Elsevier

Energy Policy

Volume 131, August 2019, Pages 370-379
Energy Policy

Heterogeneity in German Residential Electricity Consumption: A quantile regression approach

https://doi.org/10.1016/j.enpol.2019.03.045Get rights and content

Highlights

  • This paper estimates the electricity consumption rates of individual appliances.

  • We draw on data of the German Residential Energy Consumption Survey (GRECS).

  • We combine the conditional demand approach with quantile regression methods.

  • Our results indicate substantial differences in consumption rates and end-use shares.

Abstract

In the absence of sufficient coverage of metering data on the electricity consumption of individual devices, this paper estimates the contribution of individual appliances to overall household electricity consumption, drawing on the most recent wave of the German Residential Energy Consumption Survey (GRECS). Moving beyond the standard focus of estimating mean effects, we combine the conditional demand approach with quantile regression methods to capture the heterogeneity in electricity consumption rates of individual appliances. Our results indicate substantial differences in these rates, as well as the end-use shares across households originating from the opposite tails of the electricity consumption distribution. This outcome highlights the added value of applying quantile regression methods in estimating consumption rates of electric appliances and indicates some scope for realizing conservation potentials.

Introduction

Growing concern about climate change has incited widespread consensus about the need for political action and an intense debate on mitigation measures. In fact, policymakers all around the world have stipulated programs to mitigate climate change. For instance, the European Union (EU) strives for a 40% reduction in greenhouse gas emissions by 2030 relative to 1990, which to a large extent is expected to be achieved by the electricity generation sector. Since households consume a substantial share of electricity, about 30% in the EU (Eurostat, 2018), spurring households to curtail their consumption appears to be a promising approach to reach emission reduction targets.

Although there are a few studies that resort to smart metering technologies that allow measuring the electricity consumption of individual appliances (e.g. Schleich et al., 2013; Chen et al., 2015), little evidence exists on the amount of electricity used by different purposes (for a review of factors that affect electricity consumption, see Jones et al., 2015). To close this void, empirical studies are required that infer a household's total electricity consumption from both the stock of electrical appliances and the electricity consumption rates of individual appliances.

In the absence of sufficient coverage of metering data on the electricity consumption of individual devices, which presumably will not become standard for at least another decade, empirical studies necessarily resort to econometric methods, such as the widely used conditional demand approach (CDA) as suggested, for instance, by Parti and Parti (1980), Aigner et al. (1984), and Lafrance and Perron (1994) and more recently applied by Larsen and Nesbakken (2004) and Dalen and Larsen (2015). This approach draws on dummy variables that indicate the ownership of electric appliances, such as washing machines and dishwashers, and rests on the idea that the corresponding coefficients can be interpreted as the mean electricity consumption related to each type of appliance.

Based on a unique data set originating from the most recent wave of the German Residential Energy Consumption Survey (GRECS) and a subsequent survey on the individual stock of electrical appliances among a subsample of about 2100 households, this paper investigates the heterogeneity in household electricity consumption, which is due to differences in appliance stocks and consumption behavior, by employing quantile regression methods and combining them with CDA. Refining the approach of Larsen and Nesbakken (2004) and Dalen and Larsen (2015) by explicitly including variables that indicate the frequency of appliances and the intensity of use instead of mere appliance ownership, we estimate bandwidths for the electricity consumption rates of individual appliances, thereby accounting for both user behavior and the heterogeneity in electric appliance stocks of households. In addition, we gauge the shares of diverse end-use purposes for households located in different parts of the electricity consumption distribution.

In contrast to previous studies that employ quantile regression methods, our analysis is based on consumption rather than expenditure data (e.g. Huang, 2015) and on the appliance stock, rather than household attributes alone (e.g. Kaza, 2010; Valenzuela et al., 2014; Yao et al., 2014). While an advantage of the CDA is that electricity consumption rates of appliances and end-use shares are directly estimated (Larsen and Nesbakken, 2004), some CDA studies find implausible or even negative consumption rates of electric appliances (e.g. Caves et al., 1987).

Our analysis demonstrates large heterogeneity in residential electricity consumption, which is even evident for households of the same size. It may not only reflect differences in appliance stocks and intensities of use, but also significant discrepancies in both the electricity consumption rates of the appliances and heterogeneous consumer behavior. It turns out that employing quantile regression methods allows for eliciting the spectrum of consumption rates for each type of appliance, which covers the whole range from less energy-efficient to highly efficient appliances. In addition, while these results clearly reflect correlations, rather than causal relationships, we find substantial differences in the end-use shares across households originating from the opposite tails of the electricity consumption distribution, highlighting the added value of applying quantile regression methods in estimating consumption rates of electric appliances.

