The German
Residential Energy Consumption Surveys
Rheinisch-Westfälisches Institut für Wirtschaftsforschung
(RWI)
forsa
Gesellschaft
für
Sozialforschung
und
statistische
Analysen mbH
Extended Abstract for Research Project No. 54/09 commissioned
by Federal Ministry of Economics and Technology and prepared for
MESH
August 2012
Team:
RWI: Prof. Dr. Manuel Frondel (Project Leader), Nolan Ritter, Dr.
Harald Tauchmann, Prof. Dr. Colin Vance
forsa: Dr. Peter Matuschek, Ute Müller
Executive Summary
This document describes the salient methodological aspects of a
series of surveys commissioned by the Federal Ministry of Economics and Technology to estimate the energy consumption of German
private households. In an on-going effort to establish an energy
panel in Germany, the most recent documented survey covers the
energy consumption of 6,700 households nationwide from the
years 2006 to 2008 , with the target years for the first and second
waves having been 2003 and 2005, respectively. Data for two additional waves for the years 2009 and 2010 were gathered at the
outset of 2012 and are currently being evaluated, with the results
documented in a report that will be available by the end of 2012.
The assembled data set is the only panel of its kind that covers
energy consumption of households over multiple years in Europe.
Among other things, such as the individual households’
consumption of fossil fuels, these surveys elicit information on
housing conditions, the characteristics of the residential dwelling,
and the endowment with electric appliances. This information
serves as the foundation to determine the consumption of various
fuels for residential purposes and due to private car usage. The
sample estimates were extrapolated to derive consumption figures
for the entire population of private households in Germany.
Special attention is devoted to renewables, which have
gained increasing importance in the residential sector. In order to
gauge the usage of renewable energy technologies in the household sector, a telephone survey from among some 80 000 households was conducted to obtain a precise picture of the status quo
at the end of 2006. From this experience, we learned that it would
suffice to use a much smaller sample of households. To update this
picture, at the outset of 2012, we merely surveyed 30,000 households.
While the major task of these surveys is to gain a deeper
understanding of the energy consumption behavior of the interviewed households, a new survey called the “Energy Diary” was
recently designed in order to record the equipment and usage of
electric appliances on a regular basis. The first wave of the gath2
ered information was used to estimate the electricity consumption
for various purposes, such as water heating, lighting, and usage of
appliances.
The project has yielded several scientific publications, working papers, and project reports, many of which are available for
download from the project web site at:
http://www.rwi-essen.de/forschung-und-beratung/umwelt-undressourcen/projekte/39/.
Sampling
The samples are drawn from the population of German-speaking
residents aged 14 to 69, with the aim of assembling a representative panel. Recruitment proceeds according to a multistage, random-sampling procedure based on computer-supported telephone
interviews. For the 2006-2008 survey, for instance, an initial sample of 9,575 households was selected, of which 6,715 ultimately
participated in the survey, yielding a participation rate of roughly
70%. The spatial distribution of households, presented in Figure 1
by way of zip codes, spans across Germany and is particularly concentrated in regions of high population density, such as the capital
Berlin.
3
Figure 1: Survey coverage
Survey instrument
Each participating household is equipped with an interface that
projects the questions on the respondent’s television screen and
allows an easy implementation of complex questionnaires by filter
techniques and visual assistance. Respondents may answer the
questions at any time of their choosing; they are not under time
pressure and are not obligated to complete the questionnaire in a
single session. The forsa tool, which immediately saves the collected information on a server, further allows for automatic consistency and validity checks during the data input by the participant.
This ensures that the collected data is of high quality.
Notoriously critical points for surveys on energy consumption are the complex heating bills for dwelling in multi-family
homes with central heating. To address this challenge, an interview tool was developed that draws extensively on the visualizing
and filter abilities of the forsa tool. After a household has indicated
4
its respective billing company, the survey procedure subsequently
presents an exemplary bill issued by the company. The tool goes
step by step through the details of the exemplary bill, while highlighting the parameters of interest (see Figure 2). By this means,
both the error rate and drop-out quota of households living in multi-family houses with central heating were reduced substantially.
