Behavioral and Appliance-specific Determinants of the Effectiveness of Residential
Electricity Demand Response Programs
L. Zhao, The Johns Hopkins University, Baltimore USA, +1-302-442-0564, [email protected]
B.F. Hobbs, The Johns Hopkins University, Baltimore USA, + 1-410-516-4681, [email protected]
Abstract
Electricity demand response (DR) programs use price incentives to induce a change in customer
electricity demand in response to system conditions, usually with several hours or a day’s notice.
Effective DR programs can reduce system costs and, in some cases, reduce the exercise of market power
by generation. In particular, dynamic pricing provides an incentive for customers to reduce demand
during peak load or high congestion hours; lessening the need for costly energy from peaking generators.
The analysis of this paper identifies the economic and behavioral aspects of customers’ response to DR
programs while considering their interaction with household characteristics.
Past studies measured customers’ response to price incentives through demand models that relate
hourly demand for electricity to various types of dynamic prices [2-10]. We hypothesize that there are
also more subtle reactions to DR programs. One example is changes in load during off peak hours in
addition to peak hours due to cross-price elasticities or spillover effects. Others are non-price behavioral
responses to information and other cues. Our analysis addresses both price and behavioural (non-price)
time-varying influences on demand.
We also address influences that do not vary with time. With a few exceptions, past demand
response studies have not analyzed how customer response depends on demographic characteristics and
appliance ownership patterns. The analysis of this paper addresses some of those dependencies as well.
1. Introduction
Electricity demand response (DR) programs use price incentives to induce a change in customer
electricity demand in response to system conditions, usually with several hours or a day’s notice.
Effective DR programs can reduce system costs and, in some cases, reduce the exercise of market power
by generation. In particular, dynamic pricing provides an incentive for customers to reduce demand
during peak load or high congestion hours; lessening the need for costly energy from peaking generators.
DR can also mitigate supplier market power by increasing the elasticity of demand [1]. A particular type
of DR program focussed on here is critical peak pricing programs (CPP), which significantly raises the
price of a electricity during a specified set of hours during an individual “critical peak” day that is
declared by the utility one day ahead of time.
Many detailed microeconomic studies of short-run electricity demand from the past have already
tried to estimate how variables such as appliance ownership, housing stock, prices, and demographic
characteristics affect electricity demand. However, there are improvements that can be made on these
studies. Firstly, pricing data used by some of these studies are averages from monthly bills [11-16] rather
than hourly prices that, for instance, may be distinct for peak and off-peak time periods. Also, some of the
studies scaled energy usage as kWh of usage per square foot rather than the absolute amount [17-23]. In
our study, we only scaled air conditioning since it is the only appliance affected by square footage.
Furthermore, most demand and demand response papers that have not examined the interaction of
behavioral factors and price response behavioral aspects of price response [24, 9, 11, 25, 8, 13, 23, 15,
16].
Research resulting from demand response pilots [1-3,6-7] are an improvement over the studies
mentioned above. With the exception of Faruqui and George [6], they do not pay attention to household
appliance characteristics. Even though this paper analyses the same dataset as Faruqui and
Herter[1,3,6,7], we approximated demand differently and uncovered new information on the factors that
influence electricity demand. Our approach resembles those in [27-29].
In contrast to [27] the income distribution of our sample size is not skewed to those earning above
$50,000. Furthermore, we do not use a step-wise algorithm to rely on software to select important
variables. Meanwhile, [28] made inferences through a much smaller dataset of survey responders. [29]
did not consider electric appliances other than air conditioning. Following the example of [30], who
investigated the effect of feedback, we have utilized electricity bills as proxies for feedback to test the
importance of that behavioral variable.
This paper pools a dataset that includes a group of customers who were subject to critical peak
prices (CPP), a group under time of use (TOU) pricing, a group under standard rates and a control group
received notifications on critical peak prices but who did not have price increases. One CPP group was
charged a critical price of 59 cents and an off-peak price of 9 cents per kilowatt-hour. Another CPP group
was charged 65 cents during critical peak hours and an off-peak price of 65. The TOU group was subject
to a peak price of 22 cents and an off-peak price of 10 cents per kilowatt-hour. Finally, the group under a
standard tariff and the control group were subject to an average standard rate of 13 cents per kilowatthour. Applying panel data econometric methods of the dataset, this paper aims to find the effect and
magnitude behavioral influences can have on electricity demand in a demand response program.
Specifically, we aim to find out whether information alone, in the form of electricity bill and of critical
day notifications, can have an influence on electricity demand. Other papers analyzing this dataset
[1,3,6,7] have looked into the effect price increases can have on electricity demand. Our study would
point out whether prices are necessary or sufficient triggers.1 Our results show that although prices
definitely have an impact on electricity use, household appliance ownership, information, and
environmental attitudes of the customers are also influential factors.
Section 2 introduces the model. Section 3 describes the data. Section 4 shows the results, and
Section 5 concludes.
1
We have more variability in prices than those other studies that used the same dataset because we pooled together
data for the different rate groups, including TOU, CPP and Standard rates.
2. Model
The model can be generally summarized as follows:
RAC=g(Ppeak, Poff-peak, AC, SQFT,CDD, Weekend, Z)
RAPPLIANCES=h(Ppeak, Poff-peak, PPHH, Weekend, Z)
Substituting in yields:
where:

