysisreport20062007.pdf
2.7
3.0 Methodology
The national and regional estimates of lighting usage produced in this study are based on the outputs
of the HOU models developed from the CA RLMS dataset, as described in Section 2.1. These models
were leveraged here to take advantage of the unique qualities of this dataset, including its broad scope
(i.e., coverage of residence types, room types, and lamp types), large sample size, and uniform collection
methodology. Reuse of these models requires input data that meet defined pre-conditions – specifically
that the data contain the variables used by the ANCOVA models, or variables that can be recoded to those
used by the ANCOVA models. It was possible to use the ANCOVA HOU models in this study because
the model inputs defined by the CA RLMS were available collectively in the selected data sources. While
not all inputs were available in each data source, methods for creating linkages between the datasets were
identified that allowed for a combination of the RECS, the AHS, and the multi-state CFL study data
(combined with the CA RLMS data) to generate regionally representative inputs for the ANCOVA HOU
models.
The HOU models were thus applied to regionally varying input data, referred to here as the estimation
framework, rather than only California households. The estimation framework is the 2009 RECS sample,
expanded by imputing additional measures for each housing unit in the sample using data from other
sources. These extensions were made by linking fields in multiple datasets containing national and
regional information with equivalent fields used as inputs to the HOU models. After creating these
linkages, relevant data from each data source were extracted, resulting in a composite dataset of housing
units described by all of the inputs needed by the HOU models, and statistically representative of the
entire United States and its sub-regions. The following sections describe the creation of this estimation
framework in more detail, as well as the methods used to apply the HOU models to the estimation
framework, generate lamp-level usage and energy consumption estimates, and aggregate those and other
estimates to various levels.
3.1 Estimation Framework Characteristics
Each housing unit in the estimation framework is described by its household characteristics, lighting
spaces, and lighting inventory. The following sections describe the variables collected for each of these
categories, including parent datasets, valid variable values, and application in this study. Valid values for
variables used by the HOU models were established to be consistent with those used in the CA RLMS.
3.1.1 Household Characteristic Variables
Household characteristics include housing unit characteristics, occupant demographics, and
lighting/room space types. Housing unit characteristic and occupant demographic data are available with
consistent definitions and in consistent formats from the RECS, AHS, multi-state CFL study, and CA
RLMS datasets, making these variables useful for linking the datasets together for the purpose of
assigning data from one dataset to housing unit samples of another. The lamp HOU models use four of
the seven housing unit and occupant demographic variables in the estimation framework, summarized in
Table 3.1 and Table 3.2, respectively. Estimates were generated in this study at every valid value level for
all household characteristic variables, with the exception of the Rooms variable.
3.1
Table 3.1. Housing Unit Characteristic Variables in the Estimation Framework
Variable
Description
Valid Values
Dwelling Type!!
Home building classification
Single Family
Multi-Family
Mobile Home
Bedrooms +!
Total number of bedrooms in the home
1
2 to 3
4+
Bathrooms +!
Total number of bathrooms in the home
1
2
3+
Rooms
Number of lighting spaces meeting the RECS
Numeric
definition for rooms
(a)
+!Lamp HOU Model Variable !!Estimation Level
(a) Note that RECS rooms are different from lighting spaces. For example, bathrooms are not considered rooms
according to RECS, and therefore do not contribute to total rooms counts. The lighting spaces given in the AHS
that meet the RECS definition for rooms are bedrooms, kitchens, living rooms, dining rooms, family rooms,
recreation rooms, dens, and other finished rooms.
Table 3.2. Occupant Demographic Variables in the Estimation Framework
Variable
Description
Valid Values
Own/Rent !
Household ownership
Own
Rent
Composition +!
Presence of kids (0-17 years in age) in the
Kids
household
(a)
No Kids
Education Level +!
Highest education level of the household
respondent
Less than High School
High School Graduate
College
Post-Graduate
+ Lamp HOU Model Variable ! Estimation Level
(a) The AHS, RECS, and multi-state CFL study contain this information as distribution of age among household
members.
Lighting spaces are rooms and other areas inside and outside of housing units where lighting is used.
The RECS samples only contain data for Number of Bedrooms, Number of Bathrooms, and (Total)
Number of Rooms, which means that other lighting spaces (i.e., beyond bedrooms and bathrooms) must
be assigned to each housing unit sample using data from other datasets. Lighting space type is one of the
lamp HOU model variables, and estimates were generated in this study for each lighting space type and at
the household level. The lighting spaces in the estimation framework are summarized in Table 3.3.
3.2
Table 3.3. Lighting Space Types in the Estimation Framework
Variable
Description
Valid Values
Lighting Space Type +!
Type of room or location, inside or outside of
the home, where the lamp is located
Bedroom
Bathroom
Dining Room
Living Room
Kitchen
Office
Hall Entrance
Garage
Exterior
Other
+ Lamp HOU Model Variable ! Estimation Level
3.1.2 Lighting Inventory Variables
Lighting inventory data for each housing unit in the estimation framework includes lamp
characteristics and aggregations by fixture type and (lighting space type) location, as well as household
aggregations of sockets and lamps by lamp type. Lighting inventory data is not available in the RECS or
AHS datasets, but was collected for the CA RLMS and multi-state CFL study. Lighting inventories were
assigned to each housing unit sample in the estimation framework using data from these two studies.
Lighting inventories consist of lamp and fixture records. Lamp records contain lamp characteristics,
as summarized in Table 3.4. Fixture records contain fixture characteristics, as summarized in Table 3.5.
Note that the lamp HOU models use only fixture characteristics, and are independent of any lamp-level
characteristics. Lamp power is both an estimated parameter, and a variable used for estimating lamp
energy consumption, which is simply calculated as lamp power x lamp HOU.
Table 3.4. Lamp Characteristics in the Estimation Framework
Variable
Description
Valid Values
Location
The location of the lamp in the home
Fixture
Storage
Socket Type !
Socket used to install the lamp in a fixture
Screw-Base
Pin-Base
Other Base
Control Type !
Control used to operate the lamp
On/Off control
3-way control
Dimming control
Other control
Lamp Type !
Lighting technology used by the lamp
Incandescent
CFL
Other (e.g., LED)
Lamp Power "
Rate of lamp energy consumption, in Watts
Numeric
! Estimation Level " Estimated Lighting Measure
3.3
Table 3.5. Fixture Characteristics in the Estimation Framework
Variable
Description
Valid Value
Fixture Location +!
Lighting space where the fixture is installed
Lighting Space Type
(see Table 3.3)
Fixture Type +!
Mounting location of the fixture housing the lamp
Ceiling
Non-Ceiling
Fixture Sockets
Total number of sockets in the fixture
Numeric
+!Lamp HOU Model Variable !!Estimation Level
Lighting inventory aggregations done at the household level are summarized in Table 3.6. These
aggregations are generated from lamp and fixture records, and all result in a calculated numeric variable
value.
Table 3.6. Lighting Inventory Aggregations in the Estimation Framework
Variable
Description
Number of Fixtures "
Total number of fixtures in the home
Number of Sockets +
Total number of sockets in the home, whether occupied by a lamp or not
(a)
Lamps by fixture type "
Total number of installed lamps in the home, by fixture type
Lamps by lamp type "
Total number of installed lamps in the home, by lamp type
Lamps by lighting space "
Total number of installed lamps in the home, by lighting space
Number of CFLs +
Total number of installed CFLs in the home
CFL saturation +
Total number of medium screw-base CFLs in the home (whether installed or in
storage) divided by the total number of medium screw-base lamps in the home
(whether installed or in storage)
(b)
+ Lamp HOU Model Variable " Estimated Lighting Measure
(a) Empty sockets were not collected in all of the Multi-State CFL Modeling studies, so total number of installed
lamps was used as a proxy for number of sockets in this study.
(b) The accuracy of this variable, and consequently any HOU calculation that uses it, relies heavily on the
assumption that all lamps in the home were accounted for in the lighting inventory. Calculations for this study
were made using only installed lamps.
3.1.3 Regional Variables
The CA RLMS found that, in order to achieve its accuracy goals and meet the needs of study
stakeholders, a regional variable was required in the HOU models. As noted in Table 2.1, the IOU that
served the household was used as a categorical variable to satisfy this requirement. The ANCOVA
coefficients generated by the CA RLMS modeling effort for this variable were specific to the IOU
regions. Although it is possible to determine and assign an IOU variable to each housing unit in the
estimation framework, no method was defined for matching service utility types or otherwise generating
an appropriate ANCOVA coefficient for this variable. Consequently, to facilitate use of the existing CA
RLMS HOU models, the average of the three HOU model variables corresponding with the three
California IOUs was used in this study. The collection of new end-use metering data in any region would
allow for this regional variable to be calibrated, and thereby represents one of the primary opportunities
for improving the estimates in this study.