Furthermore, our results indicate energy saving that can be realized by focusing conservation policies on particular appliances, such as refrigerators. For example, encouraging the purchase of energy-efficient models through subsidies can help to reduce both electricity consumption and related greenhouse gas emissions. Another large saving potential is found for lighting, which is caused by both less efficient light bulbs and inefficient utilization. This calls for a more intensive information policy on energy efficiency that aims at both revealing conservation potentials and on behavioral changes. The latter is important given the existence of the so-called rebound effect that undermines the realization of savings, at least to some degree (Sorrell et al., 2009).

The following section describes the data set underlying our analysis. Section 3 presents the methodology, followed by a presentation of the estimation results in Section 4 and of end-use shares in Section 5. The last section summarizes and concludes with policy recommendations.

Section snippets

Data

To estimate the consumption rates of households' electrical appliances, we draw on data obtained from two surveys that were conducted jointly by RWI – Leibniz Institute for Economic Research and the professional German survey institute forsa. As part of the German Residential Energy Survey (GRECS) that was established in 2005 (RWI and forsa, 2005), the first survey took place at the outset of 2014 and gathered data on the electricity consumption of 8500 private households, as well as

Methodology

The conditional demand approach (CDA) employs data on appliance stocks to quantify the effect of an appliance type on the electricity consumption level, conditional on possessing this appliance. In CDA studies (e.g. Hsiao et al., 1995; Halvorsen and Larsen, 2001; Larsen and Nesbakken, 2004; Reiss and White, 2005; Dalen and Larsen, 2015), dummy variables Dij play a key role in explaining the electricity consumption yi of household i, where Dij equals unity if household i possesses appliance j

Results

As a starting point of our analysis, we analyze the conditional demand model by estimating Equation (1) via OLS. We find that few coefficients ρjm and ρkm of the interaction terms between the appliance dummies Dij or the count variables Nik with the household characteristics Cim are statistically different from zero. Furthermore, for most coefficients, the inclusion of these interaction terms has only a negligible bearing on the other coefficient estimates (see Table A2 in the appendix). More

End-use shares

Using the quantile regression estimates reported in Table 4 and the formulae sˆj(τ):=yˆj(τ)y(τ) and sˆk(τ)=yˆk(τ)y(τ) explained at the end of the methodology section, we now present the shares of electricity consumption that can be attributed to diverse end-use purposes for households belonging to different parts of the consumption distribution (Fig. 4, Fig. 5, Fig. 6). Note that heating purposes do not appear in these figures, as households heating solely with electricity were not invited to

Conclusion and policy implications

Employing the conditional demand approach and combining it with quantile regression methods, this paper has estimated the contribution of common household appliances to electricity demand from a sample of about 2100 German households. Moving beyond the standard focus of estimating mean effects via OLS, we have applied quantile regression methods to capture the heterogeneity in the electricity consumption rates of individual appliances across quantiles of the electricity consumption

Acknowledgements

For helpful comments and suggestions, we are grateful to three anonymous reviewers and participants of the EAERE Conference 2016 in Zurich, Switzerland, and the 7th Atlantic Workshop on Energy and Environmental Economics in A Toxa, Spain. Furthermore, we gratefully acknowledge financial support by the Federal Ministry of Education and Research (BMBF) under grant 03SFK4B0 (Kopernikus Project ENavi) and the German Association of Energy and Water Industries (BDEW), as well as the German Council

References (30)

  • S. Sorrell et al.

    Empirical estimates of the direct rebound effect: a review

    Energy Policy

    (2009)
  • C. Valenzuela et al.

    An analysis of monthly household energy consumption among single-family residences in Texas, 2010

    Energy Policy

    (2014)
  • X.-L. Yao et al.

    A quantile approach to assess the effectiveness of the subsidy policy for energy-efficient home appliances: evidence from Rizhao, China

    Energy Policy

    (2014)
  • D.J. Aigner et al.

    Conditional demand analysis for estimating residential end-use load profiles

    Energy J.

    (1984)
  • D.W. Caves et al.

    A Bayesian Approach to Combining Conditional Demand and Engineering Models of Electricity Usage

    (1987)
  • Cited by (47)

    View all citing articles on Scopus
    View full text