Figure 2: Screenshot of an exemplary bill
To glean an impression regarding the validity of the data obtained
from this approach, billing information for a sub-sample of households from the year 2005 was requested from the firm that issued
the bill. A comparison of the original bill with the exemplary bill
used in the survey revealed a high degree of validity, providing
some assurance of the reliability of the information.
Measuring consumption is particularly challenging for stockable fuels, such as fuel oil. Typically, households are able to provide information about the purchased quantity, but the delivered
amount is likely to deviate from the quantity that is actually consumed. To mitigate such problems, for the target years 2006 to
2008, for instance, the households’ deliveries were surveyed for
the period from 2005 to 2009, and the household’s fuel consumption was calculated using information of heating-degree days
(HDD) of the corresponding years. The household-specific climate
conditions were obtained by a grid of climate stations operated by
Deutscher Wetterdienst, and geographically interpolated to the
households’ places of residence.
5
By summing the quantity of fuel delivered while simultaneously taking into account the HDD, it is possible to estimate the
yearly consumption for a household according to the following
formula for the year 2006:
2009
Consumption2006,i  HDD2006,i *
 Quantity Delivered
t  2005
t ,i
2009
 HDD
t  2005
t ,i
where HDDt ,i is the household-specific heating degree days in year
t at the location of household i, which is in turn calculated as the
sum of household-specific degree days within the year t:
HDD t ,i   Gd ,i
d t
Following DIN 3807, a degree day (Gd,i) is calculated as the difference
between
an
assumed
average
room
temperature
of
20° Celsius and the average daily outdoor temperature at the location of household i, conditional on an average temperature (Ad,i) of
less than 15° Celsius:
Gd ,i
(20 C  Ad ,i ) for Ad ,i  15 C

0
for Ad ,i  15 C

Consumption data and conversion factors
Consumption data are entered for each fuel according to the typical units of measurement. Heating oil, for example, is recorded in
liters, while coal consumption is recorded in kilograms. In some
cases, there may be several measurement units for the same fuel.
For example, the consumption of liquefied gas may be measured in
kilograms or by volume in liters.
6
To establish comparability and enable summation over different fuels like oil or gas, it is necessary to convert the various
measurement units into a single common unit. We use the unit
kilowatt hour (kWh), which is a standard unit for representing
household level and per square meter energy consumption. When
extrapolating the results, the unit Petajoule (PJ) is instead used to
maintain conformity with the Law on units in measurement, according to which 1 billion kWh=1TWh=3.6 PJ.
While the consumption of electricity, district heating, and
natural gas is generally measured in kWh or similar energy units,
Table 1 presents the conversion factors applied for other fuels that
are usually measured by weight or volume. Recognizing that different countries in Europe apply different conversion factors, it is
noted that those applied here are largely based on the heating
values of the Arbeitsgemeinschaft Energiebilanzen (AGEB 2010).
7
Table 1: Employed Heating Values of different Fuels in kWh
Maßeinheit
Gas
kWh
m3
8,816
Liter
10,030
kg
11,870
Liter
6,627
kg
12,944
Brown coal
kg
5,448
Black coal
kg
8,723
Wood pellets
kg
4,900
Wood briquettes
kg
4,900
Wood chips
kg
3,976
Wood slabs
kg
3,976
Heating oil
Liquified gas
Data cleaning
Incorrect responses, of course, are inherent to any data collection
effort. The identification and correction of false entries is consequently an essential component in the evaluation of survey data.
One of the key challenges when collecting data on energy demand
is to distinguish incorrectly entered values from those that truly
have an unusual magnitude. This issue is addressed by applying
an iterative data cleaning process that simultaneously references
two values, the energy consumption per area of living space, expressed in kWh/m2, and the costs per kilowatt hour, expressed in
Cent/kWh. The procedure identifies an observation as an outlier
when both of these values lie outside of a predefined plausibility
interval. Under the assumption that the majority of households
entered correct data, this interval was defined by taking the arithmetic mean of the sample and adding and subtracting two standard deviations. Observations for which the fuel consumption and
cost figures both lie outside the interval are excluded from the data set.