3.4
3.2 Input Dataset Preparation
Creating the national and regional estimates for this study required combining information from
several datasets into a composite dataset, or estimation framework. Once datasets with key information
were identified, it was essential to ensure that they also had a common set of linking variables that
allowed for the assignment of data from one dataset to another. Careful examination of the information
contained in each dataset was critical.
The following sections describe the steps taken to prepare all the input datasets for use in creating the
composite dataset. Each input dataset was initially checked for scope and completeness, including its
sample size, geographic coverage area, and data type value (e.g., household characteristics, lighting
inventory). Variable names and values were then recoded, if necessary, to be consistent with
Tables 3.1-3.5. Finally, key variables were identified, including HOU model variables and linking
variables. Following preparation, all input data were both ready to be used as inputs to the CA RLMS
HOU models, and assigned to housing units in the estimation framework using common linking variables.
3.2.1 2009 RECS Microdata
The estimation framework is fundamentally rooted in the 2009 RECS housing sample. The 2009
RECS microdata contains information for each of the household characteristic variables listed in
Table 3.1 and Table 3.2. The RECS was designed for statistical coverage of all U.S. households, as well
as households in geographic sub-regions referred to as reportable domains. The RECS microdata contain
sample expansion weights. These weights can be used to aggregate household estimates by any
geographic sub-region supported by the RECS, as well as by other household characteristics.
All RECS data come in a consistent format, making recoding a simple effort. Although the 2009
RECS household characteristics microdata were released in November 2011, square footage microdata
were not made available until October 2012, after the estimates defined by this study had already been
generated. Consequently, this study was unable to take advantage of the square footage data. However,
the CA RLMS found that number of bedrooms and number of bathrooms, both available in the household
characteristics microdata, served as good proxies for total square footage.
3.2.2 2009 AHS Microdata
The 2009 AHS microdata contain household characteristics, space type distributions, and amenity
data. The key AHS data leveraged by this study were the counts of several space types not collected in the
RECS. The AHS microdata, however, only contain geographic identifiers for the four census regions:
Northeast, Midwest, South, and West.
All AHS data also comes in a consistent format, again making recoding a simple effort. To make use
of the key space type data available in the AHS, a method was developed for assigning space type
configurations to housing units in the estimation framework. A separate profile for each census region
was produced by calculating the average distribution of all lighting space types collected in the AHS for
particular combinations of housing unit characteristics (Table 3.1) available in the RECS. For example, in
a RECS single-family home with eight RECS defined rooms, three bedrooms, and three bathrooms, the
AHS data were used to determine the distribution of combinations for the remaining five rooms in the
3.5
home (other than the three bedrooms) and other space types where lighting is used (including bathrooms)
that do not meet the RECS definition for a room. The housing unit characteristics chosen as linking
variables were selected for two reasons. First, they provide information about the physical structure of the
home. Second, they strike a balance between matching by too few variables, in which case the matched
AHS households could have very little in common with the RECS households that they were being
matched to, and too many variables, which could lead to valid household matches being excluded.
3.2.3 2009-2010 Multi-State CFL Study Microdata
The 2009-2010 multi-state CFL study microdata contain household characteristics and, most
importantly, lighting inventory data by lighting space type. The lighting inventory data were used to
regionally assign fixtures and lamps to housing units in the estimation framework with lighting space
distributions assigned by the AHS data.
Although the two rounds of regional studies that comprised the multi-state CFL study had a common
goal and management, numerous evaluation contractors collected and processed the data. As a result,
additional effort was required to process, recode, and make use of the 30 separate datasets made available
to this study. The following sections explain some of the methods employed to prepare the multi-state
CFL study data.
3.2.3.1 Linking Household and Lighting Inventory Data
Each regional study in the multi-state CFL study provided two datasets – a household characteristics
dataset and a lighting inventory dataset. The inventory dataset was typically a subset of the characteristics
dataset. The first step in preparing the multi-state CFL data was to verify that housing units in both
datasets could be matched, usually with a case identification number. All inventory dataset housing units
were successfully linked to a characteristics dataset housing unit, with the exception of four units in the
2010 Illinois – ComEd study and one unit in the 2010 Massachusetts study. The Massachusetts study had
two characteristics datasets and two inventory datasets, one of each for 2009 and one of each for 2010.
The 2010 datasets included a matching variable, but the 2009 datasets did not, resulting in the exclusion
of the 2009 Massachusetts data.
3.2.3.2 Recoding Variables
Although regional participants in the multi-state CFL study typically collected the same information,
variables were often recorded in different formats. Fortunately, common formats were used for subsets of
data. For example, upon initial investigation, it was determined that the lighting inventories followed one
of four formats. After sorting the regional study data sets by the format employed, the datasets could be
efficiently recoded in groups. An example of how lighting spaces were recoded for a subset of regional
study data is shown in Table 3.7
3.6
Table 3.7. Recoding of Lighting Space Types for the 2009 Four-State Study (CA, KA, GA, PA) Datasets
Original Space Type
Recoded Space Type
Bathroom 1
Bathroom
Bathroom 2
Bathroom
Bathroom 3
Bathroom
Bedroom 1
Bedroom
Bedroom 2
Bedroom
Bedroom 3
Bedroom
Bedroom 4
Bedroom
Closet 1
Closet
Closet 2
Closet
Closet 3
Closet
Closet 4
Closet
Formal/Separate Dining Room
Dining
Garage
Garage
Hallway/Entry 1
Hallway
Hallway/Entry 2
Hallway
Hallway/Entry 3
Hallway
In Storage
Storage
Kitchen/Dining Area
Kitchen
Laundry/Utility Room
Laundry
Office/Den
Office
Other
Other
Other/Secondary Living Space
Other
Outside Lamps
Exterior
Primary Living Space
Living
3.2.3.3 Imputing Missing Values
In a few instances, a multi-state CFL study dataset was missing a variable needed for the analysis. In
these cases, a procedure was developed to impute the missing values based on patterns found in the other
studies. To impute categorical variables, a logistic regression was applied and the most likely valid value
was assigned to the variable. Ordinary least-squares regression was used to impute continuous variables.
Because the Michigan – Consumers Study did not collect fixture type information (ceiling versus
non-ceiling), it was imputed by using a logistic regression based on dwelling type, space type, and
number of lamps per fixture. Inventory data from all of the multi-state CFL studies were used for this
estimation.
The 2009 coordinated studies in CA, KS, GA, PA, collectively referred to as the 2009 four-state
study, did not collect lamp power during the inventory visits. Two options for proceeding were
considered. The first and simplest option was to assume a single lamp power for each lamp type (e.g.,
60 W for all incandescent lamps). Given the high likelihood that lamp power varies according to where it
is installed, however, it was decided to impute lamp power using a regression based on the dwelling type,
lighting space type, fixture type, and lamp type.
Table 3.8 and Table 3.9 present the combined multi-state CFL study distributions before and after
imputation. The tables indicate that imputation did not significantly affect the overall fixture type
distribution or average lamp power by lamp type.
3.7
Table 3.8. Fixture Type Distribution, Before and After Imputation in Michigan – Consumers Study
Territory
Multi-State CFL
Multi-State CFL Study After
Study Before
Imputation in Michigan –
Michigan – Consumers
Fixture Type
Imputation
Consumers Territory
Territory After Imputation
Ceiling
55.2%
55.4%
59.8%
Other
44.9%
44.6%
40.2%
Table 3.9. Average Lamp Power (W) by Lamp Type, Before and After Imputation in 2009 Four-State
Study Territories
Multi-State CFL Study
Multi-State CFL Study
After Imputation in
2009 Four-State Study
Before Imputation
2009 Four-State Study
After Imputation
CFL
16.7
16.4
15.8
Incandescent
61.4
61.8
62.4
Other
46.6
52.2
59.9
3.2.3.4 2009 Four-State Study Data (CA, KS, GA, PA)
Following the data examination process, it was determined that the 2009 four-state study could not be
used for estimating HOU for a number of reasons. First, the four-state study collected lighting
information only at the room level, not at the fixture level as the CA RLMS and the other multi-state CFL
studies had done. Second, the four-state study only recorded counts of CFLs and total number of lamps,
without distinguishing incandescent and other lamp types. Finally, the four-state study did not record
socket types.
Although the missing data could have been imputed by leveraging other sources, little information
would be gained by doing so. Usable information from the four-state study (i.e., demographics and space
type) was also available from the RECS and AHS. Furthermore, a reconstructed dataset would not be
significantly different from an average of the other studies.
3.3 Combining Datasets
The RECS, AHS, and multi-state CFL study datasets were combined to create a composite dataset, or
estimation framework, from which all lighting estimates were generated. More specifically, the data
representing each RECS case was extended with data from each of the other two sources so that each
RECS case could be described using all the variables in the ANCOVA model.