Qi is average electricity demanded in kilowatt-hours/hour for a particular customer on a particular
day during a particular period (peak or off-peak)

R is the utilization rate of appliance (either air conditioning (AC) or other (APPLIANCES)

Z refers to behavioral triggers such as notification of a critical peak day being provided to
customers not in the CPP program (dummy “Information”) and the recent arrival of an electricity
bill, either of which might be expected to make electricity use more salient and perhaps trigger
decisions to reduce electricity use. As another example, we calculated a dummy variable “Bill
Increase” to reflect whether (1) or not (1) a bill that the customer received is greater than or less
than the average of 3 previous bills.

A is the capital stock of each appliance, assumed to be fixed in this study

Ppeak is the average price during peak hours

Poff-peak is the average price during off-peak hours

SQFT is the square footage of the residence

AC is a dummy for air conditioning ownership

CDD is cooling degree days in degrees Fahrenheit

PPHH is the number of people in a household

Weekend is a dummy for weekend days. Dummies can also be included for month or season.
In addition to the variables above, an independent variable called “green” has been included,
which we created out of a survey on the participants to measure their attitudes towards the environment.
Questions asked were whether the customer believes that climate change is a problem, whether everyone
should pay something to contribute to a better environment, or whether the customer thinks that
environment protection is hurting jobs.
Since the panel data includes both cross-sectional (across residential electricity customers) and
time-series data (daily demand by period over two years for each customer), we have used both “randomeffects” and “fixed effects” estimation routines to estimate the demand relationships. The former assumes
independent error terms, while the latter removes unobserved time–invariant errors among individual
households. While fixed effects estimation has the advantage of controlling for customer specific timeinvariant factors, it does not allow one to estimate the effect of explanatory variables that differentiate
among customers but do not vary over time. For instance, if pool ownership is suspected of being a
significant explanatory variable for electricity demand, because such ownership is unchanged over all
observations for a given household, the fixed effects model’s estimation of a separate intercept for each
household means that it is not possible to separately identify a pool ownership effect. For this reason, both
fixed and random effects results are reported.
3. Data
The dataset used in this study comes from the Statewide Pricing Pilot (SPP) conducted in
California in 2003 and 2004 involving around 2,500 customers [6]. This pilot tested the traditional timeof-use (TOU) rate and two varieties of critical peak pricing (CPP). One CPP tariff charges high peak price
for 2 P.M. to 7 P.M. for a small number of announced days per summer (19 days in our data set). CPP
prices could be as high as five times the normal level. During non-CPP event days, the tariff applied to
customers is the TOU rate. The other CPP price applies a high peak price to a variable length peak period
within the 2 P.M. to 7 P.M. time frame, ranging from two to five hours. The experiment also included a
group of customers that remained on a standard tariff. Finally, to test the usefulness of information in
inducing a demand change, a subset of the customers on the standard tariff received notifications for CPP
days but did not have CPP rates charged during CPP days.
4. Results
All our estimations include price, appliance, temperature, and building size. Price enters the
estimates in two ways: in a term involving the product of price, CDD, size of the house, and a dummy for
the presence of central air conditioning; and in a term involving price multiplied by the number of other
electric appliances and the number of people in the household. We present several sets of results