This process is then repeated with the newly created sample, recognizing that the originally created plausibility interval is no
longer valid. Based on the mean of the revised sample, a new in-
8
terval is created that is narrower than the original because of the
smaller standard deviation. In this way, observations that were
maintained after the first pass may now be designated as outliers
with the subsequent pass and removed from the data. The process
is continually repeated until no values fall outside the interval.
The described data cleaning procedure is undertaken for
each fuel and energy source separately and stratified by single-,
two-, and multi-family residences. The boundaries of the plausibility interval are then defined for each energy source within the particular stratum. For the special case of electricity, the stratification
is by household size, except when electricity is used for heating.
Weighting and corrections for non-response
Even when the original sample is representative of the overall
population, there will always be cases of non-response, nonparticipation, and incomplete questionnaires that result in a final
sample that is no longer representative. In the 2006-2008 survey,
for example, single-person households were more heavily represented in the original sample than in the final, presumably owing
to the typically longer working hours of individuals who live alone,
which prevented them from completing the questionnaire. A
weighting scheme was developed to correct for over- or underrepresentation, with the weights chosen such that the final
sample maintained the same proportionality as the German Microcensus with respect to household size and residential location. The
weighting thus ensured that the final sample captured a representative picture of the population.
An additional source of systematic bias may emerge from
the fact that certain segments of the population have a higher
probability of answering questions related to energy consumption,
leading to self-selection bias. Self-selection might occur if households who are hardly aware of their energy consumption are also
less diligent with keeping their energy bills, and are therefore unable to quantify their energy consumption. Consequently, the sample would consist systematically of households with a low consumption, and the “true” residential energy consumption would be
underestimated.
9
For the example of the year 2008, Table 2 presents the total number of users and the number (share) who provided responses on consumption. To the extent that these shares do not
comprise randomly selected households, the obtained consumption
information will not reflect that of the population, yielding biased
estimates.
To overcome possible problems with self-selection effects in
the data, econometric discrete choice models were used to derive
a weighting scheme. The dependent variable was defined as 1 if a
response to a certain question was provided and zero otherwise.
Thus, for each household the model estimated the probability that
the household provided data on the consumption of a certain fuel.
The explanatory variables in the model included socioeconomic
characteristics such as age, sex, nationality, education level, employment status, and income. The inverse of the estimated probabilities ultimately served as an item non-response correction,
whereby households with a low probability receive a correspondingly high weight.
Table 1: Total Users and Responses by Energy Source for 2008
Total Users
Responses (Share)
Electricity
6,715
2,711
(40,4 %)
Gas
3,389
894
(26,4 %)
Heating Oil
1,733
474
(27,4 %)
District Heating
808
74
(9,2 %)
Brown Coal
240
149
(62,1 %)
Liquified Gas
203
126
(62,1 %)
88
56
(63,6 %)
Hard Coal
Extrapolation
Generally, all results obtained by a sample survey entail a certain
degree of statistical uncertainty, which has implications for the
extrapolation procedure. By definition, a sample consists of only a
fraction of the population, and the sample estimates will therefore
randomly deviate from the parameters in the population. Therefore, in addition to the estimated average consumption, our sur10
veys provide for 95% confidence intervals, which are defined to
include the true, but unknown population parameter with a probability of 95%. These confidence intervals are indispensible when
extrapolation results should be compared to results of other data
sources, for instance to the consumption figures published in the
German energy balances.
Usage of Renewable Energies
An extensive computer-assisted telephone survey with respect to
the market penetration of renewables in households was conducted in the fall of 2006. In total, more than 80,000 households were
interviewed about whether they use a heat pump, solar heating,
wood pellets, or a photovoltaic device. Based on these results for
2006, the market penetration of renewables in 2007 and 2008 was
updated by the developments in the sample of the 6,700 participating households of the forsa panel. For instance, if the number
of solar heating systems rose by 2% between 2006 and 2007 in
the forsa household panel, the telephone survey results were updated in a similar manner.