3.3.1 Extending RECS Housing Units with AHS Data
RECS housing units were extended with AHS data to augment the lighting space information in the
RECS case. The RECS only collects number of bedrooms, number of bathrooms, and number of other
rooms, but the AHS provides data on more specific lighting space types. Although the RECS provides
sample units for 27 domains, each AHS household is only categorized by 1 of 4 census regions. Census
region was therefore used as the geographic variable linking the RECS and the AHS datasets.
Lighting space configurations are possible combinations of lighting space types for a housing unit.
Lighting space configurations from the AHS were assigned to RECS cases using a total of five linking
3.8
variables: census region, dwelling type, number of bedrooms, number of bathrooms, and total number of
rooms in the household.
1
Because the AHS microdata include many more sample cases than the RECS,
and the AHS geographic identifiers are at a higher level than those of the RECS, multiple space type
configurations are generally linked to each RECS case. Rather than choosing a single AHS space type
configuration to match to each RECS case, a separate RECS replicate housing unit was created for each
of the multiple configurations for analysis. The AHS weight was used to give the level of contribution to
each configuration.
For example, consider RECS ID 10284. The linking variables and values used to connect this
household to space type configurations from the AHS are as follows:
Census Region: Midwest
Dwelling Type: Single-family
Bedrooms: 3
Bathrooms
2
: 3
RECS Rooms: 8
The replicate cases all have the exact same linking variable values as RECS ID 10284, but also
contain space type configurations from actual AHS housing units that match on these linking variable
values. Table 3.10 gives the top five matched AHS space type configurations (in terms of the total
associated AHS weights) for this RECS ID. There were 54 unique configurations that were matched to
this RECS case, and therefore 54 replicates were created for it, each with representation proportional to
the sum of the AHS weights for the configuration
3
, shown in the second column. For each RECS case, an
average number of occurrences for each lighting space type was computed using the weights in the
second column.
Table 3.10. Top Five Lighting Space Configurations for a RECS ID 10284
AHS Space Types
Lighting Space Types Meeting RECS Room Definition
Other Lighting Space Types
Lighting Space
Configuration
Σ AHS Weights
for configuration
Bedrooms
Kitchens
Living rooms
Dining rooms
Family Rooms
Recreation
Rooms
Dens
Other Finished
Rooms
Total RECS
Rooms
Bathrooms
Half Bathrooms
Business or
Personal Use
Exclusive
Business
Laundry Rooms
1
14,747
3
1
1
1
1
0
1
0
8
2
1
0
0
1
2
13,923
3
1
1
1
1
1
0
0
8
2
1
0
0
1
3
12,900
3
1
1
1
1
0
0
1
8
2
1
0
0
1
4
10,698
3
1
1
1
1
1
0
0
8
2
1
0
0
0
5
9,528
3
1
1
1
1
0
1
0
8
2
1
0
0
0
1
The difference between RECS rooms and lighting spaces is explained in Section 3.1.1.
2
Recall that bathrooms are not considered rooms according to RECS, and therefore do not contribute to the total
rooms count.
3
Each lighting space configuration may represent many AHS households.
3.9
Out of 12,083 households in the 2009 RECS, only 178 (or 1.5 percent) could not be matched to any
AHS configuration using this approach. These cases were discarded from the analysis, and the weights for
the remaining RECS cases were adjusted so that the sum of the weights before and after the drop was the
same.
3.3.2 Extending RECS Housing Units with Multi-State CFL Study Data
The next step in the creation of the estimation framework was to combine the newly formed RECS-
AHS housing unit samples with the multi-state CFL study data. The RECS housing units were further
extended with lighting configurations (consisting of fixtures and lamps) based on the multi-state CFL
study lighting inventory data.
Whereas the RECS and AHS datasets covered all U.S. census divisions, the multi-state CFL study
datasets had limited and specific coverage areas. Furthermore, the distribution and sample size for each
coverage area varied. To overcome these limitations, each RECS domain was assigned one or more
individual multi-state CFL study dataset based on geographic proximity. These datasets were weighted to
indicate how much each one should be represented when aggregated to the RECS domain level. For
locations where no regional study was reasonably close, a combination of the entire multi-state CFL study
was used.
Following the development of a strategy for combining individual multi-state CFL studies to
represent each RECS domain, a number of methods for combining these datasets with the RECS-AHS
data were investigated. An iterative procedure was developed. First, two primary linking variables were
selected that were mandatory for matching: dwelling type and lighting space type. For example, the
procedure would not match lighting configuration from a multi-family household into a single-family
household, nor would it match lighting configurations from a garage into a bedroom. Then, a secondary
list of linking variables was chosen that would be used for matching when possible: own/rent,
composition, education level, bedrooms, and bathrooms.
The lighting configurations assigned from the multi-state CFL study lighting inventory data included
lamp characteristics (socket type, control type, lamp type, lamp power), fixture characteristics (fixture
type, fixture sockets), and aggregations including total number of fixtures and lamps. The iterative
process for matching and making assignments began with a list of secondary linking variables to match.
In the first iteration, all secondary variables were included. In subsequent iterations, the list was reduced
by a single variable to increase the chances of matching. Lighting inventory data with matching variables
(primary and secondary) were aggregated across the component studies assigned to each RECS domain,
weighted, and assigned. This process was repeated until each RECS housing unit had been extended with
a lighting configuration. This process took up to 23 iterations for each RECS domain.
Because the multi-state CFL study included limited data for mobile homes, all the mobile homes were
pooled into a single dataset to be used for all domains.
3.3.3 Estimation Framework Summary
Table 3.11 summarizes the different sources of information that were used for each state, including
how the 2009-2010 multi-state CFL study datasets were assigned. Each state belongs to one of the 27
3.10
Table 3.11. Summary of Data Sources by State
3.11
State
Code
RECS Domain
Census Division or
Subdivision
Census Region
Household
Characteristics
(Additional)
Lighting
Space Types
Lighting Inventory Composition, from
CFL Multi-State Study Regions
(a)
HOU
Metering
Data
AL
AL, KY, MS
East South Central
South
RECS
AHS
100.0% ALL
CA RLMS
AK
AK, HI, OR, WA
Pacific
West
RECS
AHS
83.3% ALL, 16.7% CPUC
CA RLMS
AZ
AZ
Mountain South
West
RECS
AHS
83.3% AZ, 16.7% ALL
CA RLMS
AR
AR, LA, OK
West South Central
South
RECS
AHS
62.5% ALL, 25.0% TX, 12.5% KS
CA RLMS
CA
CA
Pacific
West
RECS
AHS
100.0% CPUC
CA RLMS
CO
CO
Mountain North
West
RECS
AHS
85.7% ALL, 14.3% AZ
CA RLMS
CT
CT, ME, NH, RI, VT
New England
Northeast
RECS
AHS
50.0% MA, 33.3% CT, 16.7% NYS
CA RLMS
DE
DE, DC, MD, WV
South Atlantic
South
RECS
AHS
50.0% MD, 25.0% DC, 12.5% ALL, 12.5% OH
CA RLMS
FL
FL
South Atlantic
South
RECS
AHS
100.0% ALL
CA RLMS
GA
GA
South Atlantic
South
RECS
AHS
100.0% ALL
CA RLMS
HI
AK, HI, OR, WA
Pacific
West
RECS
AHS
83.3% ALL, 16.7% CPUC
CA RLMS
ID
ID, MT, UT, WY
Mountain North
West
RECS
AHS
87.5% ALL, 12.5% AZ
CA RLMS
IL
IL
East North Central
Midwest
RECS
AHS
50.0% ILa, 50.0% ILc
CA RLMS
IN
IN, OH
East North Central
Midwest
RECS
AHS
50.0% IN, 33.3% OH, 16.7% OHd
CA RLMS
IA
IA, MN, ND, SD
