Energy Policy 49 (2012) 552–561

Contents lists available at SciVerse ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

Consumer responses to time varying prices for electricity

Paul Thorsnes a, John Williams b, Rob Lawson b,n
a
Department of Economics, University of Otago, PO Box 56, Dunedin, New Zealand
b
Department of Marketing, University of Otago, PO Box 56, Dunedin, New Zealand

H I G H L I G H T S

c Seasonal effects in winter.

c High conservation effect from information.
c Higher peak prices no effect on peak use.
c Low off-peak prices encourage less conservation off-peak.

a r t i c l e i n f o a b s t r a c t

Article history: We report new experimental evidence of the household response to weekday differentials in peak and
Received 23 November 2011 off-peak electricity prices. The data come from Auckland, New Zealand, where peak residential
Accepted 29 June 2012 electricity consumption occurs in winter for heating. Peak/off-peak price differentials ranged over four
Available online 31 July 2012
randomly selected groups from 1.0 to 3.5. On average, there was no response except in winter.
Keywords: In winter, participant households reduced electricity consumption by at least 10%, took advantage of
Electricity pricing lower off-peak prices but did not respond to the peak price differentials. Response varied with house
Time of use and household size, time spent away from home, and whether water was heated with electricity.
Price elasticity & 2012 Elsevier Ltd. All rights reserved.

1. Introduction closely to production costs. At one extreme, the retailer could

As in most parts of the world, residential electricity prices in
New Zealand do not really reflect the variation over time in
marginal production cost. There are three types of schemes that
have a time based component. Some suppliers offer seasonal
pricing options because production costs are higher in winter due
to the widespread use of electricity for heating. Also, to take
advantage of lower-cost production at night, households have had
the option of installing a second electricity meter usually wired to
a special heater that stores heat at night, and releases it slowly
over the day. In return for a slightly lower price many households
also allow ‘ripple control’ which allows the electricity supplier
to switch their water heater off remotely (and without their
knowledge) during high-cost peak periods. However, even when
considered together all these schemes amount to small variations
on a ‘flat’ retail pricing structure.
The roll-out of new meters that record electricity consumption
at half hour intervals provides opportunities to tune prices more
n
Corresponding author. Tel.: þ64 3 479 1100.
E-mail addresses: paul.thorsnes@otago.ac.nz (P. Thorsnes),
john.williams@otago.ac.nz (J. Williams), rob.lawson@otago.ac.nz (R. Lawson).
vary prices half-hourly in response to variation in wholesale
prices and to stress on the distribution system. This ‘dynamic’
or ‘real time’ pricing challenges households to monitor prices and
adjust consumption accordingly. One option is a display mounted
in a prominent place in the home. Another option, given time and
investment, is enabling technology, such as a computer that
monitors prices and controls appliances. At the other extreme, if
patterns in electricity demand are fairly regular, the retailer could
vary prices daily across a relatively small set of time periods,
which is commonly referred to as time-of-use (TOU) pricing. The
simplest gives two prices for peak and off-peak periods.
This strategy is being actively discussed at both an industry
and a political level in many countries as a means of ‘managing
demand’. For example, Section 1252 of the 2005 United States
Energy Policy Act affirms that ‘‘y time-based pricing and other
forms of demand response, whereby consumers are provided
with electricity price signals and the ability to benefit by respond-
ing to them, shall be encouraged’’. This makes of interest the
extent to which households will respond to daily variation
in price. Studies conducted over the last forty years in various
places and under various conditions indicate that peak demand
is, on average, price inelastic. Nevertheless, even seemingly
small responses can have significant effects on marginal peak