According to Figure 3, about 12% of all single-family homes
and 11% of all two-family homes were equipped at the end of
2006 with solar heating. Heat pumps and wood pellets were used
in less than 6% of all single- and two-family homes. Basically, no
difference between single- and two-family homes existed for the
usage of photovoltaic: about 4.5% of these houses are equipped
with such a facility. Renewable technologies are generally much
less present in apartment houses, since the building owner typically does not benefit from the installed systems.
11
Figure 3: Usage of Renewable Technologies in Households 2008
14%
12%
10%
8%
6%
4%
2%
0%
Solar Heater
Heat Pump
Single-Family
Photovoltaic
Two-Family
Wood Pellets
Apartment
Decomposition of Electricity Consumption: the Energy Diary
Electricity is probably the most versatile fuel in the residential sector. Electricity can be used for various purposes, for instance lighting, cooking, hot water preparation, and the usage of electric appliances. Typically, only the total electricity consumption of a
household is known, if at all. In order to differentiate electricity
consumption into single usage purposes, an additional survey
called “Energy Diary” was designed to collect information about the
equipment and usage of electric appliances for a sample of almost
900 households. The “Energy Diary” is conducted on a more or
less regular basis, with the same set of households. By this means,
the gathered information provides for reliable insights into the
electricity consumption behavior of the residential sector. So far,
there are data from four waves originating from surveys performed
in August and November 2010, as well as in March and August
2011.
Dissemination of findings
The survey data has already yielded several scientific publications
and project reports that address various dimensions of energy
consumption. Grösche and Schröder (2011), for example, explore
the willingness-to-pay for alternative fuel mixes, with an eye to12
ward gleaning the level of public support for renewable energies.
Grösche and Vance (2009) address the issue of home retrofits,
specifically quantifying the extent to which public policies to support energy-saving retrofitting suffer from free-ridership. Also addressing the topic of retrofitting, a related paper by Frondel,
Grösche, and Schmidt (2008) explores the effectiveness of home
audits. Other publications from the project document the survey
instrument and data collect in both English (RWI and forsa, 2011a)
and German (RWI and forsa, 2011b). These can be downloaded
from the project web site, http://www.rwi-essen.de/forschungund-beratung/umwelt-und-ressourcen/projekte/39/.
13
References
AGEB (2010) Heizwerte der Energieträger und Faktoren für die
Umrechnung von spezifischen Mengeneinheiten in Wärmeinheiten, Stand 26.08.2010, Arbeitsgemeinschaft Energiebilanzen,
Berlin, http://www.ag-energiebilanzen.de.BDEW (2009), EEGJahresabrechung
2008,
Stand:
27.07.2009,
http://www.bdew.de/bdew.nsf/id/DE_EEG-Jahresabrechnungen
BDEW (2008), EEG-Jahresabrechung 2007, Stand: 26.09.2008,
http://www.bdew.de/bdew.nsf/id/DE_EEG-Jahresabrechnungen
BDEW (2007), EEG-Jahresabrechung 2006, Stand: 21.09.2007,
http://www.bdew.de/bdew.nsf/id/DE_EEG-Jahresabrechnungen
Frondel, M., P. Grösche und C. M. Schmidt (2008), Energiesparen:
Warum
wird
Beratung
gefördert?.
Zeitschrift
für
Ener-
giewirtschaft 32 (2): 97-101.
Grösche, P. und C. Schröder (2011), Eliciting Public Support for
Greening the Electricity Mix Using Random Parameter Techniques. Energy Economics 33 (2): 363-370.
Grösche, P. und C. Vance (2009), Willingness-to-Pay for Energy
Conservation and Free-Ridership on Subsidization – Evidence
from Germany. Energy Journal 30 (2): 135-154.
RWI und forsa (2011a), The German Residential Energy Consumption Survey 2006–2008. RWI Projektberichte
RWI und forsa (2011b), Erhebung des Energieverbrauchs der privaten Haushalte für die Jahre 2006–2008. RWI Projektberichte
14