West North Central
Midwest
RECS
AHS
50.0% SD, 16.7% ILa, 16.7% MO, 16.7% WI
CA RLMS
KS
KS, NE
West North Central
Midwest
RECS
AHS
83.3% KS, 16.7% MO
CA RLMS
KY
AL, KY, MS
East South Central
South
RECS
AHS
100.0% ALL
CA RLMS
LA
AR, LA, OK
West South Central
South
RECS
AHS
62.5% ALL, 25.0% TX, 12.5% KS
CA RLMS
ME
CT, ME, NH, RI, VT
New England
Northeast
RECS
AHS
50.0% MA, 33.3% CT, 16.7% NYS
CA RLMS
MD
DE, DC, MD, WV
South Atlantic
South
RECS
AHS
50.0% MD, 25.0% DC, 12.5% ALL, 12.5% OH
CA RLMS
MA
MA
New England
Northeast
RECS
AHS
87.5% MA, 12.5% CT
CA RLMS
MI
MI
East North Central
Midwest
RECS
AHS
100.0% MI
CA RLMS
MN
IA, MN, ND, SD
West North Central
Midwest
RECS
AHS
50.0% SD, 16.7% ILa, 16.7% MO, 16.7% WI
CA RLMS
MS
AL, KY, MS
East South Central
South
RECS
AHS
100.0% ALL
CA RLMS
MO
MO
West North Central
Midwest
RECS
AHS
87.5% MO, 12.5% KS
CA RLMS
MT
ID, MT, UT, WY
Mountain North
West
RECS
AHS
87.5% ALL, 12.5% AZ
CA RLMS
NE
KS, NE
West North Central
Midwest
RECS
AHS
83.3% KS, 16.7% MO
CA RLMS
NH
CT, ME, NH, RI, VT
New England
Northeast
RECS
AHS
50.0% MA, 33.3% CT, 16.7% NYS
CA RLMS
Table 3.11. (contd)
3.12
State
Code
RECS Domain
Census Division or
Subdivision
Census Region
Household
Characteristics
(Additional)
Lighting
Space Types
Lighting Inventory Composition, from CFL Multi-
State Study Regions
(a)
HOU
Metering
Data
NV
NV, NM
Mountain South
West
RECS
AHS
50.0% ALL, 25.0% AZ, 25.0% CPUC
CA RLMS
NJ
NJ
Middle Atlantic
Northeast
RECS
AHS
25.0% CT, 25.0% MD, 25.0% NYC, 25.0% NYS
CA RLMS
NM
NV, NM
Mountain South
West
RECS
AHS
50.0% ALL, 25.0% AZ, 25.0% CPUC
CA RLMS
NY
NY
Middle Atlantic
Northeast
RECS
AHS
66.7% NYS, 33.3% NYC
CA RLMS
NC
NC, SC
South Atlantic
South
RECS
AHS
100.0% ALL
CA RLMS
ND
IA, MN, ND, SD
West North Central
Midwest
RECS
AHS
50.0% SD, 16.7% ILa, 16.7% MO, 16.7% WI
CA RLMS
OH
IN, OH
East North Central
Midwest
RECS
AHS
50.0% IN, 33.3% OH, 16.7% OHd
CA RLMS
OK
AR, LA, OK
West South Central
South
RECS
AHS
62.5% ALL, 25.0% TX, 12.5% KS
CA RLMS
OR
AK, HI, OR, WA
Pacific
West
RECS
AHS
83.3% ALL, 16.7% CPUC
CA RLMS
PA
PA
Middle Atlantic
Northeast
RECS
AHS
50.0% NYS, 25.0% MD, 25.0% NYC
CA RLMS
RI
CT, ME, NH, RI, VT
New England
Northeast
RECS
AHS
50.0% MA, 33.3% CT, 16.7% NYS
CA RLMS
SC
NC, SC
South Atlantic
South
RECS
AHS
100.0% ALL
CA RLMS
SD
IA, MN, ND, SD
West North Central
Midwest
RECS
AHS
50.0% SD, 16.7% ILa, 16.7% MO, 16.7% WI
CA RLMS
TN
TN
East South Central
South
RECS
AHS
87.5% ALL, 12.5% MO
CA RLMS
TX
TX
West South Central
South
RECS
AHS
50.0% ALL, 50.0% TX
CA RLMS
UT
ID, MT, UT, WY
Mountain North
West
RECS
AHS
87.5% ALL, 12.5% AZ
CA RLMS
VT
CT, ME, NH, RI, VT
New England
Northeast
RECS
AHS
50.0% MA, 33.3% CT, 16.7% NYS
CA RLMS
VA
VA
South Atlantic
South
RECS
AHS
50.0% MD, 37.5% ALL, 12.5% DC
CA RLMS
WA
AK, HI, OR, WA
Pacific
West
RECS
AHS
83.3% ALL, 16.7% CPUC
CA RLMS
WV
DE, DC, MD, WV
South Atlantic
South
RECS
AHS
50.0% MD, 25.0% DC, 12.5% ALL, 12.5% OH
CA RLMS
WI
WI
East North Central
Midwest
RECS
AHS
87.5% WI, 12.5% MI
CA RLMS
WY
ID, MT, UT, WY
Mountain North
West
RECS
AHS
87.5% ALL, 12.5% AZ
CA RLMS
(a) ALL refers to the average lighting configuration across ALL multi-state CFL study regions
RECS domains. Household characteristics came from the 2009 RECS, with additional lighting spaces
drawn from the 2009 AHS. End-use metering data came solely from the 2008-2009 CA RLMS. The
weights for each multi-state CFL study used for the lighting inventory in each state are shown using the
study codes given in Table 2.4. Note the use of “ALL” to represent the average lighting configuration
across ALL multi-state CFL study regions.
3.4 Lighting Estimates
As described in Section 3.3, the estimation framework was created by first assigning space types to
each RECS sample respondent, or housing unit, and then assigning a lighting inventory to each space
type. Estimates of lighting usage and energy consumption were then made by applying the HOU models
to lamps in each housing unit. Next, these and other estimates were aggregated to various levels by
applying the RECS sample weights to the household measures in the estimation framework. The use of
RECS sample weights results in statistically unbiased estimates for household characteristics estimation
levels.
Estimates were made for seven lighting measures, at four geographic aggregation levels, and by 27
categorical aggregation levels, as summarized in Table 3.12 and Table 3.13. The following sections
describe the procedures for generating lighting usage and energy consumption estimates, and performing
aggregations.
Table 3.12. Estimated lighting measures and geographic aggregation levels
Estimated Lighting Measures
Geographic Aggregation Levels
Average number of fixtures
Nationally
Average number of lamps
Census Division
Average lamp power
Census Region
Average daily HOU, per lamp, all months
RECS Domain
Average daily HOU, all lamps, all months
Average daily HOU, per lamp, by month
Average daily energy consumption
Average annual energy consumption
Table 3.13. Categorical Aggregation Levels
Household
Characteristics
Lamp or Fixture
Characteristics
Cross-Classifications
Dwelling Type
Bedrooms
Bathrooms
Socket Type
Control Type
Lamp Type
Dwelling Type AND Fixture Type
Dwelling Type AND Lamp Type
Dwelling Type AND Lighting Space Type
Ownership
Fixture Type
Bedrooms AND Lighting Space Type
Composition
Bathrooms AND Lighting Space Type
Education Level
Lighting Space Type AND Fixture Type
Lighting Space Type
Lighting Space Type AND Socket Type
Lighting Space Type AND Control Type
Lighting Space Type AND Lamp Type
Lighting Space Type AND Fixture Type AND Control Type
Fixture Type AND Socket Type
Fixture Type AND Control Type
3.13
Table 3.13. Categorical Aggregation Levels
Household
Lamp or Fixture
Characteristics
Characteristics
Cross-Classifications
Fixture Type AND Lamp Type
Lamp Type AND Socket Type
Lamp Type AND Control Type
ℎ!!
!"#$
3.4.1 Lamp Usage and Energy Consumption
The HOU model ANCOVA coefficients from the CA RLMS study were applied to the estimation
framework to produce estimates of average daily lamp HOU. A more detailed explanation of the
ANCOVA model is available in the CA RLMS report.
1
Lamp HOU was estimated for each lamp type (incandescent, CFL, other) by multiplying the model
coefficients by the corresponding covariates in the analysis dataset, and combining them according to the
following equation:
= ! !
!
+ !
!
!"#$"%&'&" !
!
+ !
!
!"#$%"#&' !
!"
+ !
!
!"#$%&'"!
!"#
+ ⋯ + !
!"#$
where ℎ!!
!"#$
= Estimated lamp HOU
r = Estimation framework housing unit
! = Lighting space configuration
! = Inventory configuration
! = Day type (weekday or weekend/holiday)
!
!
, !
!
, … = ANCOVA model coefficients for day type d
!
!"#$
= Model residuals
2
ANCOVA coefficients for the CFL HOU model are available in the CA RLMS report.
3
Calculating predicted HOU for each lamp in the estimation framework resulted in a large dataset that
needed to be aggregated to the levels of interest. Because each level was weighted, data was aggregated in
multiple steps.
1. Identify the household samples in the estimation framework that have the household
characteristics and lighting inventory variables of interest, and select them for use in creating the
estimate of interest. For example, if one desires to estimate the average daily HOU for lamps in
ceiling fixtures in dining rooms in single-family homes in Missouri, the first step is to identify
and select only those household samples in the estimation framework that are single-family
homes in Missouri, only their lighting space distributions that contain dining rooms, and only the
dining rooms that contain ceiling fixtures.
1
Upstream Lighting Program Evaluation Report, Volume 1, Section 8.5.
http://www.calmac.org/publications/FinalUpstreamLightingEvaluationReport_Vol1_CALMAC_3.pdf
2
Model residuals were set to zero for all estimates made in this study.