involving different behavioral variables. Table 1 displays the effect of receiving a bill that is an increase
from the average of the previous three bills on peak (2pm-7pm) electricity demand (the dummy “Bill
Increase”) for the day after the bill is received. Table 2 studies the effect information has on the peak
electricity demand of the control group while controlling for a person’s attitudes towards the
environment, by including dummies representing each combination of the dummy variable “green” and
the dummy variable “notification”. Table 3 studies the effect of information on peak demand, and Table 4
studies the effect of the same information but taking into account a person’s environmental attitudes.2
In Tables 1-4, “Information” equals 1 only for the control group of in the Californian experiment,
since they are the only ones not experiencing higher prices when a CPP is called, but yet are informed
about its occurrence. At first glance, it looks like information itself does not have an effect in reducing
load. However, as can be seen in Tables 2 and 4, once information is interacted with a person’s beliefs
about the environment (variable “green”), information can have an effect. Also, according to the results in
Tables1 and 2, receiving a bill can induce a reduction in a person’s peak load.
2
Because weekend consumption behavior differs from weekday consumption, a weekend dummy has been created
and used in Tables 3 and 4. This dummy is not found in Tables 1 and 2 since the two datasets used in the analyses of
those tables did not use data for weekends.
Table 1: Effect of Information and Bill Increase on Peak Electricity Demand
Table 2: The Effect of Information and a Bill Increase on Peak Electricity Demand Conditioned on
Customer’s Environmental Attitudes
Table 3: Effects of Information on Off-Peak Electricity Demand
Table 4: Effects of Information on Off-Peak Electricity Demand Conditioned on Customer’s
Environmental Attitudes (“Green”)3
3
Because weekend consumption behaviour differs from weekday consumption, a weekend dummy has been created
and used in Tables 3 and 4. This dummy is not found in Tables 1 and 2 since the two datasets used in the analyses of
those tables did not use data for weekends.
5. Conclusion
For demand response to be effective, it is important to understand all the factors that influence
demand. As this study shows, both appliance specific factors and behavior matter. Policymakers or
utilities that would like to conduct such program should know well the implications of the demographics
of its customers and of the stock of appliances its customer. One weakness in this study is that our
appliance data and answers from surveys are static, for a survey was only conducted at the end of the
pilots. However, as our results have shown, the conclusions are more complex than previously assumed
when behavioral cues were not taken into account.
Acknowledgements
This work was supported by NSF grants OISE 1243482 (WINDINSPIRE) and ECCS 1230788.
We acknowledge the irreplaceable assistance of Margaret Sheridan, who provided access to the dataset,
along with advice, and the suggestions of Jian Ni of the Carey School of Business, the Johns Hopkins
University.
Works Cited
[1] Herter, K., McAuliffe, P., & Rosenfeld, A. (2007). An exploratory analysis of California residential
customer response to critical peak pricing of electricity. Energy, 32(1), 25-34.
[2] Faruqui, A., & Sergici, S. (2011). Dynamic pricing of electricity in the mid-Atlantic region:
econometric results from the Baltimore gas and electric company experiment. Journal of regulatory
economics, 40(1), 82-109.
[3] Herter, K., & Wayland, S. (2010). Residential response to critical-peak pricing of electricity:
California evidence. Energy, 35(4), 1561-1567.
[4] Neenan, B., & Eom, J. (2008). Price elasticity of demand for electricity: A primer and
synthesis. Electric Power Research Institute.
[5] Bernard, J. T., Bolduc, D., & Yameogo, N. D. (2011). A pseudo-panel data model of household
electricity demand. Resource and Energy Economics,33(1), 315-325.
[6] Faruqui, A., & George, S. (2005). Quantifying customer response to dynamic pricing. The Electricity