0301-4215/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.enpol.2012.06.062
 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561
production costs, potentially reducing investment in expensive
peak generation or additional transmission or distribution
capacity.
In this paper we report the results of a time-of-use pricing
experiment conducted in Auckland, New Zealand. The study has
at least three novel aspects. First, this study is not concerned
about summer air conditioning. Electricity demand in New
Zealand peaks in winter for heating, rather than in summer for
cooling. Auckland, the largest urban area in New Zealand, is in the
relatively mild north (about the latitude of Virginia Beach,
Virginia and Seville, Spain). Mild summer temperatures and
breezes mean very little air conditioning in summer. Winters,
however, are cool, though not cold, and most heating is done
with electricity. Wood is the most common alternative energy
source.
With a comprehensive survey of participating households we
were also able to collect a more extensive range of personal and
household data than has been collected in previous TOU experi-
ments. This has enabled the examination of some new variables
to test their influence on peak and off-peak demand. A final
difference is that our study, with the cooperation of a major New
Zealand electricity retailer was deliberately aimed at the ‘meat of
the market’ as opposed to attempting to be nationally represen-
tative. The roll-out of smart meters by geographic area means
that a nationally representative sample is not possible anyway
but the selected area involved prime customers in relatively new
housing occupied by somewhat older and relatively high-income
households. This last factor is important if TOU policies are to be
implemented on a voluntary as opposed to mandatory basis. The
area under consideration in our study would be regarded as a
prime target in which to encourage the early adoption of TOU
pricing and the residents regarded as more capable of adjusting
lifestyles to suit compared with more deprived neighborhoods,
for example.
The study lasted for one year. Each participant household was
assigned randomly to one of four pricing groups. The peak to off-
peak price ratio ranged from 1.0, no price differential, to (approxi-
mately) 1.25, 1.75, and 3.5 (e.g., roughly 28b/kW h peak/ 8b/kW h
off peak). We had access to daily peak and off-peak consumption
data and half-hour readings from both the experimental period and
the previous year. We were thus able to conduct differences-in-
differences analyses.
Results indicate no response by participants relative to a
control group that was unaware of the experiment, except in
winter when electricity consumption and monthly payments both
increase considerably. During winter, the households in the
experimental groups conserved an average of about 12% relative
to households in the control group. This level of conservation is
surprising. Faruqui et al. (2009a) report that in-home displays of
energy consumption typically encourage conservation of about
7%. Thus 12% conservation in this case is relatively large, espe-
cially given that the information households received consisted
only of a set of energy saving tips at the start of the experiment
and a chart in their monthly bill showing daily peak and off-peak
consumption over the month.
The response to TOU pricing differed between peak and off-
peak periods. During peak periods, there was surprisingly no
variation, on average, across the experimental groups in the
amount of conservation in winter; e.g., those facing a peak price
of 28b/kW h conserved at roughly the same rate as those facing
18b/kW h. On average, households in all of the groups appear to
have been taking the steps they were willing to take to conserve,
regardless of price. In contrast, those who faced lower offer-peak
prices conserved less; households who faced 8b/kW h off-peak
prices took advantage of the low price by conserving less, on
average, than those who faced 18b/kW h.
553
There was, of course, variation in the response to the pricing
across the households in each group. All else constant, peak price
elasticity increased with floor area, household size and outdoor
temperature, and decreased with time spent away from home and
electric water heating. Not surprisingly, those who spent more
time away from home (usually during peak periods) were more
likely to take advantage of lower off-peak prices. This suggests
that the average responses would likely have been different if our
sample houses and households varied more in their character-
istics, in line with the variation in the country over all.
The remainder of the paper is organized as follows. The next
sections briefly summarize the results from existing studies of time-
varying prices. After this we describe the sample and experimental
setup and then report the results from three sets of analysis of
monthly, daily, and half-hour data, respectively. Finally we discuss
implications for electricity retailing and complementary govern-
ment policy.
2. Background literature
There is a small but established body of literature on the
responsiveness of households to changes in the price of electri-
city. In this literature there are studies that examine TOU pricing
(e.g., Faruqui and George 2002) and others that examine critical
period pricing (CPP) (e.g., Matsukawa, 2001; Herter et al., 2006;
Herter, 2007). With TOU pricing plans customers know when and
by how much price varies. With CPP, the customer agrees to a
pricing plan where the retail price rises, usually substantially,
when the wholesale price reaches a critical point. Aigner (1985)
gives a review of early work on TOU experiments. Conclusions
from the more recent literature on the price elasticity of elec-
tricity demand can be summarized as follows:
 Demand is price inelastic: typically, a 100% price change might
produce around a 20% change in demand (EPRI, 2008). Faruqui
and Sergici (2009b) report responses to peak period prices
under a standard TOU pricing system of 2% to 6%, translating to
own price elasticities of 0.02 to 0.1. This seems unsurprising:
electricity prices are typically low enough that electricity
payments make up only a small portion of the average house-
hold’s budget, and consumers likely perceive electricity as a
necessity during peak times. Price elasticities within TOU
schemes seem to increase over time as people are able to
adjust their lifestyle to suit the scheme. People become both
more experienced at managing their life around regularly
time-varying electricity prices and also are able to acquire
appliances that enable them to control the timing of electricity
usage (Filippini, 2011).
 Several TOU studies report overall conservation effects in
electricity usage (EPRI, 2008). This may reflect a ‘‘Hawthorne’’
effect induced by knowingly participating in an experiment.
As noted earlier, it may also result from receiving better