3
Upstream Lighting Program Evaluation Report, Volume 1, Table 82.
http://www.calmac.org/publications/FinalUpstreamLightingEvaluationReport_Vol1_CALMAC_3.pdf
3.14
2. Calculate and average the HOU for the identified and selected lamps, using appropriate multi-
state CFL study lighting inventory weights. For example, if 10 multi-state CFL study households
– 5 from Missouri and 5 from Kansas – were matched to one of the identified and selected RECS
single-family Missouri households with a dining room and a ceiling fixture, the calculated HOU
for the lamps in those 10 cases would be averaged using the weights assigned for the multi-state
CFL study data in Missouri: 87.5 percent for the 5 multi-state CFL study Missouri households
and 12.5 percent for 5 multi-state CFL study Kansas households. This produces an estimate of
average daily HOU for lamps in ceiling fixtures, for each lighting space configuration containing
a dining room, for each single-family home in Missouri in the estimation framework.
3. Average the daily lamp HOU produced in the previous step across the lighting space
configurations matched to each identified and selected household in the estimation framework,
using appropriate AHS lighting space distribution weights. Continuing the example, the average
daily HOU for lamps in ceiling fixtures produced in the previous step would be averaged across
all the matched lighting space configurations assigned to each single-family home in Missouri in
the estimation framework, using their AHS weights. This produces an estimate of average daily
HOU for lamps in dining room ceiling fixtures, for each single-family home in Missouri in the
estimation framework.
4. Average the daily lamp HOU produced in the previous step across the geographic region of
interest, using appropriate RECS sample design weights. Continuing the example, the average
daily HOU for lamps in dining room ceiling fixtures produced in the previous step would be
averaged across all single-family homes in Missouri in the estimation framework, using their
RECS sample design weights. This produces an estimate of average daily HOU for lamps in
ceiling fixtures in single-family homes in Missouri.
Note that the lamp HOU model calculates different results for weekdays and for weekends or
holidays. To derive an annual average, weighting by number of days is applied using the assumptions in
Table 3.14.
Table 3.14. Assumed Day Type Distribution for Estimates of Annual Usage
Day Type
Days in a Year
Weekdays
250
Weekends/Holidays
114
Total
364
Finally, estimates of energy consumption for each lamp is given by:
!"#$ !
!"#
= !" #$$%& '!
!"#
×ℎ !!
!"#
where !"#$ !
!"#
= estimated lamp energy consumption
!"#$$% & '!
!"#
= lamp power (watts)
ℎ!!
!"#
= estimated lamp HOU
! = estimation framework housing unit
! = lighting space configuration
! = inventory configuration
3.15
Average lamp power and energy consumption estimates were aggregated from the composite dataset
using the same procedure used for aggregating lighting usage.
3.4.2 Seasonal Variation
The ANCOVA models produced by the CA RLMS estimated the average lamp HOU per day over the
course of a year. These models were created by annualizing each set of end-use metering data (which
typically was only collected over a portion of the year) using a sinusoidal regression model. Re-applying
the sinusoidal fit to the annualized estimate can create estimates of HOU within the year. This procedure
was used to generate estimates of average daily HOU per lamp, by month. The annualization procedure is
described in greater detail in the CA RLMS report.
1
3.4.3 Number of Fixtures and Lamps
Estimates for the number of fixtures and lamps were generated by aggregating from the composite
dataset to the levels of interest using the same procedure delineated in the previous section for creating
lamp power, lamp usage, and energy consumption estimates.
1
Upstream Lighting Program Evaluation Report, Volume 1, Section 8.5.
http://www.calmac.org/publications/FinalUpstreamLightingEvaluationReport_Vol1_CALMAC_3.pdf
3.16
4.0 Initial Estimation Highlights
This study distinguishes itself from previous residential lighting stock and energy consumption
studies in that the estimates were generated using a bottom-up analysis and produced for both the entire
United States and at various regional levels. The full set of estimates produced by this study, as described
in Section 3.4, is provided in a companion spreadsheet. A select number of these estimates are presented
here, focusing on comparisons of high-level estimates with previous studies, the ability to see regional
variation courtesy of the RECS-based estimation framework, and the ability to produce cross-
classifications afforded by the bottom-up analysis.
By definition, all statistical estimates have limitations. As described previously, the estimates
produced by this study were not derived from new primary data or a single dataset. They are the result of
a meta-analysis of datasets from previous residential lighting studies. While each of the datasets used was
the best and most recent of its kind, the studies that produced them all had a more limited scope than this
study set out to achieve. As a result, all estimates presented here should be viewed in the context of their
accuracy and validity, which is discussed in detail in Section 5.0.
Finally, a consequence of using data sources from studies conducted over a span of years is that the
estimates do not reflect one specific year. The data used in this study were collected between 2008 and
2010. The 2009 RECS household interviews began in early 2010 and were completed later that year.
1
The
2009 AHS data was similarly collected in 2010, while the multi-state CFL study and CA RLMS were
conducted over 2008-2009. Because all data sources collected data in 2010, the results are best presented
as 2010 estimates, with the aforementioned caveats.
4.1 Total Energy Consumption
The residential sector consumed an estimated 194.1 TWh of electricity for lighting annually in and
around 2010, as computed using the bottom-up aggregation described previously. A top-down estimate of
total lighting energy consumed results in 206.6 TWh (i.e., 1.6 HOU per lamp × 47.7 watts per lamp ×
67.4 lamps per home × 365 days per year × 113,566,400 homes in the United States). The bottom-up
estimate is more accurate because it incorporates specific lamp-level attribute relationships from the
source data that are lost in top-down energy consumption calculations.
This bottom-up estimate was greater than the recent 2010 Lighting Market Characterization (LMC)
Study residential estimate of roughly 175 TWh. The LMC estimated a larger HOU (1.8 vs. 1.6) and a
lower average number of lamps per home (51.4 vs. 67.4). The average lamp power between the two
studies was very close, with 46 W estimated in the LMC and 47.7 W estimated in this study.
Table 4.1 gives an estimation of the average daily energy consumption (in kWh) for lighting for each
of the 27 domains available in the 2009 RECS, along with estimates of the average number of lamps per
housing unit, average daily HOU per lamp, and average lamp power. The 2009 RECS Mountain North
domain, consisting of Idaho, Montana, Utah, and Wyoming, had the highest estimated average daily
lighting energy consumption per household. This was driven primarily by the high percentage (86
1
Energy consumption was collected through the RECS Energy Supplier Survey for the 2009 calendar year, to be
subsequently linked to the household characteristics data collected in 2010.
4.1
percent) of single-family homes in those states. The next several RECS domains had both high number of
lamps per home and a high average lamp power. New York, California, and the New England states (i.e.,
CT, ME, NH, RI, VT, and MA) had the lowest estimated average household energy consumption for
lighting. These states have lower than average lamp power and/or a higher than average proportion of
multi-family homes, which generally have fewer lamps per household.
Table 4.1. Household Average Daily Energy Consumption (sorted from high to low), Number of Lamps,
Daily HOU per Lamp, and Lamp Power, by RECS Domain
Energy
Consumption
RECS Domain
(Wh)
Number of Lamps
HOU per Lamp
Lamp Power (W)
ID, MT, UT, WY
6,411
85.7
1.5
48.5
MO
6,289
89.0
1.4
53.0
AZ
6,161
75.8
1.5
48.2
KS, NE
5,353
75.8
1.5
51.2
GA
5,330
78.2
1.5
47.1
MI
5,271
71.1
1.6
49.4
CO
5,151
73.9
1.5
47.9
NV, NM
5,143
69.4
1.5
49.3
IA, MN, ND, SD
5,095
77.1
1.5
53.2
IL
5,061
77.6
1.5
53.5
AL, KY, MS
4,987
69.8
1.5
47.8
WI
4,977
73.8
1.5
49.6
IN, OH
4,874
71.2
1.5
51.2
VA
4,752
77.9
1.5
47.1
All US
4,679
67.4
1.6
47.7
FL
4,669
64.3
1.6
47.5
NC, SC
4,654
64.7
1.6
47.5
TN
4,638
64.6
1.6
47.7
TX
4,617
63.4
1.6
47.3
AR, LA, OK
4,598
59.8
1.6
47.6
AK, HI, OR, WA
4,572
65.0
1.6
48.6
NJ
4,551
70.2
1.5
42.4
PA
4,547
69.0
1.6
42.5
DE, DC, MD, WV
4,402
73.3
1.5
46.7
CT, ME, NH, RI, VT
4,289
63.0
1.6
42.9
CA
3,804
57.3
1.6
48.9
NY
3,783
53.6
1.6
40.5
MA
3,405
51.9
1.6
46.5
4.2
4.2 Lamp-Level Attributes
Table 4.2 gives estimates of average lamp counts, daily HOU per lamp, and energy consumption by
dwelling type and RECS domain. Lamp usage (HOU) within dwelling type was generally less variable
across regions than average household lamp counts. In general, single-family homes have more lamps
with lower average daily HOU and energy consumption per lamp than either mobile homes or multi
-
family homes.