Journal, 18(4), 53-63.
[7] Herter, K. (2007). Residential implementation of critical-peak pricing of electricity. Energy
Policy, 35(4), 2121-2130.
[8] Aigner, D. J., Sorooshian, C., & Kerwin, P. (1984). Conditional demand analysis for estimating
residential end-use load profiles. The Energy Journal, 5(3), 81-98.
[9] Carter, A., Craigwell, R., & Moore, W. (2012). Price reform and household demand for
electricity. Journal of Policy Modeling, 34(2), 242-252.
[10]McLoughlin, F., Duffy, A., & Conlon, M. (2012). Characterising domestic electricity consumption
patterns by dwelling and occupant socio-economic variables: An Irish case study. Energy and
Buildings, 48, 240-248.
[11]Bartusch, Cajsa, et al. "Exploring Variance in Residential Electricity Consumption: Household
Features and Building Properties." Applied Energy 92 (2012): 637-43.
[12]Aigner, Dennis J., Cynts Sorooshian, and Pamela Kerwin. "Conditional Demand Analysis for
Estimating Residential End-use Load Profiles." The Energy Journal 5.3 (1984): 81-98.
[13]Barnes, Roberta, Robert Gillingham, and Robert Hagemann. "The Short-Run Residential Demand for
Electricity." The review of economics and statistics (1981): 541-52.
[14]Carter, Adrian, Roland Craigwell, and Winston Moore. "Price Reform and Household Demand for
Electricity." Journal of Policy Modeling 34.2 (2012): 242-52.
[15]Parti, Michael, and Cynthia Parti. "The Total and Appliance-Specific Conditional Demand for
Electricity in the Household Sector." The Bell Journal of Economics (1980): 309-21.
[16]Wilder, Ronald P., and John F. Willenborg. "Residential Demand for Electricity: A Consumer Panel
Approach." Southern Economic Journal 42.2 (1975).
[17]Baltagi, Badi H., Georges Bresson, and Alain Pirotte. "Comparison of Forecast Performance for
Homogeneous, Heterogeneous and Shrinkage Estimators: Some Empirical Evidence from US
Electricity and Natural-Gas Consumption." Economics Letters 76.3 (2002): 375-82.
[18]Haas, Reinhard, and Lee Schipper. "Residential Energy Demand in OECD-Countries and the Role of
Irreversible Efficiency Improvements." Energy Economics 20.4 (1998): 421-42.
[19]Halvorsen, Robert. "Residential Demand for Electric Energy." The review of economics and statistics
(1975): 12-8.
[20]Hirst, Eric. "A Model of Residential Energy use." Simulation 30.3 (1978): 69-74.
[21]Houthakker, Hendrik S. "Residential Electricity Revisited." The Energy Journal 1.1 (1980).
[22]Kamerschen, David R., and David V. Porter. "The Demand for Residential, Industrial and Total
Electricity, 1973–1998." Energy Economics 26.1 (2004): 87-100.
[23]Dubin, Jeffrey A., and Daniel L. McFadden. "An Econometric Analysis of Residential Electric
Appliance Holdings and Consumption." Econometrica: Journal of the Econometric Society (1984):
345-62.
[24]McLoughlin, Fintan, Aidan Duffy, and Michael Conlon. "Characterising Domestic Electricity
Consumption Patterns by Dwelling and Occupant Socio-Economic Variables: An Irish Case Study."
Energy and Buildings 48 (2012): 240-8.
[25]Larsen, Bodil Merethe, and Runa Nesbakken. "Household Electricity End-use Consumption: Results
from Econometric and Engineering Models." Energy Economics 26.2 (2004): 179-200.
[26]Hobby, John D., and Gabriel H. Tucci. "Analysis of the Residential, Commercial and Industrial
Electricity Consumption." Innovative Smart Grid Technologies Asia (ISGT), 2011 IEEE PES.
[27]Kavousian, Amir, Ram Rajagopal, and Martin Fischer. "Determinants of Residential Electricity
Consumption: Using Smart Meter Data to Examine the Effect of Climate, Building Characteristics,
Appliance Stock, and Occupants' Behavior." Energy 55 (2013): 184-94.
[28]Gram-Hanssen, Kirsten. "Residential Heat Comfort Practices: Understanding Users." Building
Research & Information 38.2 (2010): 175-86.
[29]Steemers, Koen, and Geun Young Yun. "Household Energy Consumption: A Study of the Role of
Occupants." Building Research & Information 37.5-6 (2009): 625-37.
[30]Brandon, Gwendolyn, and Alan Lewis. "Reducing Household Energy Consumption: A Qualitative
and Quantitative Field Study." Journal of Environmental Psychology 19.1 (1999): 75-85.