information about electricity consumption.
 There appears to be considerable variation in sensitivity to
electricity prices across segments of the market. Reiss and
White’s (2008) six year study of the Californian market
indicates that nearly half (44%) of households show no short-
run response to price fluctuations. Households who heat with
electricity tend to be more sensitive and higher-income house-
holds less sensitive to price. Bernstein and Griffin (2006)
report differences in price elasticities across the regions of
the United States which they speculated result from variation
in climate, available substitutes and demographic character-
istics Archibald et al. (1982) reports that price elasticities vary
seasonally.
554 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561
 Some studies report estimates of the elasticity of substitution
between higher-price peak and lower-price off-peak periods.
Caves et al. (1984, 1989) report average within day substitu-
tion elasticities of 0.12, but with considerable variation across
households. A later study by Midwest Electricity Systems
reports a somewhat larger range in substitution elasticities,
0.39 to zero, partially explained by ownership of major
appliances (Baladi and Herriges, 1998) while a more recent
study in California reports large differences between house-
holds with and without air-conditioning, a response first
identified by Lillard and Aigner (1984).
3. Experimental design
We partnered with Mercury Energy, one of the major New
Zealand electricity retailers, to undertake an experiment in TOU
pricing. Mercury was in the process of replacing conventional
residential electricity meters with meters that send readings
wirelessly every half hour.1 Approximately 4000 meters had been
installed at the time of the study, all in a suburban area to the
south of the Auckland CBD.
Invitations to participate in the study were sent to 1400 of these
4000 households. Recruiting ceased when 400 households had
agreed to participate. This was the maximum sample size that the
company was willing to support over the course of the project.
The process began with a face-to-face survey of the member of
each household who was responsible for managing the electricity
account. The survey was conducted by professional staff
employed by a market-research company. Data were collected
on the characteristics of:
 the house
 energy-using appliances
 household energy behaviors
 household composition and demographics.
 householder attitudes toward energy and the environment2
3.1. Sample characteristics
Table 1 provides information about household size. Most of the
single-person households were older females while the largest
single type of grouping is married (or partner couples) without
children. On average, participants in the sample are older (median
age is in the 50–54 bracket) while the mode lay in the 60–64
bracket. They are more affluent than the average New Zealand
household with the median before-tax household income for the
sample falling between $90,000 and $100,000 per annum before tax.
Consistent with the relatively high income, only 62 (18.7%) of
the households include someone who does not work at least part
time. About a third of households (35.5%) have at least two people
in full-time paid employment. Across all of the people in all of the
households, the average time spent away from home on a normal
weekday is six hours.
There are few Māori (22) or Pacific Island (12) respondents in
the sample. New Zealanders of European descent constitute by far
the largest proportion of the sample (72.1%). Respondents from
1
The signals from sent from each meter are received using antennas places
strategically around the neighbourhoods. If a signal is blocked, the meters store
information temporarily or relay the signal through meters on neighbouring
houses.
2
The environmental values scale used was the New Environmental Paradigm
Scale (NEP) developed by Dunlap et al. (2000).
Table 1
Household composition.
Number of people Number of households Percent of sample
1 42 12.7
2 116 34.9
3 60 18.1
4 69 20.8
5 or more 45 13.5
Total 332 100.0
233 of the 322 households have lived the majority of their lives in
the city and a further 21 households have moved to the study area
from other parts of New Zealand. Residents in the other 68
households have moved to the area from a wide variety of
countries with the main origins being China (19), India (10) and
the U.K. (9).
Comparing these figures to the New Zealand census data, the
sample is older, wealthier and with fewer Māori and Pacific Islanders
than the national average.
4. Experimental setup
Each of the participant households received two types of
information. Prior to the start of the TOU pricing period, each
household received a one-page list of tips for conserving energy.
In addition, the monthly power bill included a simple bar chart
where the height of the bar showed total daily power consump-
tion and colors indicated the proportion of the total consumed
during peak and off-peak periods.
The price paid per kilo-watt hour prior to the experiment
varied across participant households. Mercury customers choose
from an array of pricing plans. Some plans are tailored for low
electricity users. Some plans trade off a higher fixed daily charge
for a lower unit charge. A household can decrease their unit
charge by agreeing to allow the power to their water heater to be
switched off remotely, and without their knowledge, during
periods of peak demand. Prices varied from essentially 17b to
21b, with the average and the mode being just over 19b/kW h.
Roughly three-quarters of the households paid either 19.9b or
17.4b/kW h.
Each of the 400 participant households was assigned randomly
by Mercury staff to one of four groups:
1. Information-only group: these households were interviewed
and received the initial information sheet and monthly con-
sumption chart, but they did not face time-varying prices, so
the peak/off-peak price ratio equals 1.
2. Low price differential group: a peak/off-peak price differential
of 4b. That is, peak and off-peak prices equal the household’s
current price plus or minus 2b, respectively.
3. Medium price differential group: a peak/off-peak price differ-
ential of 10b.
4. Large price differential group: a peak/off-peak price differen-
tial of 20b.
The experiment started on 1st August 2008 (winter time in the
southern hemisphere) and finished at the end of July 2009. The
peak period was defined as 7 am to 7 pm Monday to Friday. All
other times and public holidays were off peak.
Of the 400 initial participants, 332 remained in the study through
the duration of the trial. Mercury supplied daily peak and off-peak
electricity consumption and half-hour consumption for each house-
hold over both the year of the experiment and the previous year.
Mercury also supplied similar consumption information for 55
 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561 555