Estimates of average lamp counts, daily HOU per lamp, and energy consumption by lamp type and
RECS domain are provided in Table 4.3. Estimates of average HOU for CFL lamps exceeded that of
incandescent and other lamp types, ranging between 1.8 and 2.1 hr per lamp, as compared to 1.0 and 1.3
hr per incandescent and approximately 1.5 hr (in most regions) for other kinds of lamps.
Table 4.4 presents estimates by lighting space type and lamp type. Lamps in kitchens and living
rooms had the highest estimated HOU. CFLs had the highest saturation in bedrooms, bathrooms, and
other rooms, which includes spaces such as closets. Although the estimated HOU of incandescent lamps
tended to be well below that of CFLs, their saturation was considerably higher in all lighting spaces. It
should be noted that estimates for the lighting space types are only for homes with that lighting space
type. Because not all homes have each lighting space type, the sum of the average measures across all
lighting space types will exceed the “All US” measure. Furthermore, estimates for the lighting space
types are for all instances of the given type, rather than per instance. For example, estimates for
bathrooms are for all bathrooms in the home, rather than per bathroom. The companion Microsoft Excel
spreadsheet can be used to view estimates for different bathroom or bedroom categorical levels. A
comparison of the estimates for 0-1, 2, and 3 or more bathrooms can provide both incremental or per
room information as well as some estimation of the effect of increasing home size on lighting usage,
given that number of bathrooms (and number of bedrooms) are good proxies for home square footage.
Table 4.5 provides estimates by lighting space type and fixture location (ceiling vs. non-ceiling).
Overall, U.S. homes have more lamps in ceiling than non-ceiling fixtures on average, but lamps in ceiling
fixtures tend to be used fewer hours per day and consume less total energy per unit per day than lamps in
non-ceiling fixtures.
Table 4.6 also provides estimates by lighting space type and fixture location, but is limited to
dimmable lamps. Because the number of dimmable lamps in garages in the end-use metered dataset was
very small, estimates for garages are not reported. These estimates are the result of the only three-way
cross-classification (Table 3.13) produced in this study, whereby lamp-level results were aggregated by
Lighting Space Type AND Fixture Type AND Control Type. Although ceiling-mounted dining room
fixtures (likely to be dining table chandeliers) have, on average, the second greatest number of dimmable
lamps (after bedrooms), their net usage (in lamp-hours) lags behind dimmable lamps installed in
bedrooms and kitchens.
Figure 4.1 shows estimates of overall average daily lamp usage, by lighting space type and month.
Average daily HOU follows the hours of darkness, highest in December and lowest in June. Although the
overall average daily HOU for lamps has a December peak and June trough, considerable variation exists
across the different lighting space types. The average usage of dining room lamps exhibits the steepest
amplitude, exceeding 2 hr per day in December and January and approaching 1 hr in June. Bathrooms
have an opposite seasonal usage pattern as the overall stock of lamps, with a slight peak in June compared
to other months. Garage lamps and exterior lights have almost no seasonality.
4.3
Table 4.2. Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy Consumption, by Dwelling Type and RECS Domain
Dwelling Type
All Dwelling Types
Single Family
Multi-Family
Mobile Home
4.4
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
All US
67.4
1.6
4,679
85.1
1.5
5,816
24.8
1.7
1,803
38.3
1.9
3,379
CT, ME, NH, RI, VT
63.0
1.6
4,289
81.1
1.5
5,404
24.4
1.7
1,869
36.6
1.9
3,623
MA
51.9
1.6
3,405
81.3
1.5
5,138
19.1
1.7
1,447
31.6
1.8
2,727
NY
53.6
1.6
3,783
81.1
1.5
5,549
28.2
1.7
2,125
34.0
1.9
3,053
NJ
70.2
1.5
4,551
87.0
1.5
5,554
27.0
1.7
1,957
29.4
1.7
2,356
PA
69.0
1.6
4,547
84.8
1.5
5,491
25.4
1.7
1,773
37.2
1.9
3,422
IL
77.6
1.5
5,061
99.9
1.4
6,460
29.0
1.7
2,009
N/A
N/A
N/A
IN, OH
71.2
1.5
4,874
84.5
1.5
5,755
26.1
1.7
1,796
34.1
1.8
2,924
MI
71.1
1.6
5,271
88.1
1.5
6,469
22.9
1.8
1,765
36.8
2.0
3,193
WI
73.8
1.5
4,977
89.3
1.4
5,862
25.4
1.6
2,075
52.2
1.7
5,294
IA, MN, ND, SD
77.1
1.5
5,095
92.1
1.4
5,974
22.2
1.7
1,666
40.5
1.9
3,773
KS, NE
75.8
1.5
5,353
93.8
1.4
6,665
21.6
1.6
1,422
39.6
1.9
2,666
MO
89.0
1.4
6,289
103.8
1.4
7,283
31.6
1.7
2,116
38.9
1.9
3,615
VA
77.9
1.5
4,752
99.8
1.3
5,865
23.8
1.7
1,853
36.1
1.8
2,947
DE, DC, MD, WV
73.4
1.5
4,402
92.5
1.4
5,412
24.4
1.6
1,729
35.1
2.0
2,702
GA
78.2
1.5
5,330
90.5
1.4
6,128
29.8
1.7
2,128
40.7
1.9
3,157
NC, SC
64.7
1.6
4,654
81.7
1.5
5,763
27.1
1.7
1,974
40.5
1.9
3,437
FL
64.3
1.6
4,669
83.5
1.5
5,934
25.0
1.7
1,765
38.6
1.8
3,447
AL, KY, MS
69.8
1.5
4,987
78.7
1.5
5,597
23.1
1.8
1,780
30.2
1.7
2,295
TN
64.6
1.6
4,638
81.9
1.5
5,813
25.0
1.7
1,820
36.4
2.0
3,055
AR, LA, OK
59.8
1.6
4,598
72.8
1.6
5,594
21.5
1.7
1,554
37.0
1.9
3,246
TX
63.4
1.6
4,617
78.3
1.5
5,594
22.3
1.7
1,691
42.8
1.9
4,030
CO
73.9
1.5
5,151
89.5
1.4
6,165
27.8
1.8
2,061
45.8
1.8
3,688
ID, MT, UT, WY
85.7
1.5
6,411
93.9
1.5
6,941
29.0
1.8
2,131
41.8
2.0
4,545
AZ
75.8
1.5
6,161
95.8
1.4
7,749
26.6
1.7
1,843
41.1
2.0
3,894
NV, NM
69.4
1.5
5,143
80.3
1.5
5,933
29.2
1.7
2,008
35.4
1.7
3,422
CA
57.3
1.6
3,804
76.0
1.5
4,959
21.1
1.7
1,482
35.0
1.9
3,312
AK, HI, OR, WA
65.0
1.6
4,572
85.5
1.5
5,878
24.7
1.7
1,753
42.5
1.8
3,954
4.5
Table 4.3. Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy Consumption, by Lamp Type and RECS Domain
Lamp Type
All Lamp Types
Incandescent
CFL
Other Lamp Type
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
All US
67.4
1.6
4,679
41.9
1.2
2,932
14.3
1.9
411
11.2
1.5
1,341
CT, ME, NH, RI, VT
63.0
1.6
4,289
35.4
1.3
2,541
15.8
2.0
485
11.9
1.5
1,264
MA
51.9
1.6
3,405
28.5
1.3
2,067
14.0
2.1
423
9.5
1.5
916
NY
53.6
1.6
3,783
31.3
1.3
2,228
12.8
2.0
390
9.6
1.5
1,166
NJ
70.2
1.5
4,551
43.1
1.2
2,956
15.7
1.9
434
11.4
1.5
1,160
PA
69.0
1.6
4,547
41.6
1.2
2,808
15.4
2.0
435
12.1
1.5
1,304
IL
77.6
1.5
5,061
50.3
1.2
3,451
16.6
1.9
463
10.7
1.3
1,148
IN, OH
71.2
1.5
4,874
48.4
1.2
3,356
11.8
2.0
361
10.9
1.4
1,158
MI
71.1
1.6
5,271
46.6
1.3
3,350
11.9
2.0
339
12.5
1.5
1,582
WI
73.8
1.5
4,977
48.6
1.2
3,761
15.9
1.9
516
9.4
1.4
740
IA, MN, ND, SD
77.1
1.5
5,095
46.8
1.2
3,366
17.4
1.9
486
12.8
1.4
1,243
KS, NE
75.8
1.5
5,353
46.6
1.2
3,171
17.0
1.9
447
12.2
1.4
1,791
MO
89.0
1.4
6,289
53.4
1.2
3,469
18.2
1.8
483
17.5
1.3
2,338
VA
77.9
1.5
4,752
50.7
1.2
3,332
15.9
1.8
375
11.2
1.4
1,046
DE, DC, MD, WV
73.4
1.5
4,402
50.3
1.2
3,295
13.6
1.8
336
9.5
1.4
771
GA
78.2
1.5
5,330
49.5
1.2
3,396
16.1
1.8
438
12.6
1.5
1,515
NC, SC
64.7
1.6
4,654
41.0
1.2
2,907
13.5
1.9
384
10.2
1.6
1,373
FL
64.3
1.6
4,669
40.8
1.2
2,863
13.3
1.9
384
10.3
1.6
1,429
AL, KY, MS
69.8
1.5
4,987
44.1
1.2
3,110
14.2
1.9
407
11.5
1.5
1,471
TN
64.6
1.6
4,638
40.6
1.3
2,881
13.5
2.0
398
10.5
1.5
1,359
AR, LA, OK
59.8
1.6
4,598
38.3
1.3
2,901
12.3
2.0
371
9.2
1.6
1,327
TX
63.4
1.6
4,617
41.4
1.3
3,022
12.9
2.0
370
9.1
1.6
1,230
CO
73.9
1.5
5,151
45.2
1.2
3,054
15.3
1.8
412
13.4
1.4
1,690
ID, MT, UT, WY
85.7
1.5
6,411
52.4
1.2
3,747
17.6
1.9
520
15.7
1.5
2,145
AZ
75.8
1.5
6,161
41.2
1.2
2,837
17.4
1.8
496
17.2
1.6
2,851
NV, NM
69.4
1.5
5,143
41.2
1.2
2,865
15.1
1.8
437
13.1
1.6
1,841
CA
57.3
1.6
3,804
32.3
1.2
2,183
13.8
1.9
417
11.2
1.6
1,204
AK, HI, OR, WA
65.0
1.6
4,572
40.2
1.2
2,816
13.5
1.9
396
11.4
1.5
1,362
4.6
Table 4.4. Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy Consumption, by Lamp Type and Lighting
Space Type
(a)
Lamp Type
All Lamp Types
CFL
Incandescent
Other Lamp Type
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number
of Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Overall
67.4
1.6
4,679
14.3
1.9
411
41.9
1.2
2,932
11.2
1.5
1,341
Bedrooms
15.9
1.2
752
3.9
1.4
87
10.5
1.0
600
1.5
1.0
74
Bathrooms
10.4
1.2
512
2.2
1.4
44
7.4
1.1
435
0.8
1.2
39
Dining Rooms
2.8
1.6
190
0.4
1.8
10
2.3
1.6
175
0.1
1.4
10
Garages
3.2
1.1
121
0.5
1.7
16
1.3
0.5
37
1.4
1.2
80
Hallways
6.0
0.8
170
1.2
1.4
24
4.4
0.7
145
0.4
0.4
4
Kitchens
6.1
2.3
481
1.3
2.5
49
2.9
1.7
272
2.0
2.7
168
Living Rooms
5.5
1.7
472
1.4
2.1
53
3.6
1.6
377
0.5
1.4
50
Other Rooms
7.2
1.3
302
1.5
1.6
33
3.6
1.0
177
2.2
1.4
101
Offices
1.2
1.5
71
0.3
1.5
7
0.7
1.1
47
0.2
1.9
21
Exterior
9.0
2.9
1,610
1.7
3.5
104
5.3
2.0
695
2.1
3.1
965
(a) Estimates are for all instances of the given lighting space type, rather than per instance.
4.7
Table 4.5. Household Average Number of Lamps, Daily HOU per Lamp, and Daily Energy Consumption, by Fixture Type and Lighting
Space Type
(a)
Fixture Type
All Fixture Types
Ceiling
Non-Ceiling
Number of
Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number of
Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Number of
Lamps
HOU per
Lamp
Energy
Consumption
(Wh)
Overall
67.4
1.6
4,679
37.5
1.5
1,935
29.9
1.7
2,745
Bedrooms
15.9
1.2
752
8.7
1.1
381
7.2
1.3
375
Bathrooms
10.4
1.2
512
3.5
1.1
169
7.0
1.3
347
Dining Rooms
2.8
1.6
190
2.4
1.5
163
0.4
1.7
29
Garages
3.2
1.1
121
2.9
1.0
112
0.3
1.2
14
Hallways
6.0
0.8
170
5.0
0.8
138
1.0
0.9
35
Kitchens
6.1
2.3
481
4.9
2.2
405
1.2
2.4
79
Living Rooms
5.5
1.7
472
2.7
1.6
206
2.8
1.8
270
Other Rooms
7.2
1.3
302
5.6
1.2
209
1.7
1.4
95
Offices
1.2
1.5
71
0.8
1.4
44
0.5
1.6
29
Exterior
9.0
2.9
1,610
1.1
2.8
139
7.9
3.0
1,493
(a) Estimates are for all instances of the given lighting space type, rather than per instance.
4.8
Table 4.6. Household Average Dimmed Number of Lamps, Daily HOU per Lamp, Daily HOU for all Lamps, and Nominal (Un-dimmed) Lamp
Power, by Fixture Type and Lighting Space Type
(a)
Fixture Type
All Fixture Types
Ceiling
Non-Ceiling
Number of
Dimmed
Lamps
HOU per
Dimmed
Lamp
HOU all
Dimmed
Lamps
Lamp
Power
(W)
Number of
Dimmed
Lamps
HOU per
Dimmed
Lamp
HOU all
Dimmed
Lamps
Lamp
Power
(W)
Number of
Dimmed
Lamps
HOU per
Dimmed
Lamp
HOU all
Dimmed
Lamps
Lamp
Power
(W)
Overall
2.8
1.6
4.3
57.1
2.2
1.5
3.2
46.1
0.6
1.7
0.9
70.5
Bedrooms
0.7
1.2
0.8
59.0
0.6
1.1
0.6
50.5
0.2
1.3
0.2
75.3
Bathrooms
0.3
1.2
0.4
48.3
0.1
1.1
0.1
49.0
0.2
1.3
0.3
47.3
Dining Rooms
0.4
1.6
0.6
45.2
0.4
1.5
0.6
42.8
*
1.7
*
53.1
Garages
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
Hallways
0.2
0.9
0.2
44.5
0.2
0.77
0.1
50.2
*
1.0
*
35.7
Kitchens
0.3
2.3
0.8
43.2
0.3
2.2
0.7
43.3
*
2.4
0.1
36.8
Living Rooms
0.4
1.7
0.6
78.1
0.3
1.6
0.5
52.6
0.1
1.8
0.1
99.6
Other Rooms
0.3
1.3
0.4
51.1
0.3
1.2
0.4
41.9
*
1.4
*
77.9
Offices
0.1
1.5
0.1
53.4
0.1
1.4
0.1
43.3
*
1.6
*
92.2
Exterior
*
2.9
0.1
53.0
*
2.8
*
55.3
*
3.0
0.1
46.9
(a) Estimates are for all instances of the given lighting space type, rather than per instance.
* Estimate less than 0.1
4.9
Figure 4.1. National Estimates of Average Daily HOU per Lamp, by Lighting Space Type and Month
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Exterior"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Other"Room"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Office"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Living"Room"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Kitchen"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Hallway"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Garage"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Dining"Room"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Bathroom"
0"
1"
2"
3"
Jan"
Feb"
Mar"
Apr"
May"
Jun"
Jul"
Aug"
Sep"
Oct"
Nov"
Dec"
HOU$
Overall" Bedroom"
5.1
5.0 Accuracy and Validity of the Estimates
As discussed in previous sections, the estimates presented in this report and in the companion
spreadsheet are the result of an analysis linking estimates from four related studies. Each of these studies
was designed to be unbiased and achieve a certain degree of accuracy within its original domain. This
study uses various methods to extend data from these studies to other domains. The scope of each study
was limited to specific geographic regions or in the collected lighting information:
• The CA RLMS ANCOVA model relates lighting HOU to within-household unit and demographic
characteristics, for California households only. The model has not yet been calibrated to account for
this relationship in other regions of the country.
• The 2009 RECS did not collect room-level inventory of fixture characteristics or usage data from
direct metering.
• The 2009 AHS did not collect fixture characteristics and usage data within rooms of homes, and
geographic identifiers at finer levels than the four census regions were not made public.
• The Multi-State CFL Modeling Study consisted of several regional studies, corresponding with states
or electric utility service territories. The collection of these regional studies had very good coverage
in some areas of the United States and very little coverage in others. Although the studies had both
overall household and room-level luminaire characteristics, direct metering of lamps was not part of
the study protocols.
The accuracy and validity of the estimates presented in this report and the companion spreadsheet are
largely dependent on how representative the data collected in those restricted studies is for extension to
the housing stock in other geographic areas, and the validity of the assumptions underlying the methods
used to make those extensions. The bias in the estimates created by the data linking processes used is
unknown. There are, however, controls in place that flag estimates derived from fewer than ten RECS
households or having a coefficient of variation associated with an HOU estimate greater than 50 percent.
These data quality flags have been used in the 2009 and previous RECS. The CA RLMS generated HOU
estimates with an overall 16 percent margin-of-error at 90 percent confidence
1
, which would serve as a
best case when applying this model to linked data from other studies. Estimates generated here would
necessarily be less accurate if regional inventory data used in the study were not reflective of the
population or if lighting usage behavior varied significantly in different parts of the United States.
5.1 Comparison with CA RLMS Estimates
A simple method for validating the methodology used in this study is to compare the estimates for
California produced by this study with those from the CA RLMS study. Although the CA RLMS relied
solely on its sample of households to estimate HOU, this study leveraged information from various
sources, including the RECS, the AHS, the multi-state CFL studies, and the CA RLMS itself.
Nevertheless, the estimates for California from both studies are very similar. Table 5.1, Table 5.2, and
Table 5.3 show the estimates of HOU from this study along with the CFL estimates from the CA RLMS.
2
1
Upstream Lighting Program Evaluation Report, Volume 1, Table 84.
http://www.calmac.org/publications/FinalUpstreamLightingEvaluationReport_Vol1_CALMAC_3.pdf
2
Upstream Lighting Program Evaluation Report, Volume 1, Table 85.
http://www.calmac.org/publications/FinalUpstreamLightingEvaluationReport_Vol1_CALMAC_3.pdf
5.2
Table 5.1. Daily HOU by Lamp Type for California
Lamp Type
Daily HOU
DOE
CA RLMS ± 90% C.I.
CFL
1.9
1.9 ± 0.3
Incandescent
1.2
n/a
Other
1.6
n/a
Table 5.2. Daily CFL HOU by Dwelling Type for California
Dwelling Type
CFL Daily HOU
DOE
CA RLMS ± 90% C.I.
Single Family
1.8
1.8 ± 0.3
Multi-Family
2.1
2.0 ± 0.3
Mobile Home
2.1
1.9 ± 0.3
Table 5.3. Daily CFL HOU by Space Type for California
Space Type
CFL Daily HOU
DOE
CA RLMS ± 90% C.I.
Bathroom
1.4
1.3 ± 0.3
Bedroom
1.4
1.5 ± 0.3
Dining Room
1.8
1.7 ± 0.4
Exterior
3.5
3.8 ± 0.3
Garage
1.7
1.8 ± 0.5
Hall
1.4
1.4 ± 0.3
Kitchen
2.5
2.3 ± 0.3
Living
2.1
2.3 ± 0.3
Office
1.5
1.5 ± 0.4
Other
1.6
1.9 ± 0.3
Overall
1.9
1.9 ± 0.3
5.2 Statistical Precision of Estimates
The coefficient of variation (CV) is defined as the standard error of the estimate divided by the
estimate itself.
!" =
!(!)
!
where !(!) is the standard error for lighting characteristic µ and ! is the estimate of that lighting
characteristic.
5.3
To estimate the standard error of the lighting estimates from the composite sample, a decomposition
of variance was used. The ANCOVA model can be represented as:
!
!
= !
!
!!
where !
!
= the unobserved lighting usage measure for household j in the composite dataset
!
!
= the model covariate vector for household j in the composite dataset
! = the vector of estimated coefficients from the CA RLMS ANCOVA.
Then, for any domain of interest, the average lighting usage measure !
!
is estimated by:
!
!
= !
!
!!
!
!
!
!
!
= !
!
!!
!
!
!
!
!
!
! = !
!
!!
!
!
!
!
!
! = !
!
!
where !
!
= the sample expansion weight for household j in the composite sample
!
!
= the vector of model covariate means for the expanded population, estimated by !
!
across the composite sample.
In this application, the vector of model covariate means for the expanded population, !
!
, is
independent of the estimated coefficient ! from the CA RLMS ANCOVA. Therefore, the following
approximation can be used:
!"# !
!
≅ ! !
!
!"# !
!
! + ! !
!
!"# ! !
!
!
The first term on the right-hand side is the contribution of the variance estimate of !
!
from the
composite sample variability, treating the vector of ANCOVA coefficients ! as fixed. The second term is
the contribution of the variance estimate of the vector of ANCOVA coefficients, treating the estimated
mean variable vector !
!
as fixed.
The first term can be estimated directly using the RECS half-sample variance estimation methods
1
and the estimated coefficient ! from the CA RLMS. The second term is computed using the results of the
ANCOVA model fit on the CA RLMS data, as well as the estimated composite data variable of model
covariate means, !
!
, treated as fixed.
5.3 Sources of Bias and Variability Introduced in the Estimates
Other sources of estimation error could impact the estimation process related to the representativeness
of the multi-state data with respect to other households of the same dwelling type and other data linkage
variables. The incorporation of these types of error in the overall standard error computations is beyond
the scope of this study, but the standard error formula above gives an approximation of the total
estimation error. The following section lists possible impacts on the bias and variance of the estimates
presented above, due to aspects of the estimation methodology implemented in this study.
1
Half-sample weights and primary sample unit group indicators are not included on the RECS public use microdata
file. Thus, they were estimated using the full-sample weights and applying a ratio adjustment to account for the
impact of the clustering design on the variance estimates.
5.4
1. Applying CA RLMS ANCOVA to other areas
a. Bias: Regional differences in patterns of equipment ownership and use result in biased estimates.
The relationship between lighting usage and household and lighting source characteristics in
California may be very different than in other regions of the country.
b. Variance: mainly captured in the RECS-type variance estimate, as discussed.
2. Imputing lighting inventory using the inventory samples and the AHS room-type distributions
a. Bias: For comprehensive inventory samples, bias is mainly due to geographic differences
between the originally represented area and the areas the data are applied to.
b. Variability: For some multi-state study areas, the level of variability for certain subgroups of the
population, such as mobile homes or very large homes, may be lower in the samples of inventory
data than in the population.
3. Imputing room and lighting space type using AHS assignments
a. Bias: Probably minimal for aggregates at the census region level. If room-type distributions are
very different within a region, in particular if the mix of high- and low-use rooms is very
different, estimates for these finer areas will be systematically misstated.
b. Variability: The variability from AHS microdata is mainly the sampling variability of AHS itself.
As noted, some of this is captured in the observed variability.
5.4 Opportunities for Improving Estimates
The estimates generated by this study could be readily improved by the availability of new regional
data meeting defined pre-conditions and funding for its analysis. New end-use metering data would
improve the accuracy for not only the geographic region the data came from and all RECS domains
inclusive of that region, but likely also for nearby domains and possibly nationally.
For example, if New York conducted a statewide residential lighting inventory study that collected
measures required by the HOU model, the estimation framework would be updated by replacing the
inventory data previously used to represent New York, assuming this new data was determined to be an
improvement of that which was previously used. This would, at minimum, improve the study estimates
for New York, the Mid-Atlantic census division, the Northeast census region, and the United States. To
the extent it could be determined that the new data was representative of other states or RECS-reportable
domains listed in Table 3.11, the data could improve the estimates in other states. If a state was in a
RECS-reportable domain that also included one or more other states, such as Connecticut, that conducted
a statewide inventory study with relevant demographics to the HOU model, it could be possible to “break
away” Connecticut from the other states in the RECS-reportable domain. This would depend on the
sample design and other factors.
Continuing the example scenario, another level of improvement could be achieved if New York
decided to instead conduct a statewide residential lighting metering study, collecting household
characteristics, lighting inventory, and end-use metering data for each household following the protocols
used in the CA RLMS. In addition to updating the estimation framework as described above, the HOU
model could be calibrated for representative regions using the new metering data. This would eliminate
potential bias in the existing New York estimates due to behavioral differences in how New Yorkers use
5.5
lighting in their homes as compared with Californians. Additional analysis would be conducted to
determine the most appropriate way to use the New York and California metering data to improve the
accuracy for applications of the HOU models in states other than New York and California.
Two datasets, available now or in the near future, have been identified that meet the requirements for
incorporation into the estimation framework:
1. Household characteristics, lighting inventories, and end-use metering data were collected over
6 months in late 2012 from 183 households in the Mid-Atlantic census division, in an effort managed
and executed by the authors of this study.
2. Household characteristics and lighting inventories were collected from over 1,850 households
together with end-use metering data from a 101 household subset in the Pacific census division as
part of the Northwest Energy Efficiency Alliance Residential Building Stock Assessment.
1
The availability of additional funding for analyzing these datasets would lead to improved accuracy in
multiple RECS domains.
1
http://neea.org/resource-center/regional-data-resources/residential-building-stock-assessment