Table 2
Determinants of average daily electricity consumption.

Variable Average daily consumption Proportion off-peak

Before During Before During

p b p b p b p b

Month 0.000 – 0.000 – 0.000 – 0.000 –

Floor area 0.013 0.001 0.002 0.001
Electric water heating 0.002 0.404 0.000 0.611 0.002 0.033
Number of people 0.031 0.066 0.012 0.086
Income (in bands) 0.002 0.678 0.000 0.798
Special health needs 0.025 0.437 0.015 0.535
Hours away from home 0.012  0.028 0.004  0.036 0.000 0.006 0.000 0.006
Environment values 0.011  0.112 0.036  0.103
Pricing plan 0.000 – 0.000 –

‘control’ households from the same suburb who were unaware of
the experiment, were not interviewed, and did not receive any
additional information about their consumption.
Unfortunately, not all of this data was usable. Some of the half-
hour and daily data from August through November 2007 had
been ‘backfilled’ due to technical problems and were unreliable.
Our year-on-year comparisons are consequently restricted to the
eight months December through July.
5. Year-on-year analysis of monthly averages of daily
consumption
Our overall analysis strategy was to account for as much
variation as possible in both (a) overall electricity consumption;
and (b) the proportion of consumption at peak times before we
assessed the impact of TOU pricing. This approach helps to
alleviate problems due to the allocation into experimental groups
being confounded with pre-existing group differences in electri-
city consumption. The more common approach to analysis of
experimental data, ANOVA, cannot detect this possibility. We
start by analyzing observations on the daily household consump-
tion of the participants in the experiment averaged over each
month. Median daily consumption was used instead of total
consumption to account for missing data. Because we have panel
data (multiple observations from the same experimental units)
we estimated a linear mixed effects (LME) model with dummy
variables to capture the effects of each household in order to
accommodate unobserved individual heterogeneity.3
Using notation following Baltagi (2005), the model is:
yit ¼ ai þ bi xit þui þ Eit i ¼ 1,. . .,n; t ¼ 1,. . .,T; E½u ¼ 0; E½E ¼ 0
where y is value of the dependent variable in household i at time
t, ai is the mean level of y for household i when all explanatory
variables are zero, x is a vector of explanatory variables and bi
their associated coefficients for household i. Finally, ui is known as
the individual error, while Eit is known as the idiosyncratic error.
The idiosyncratic errors are assumed to be uncorrelated with both
the individual errors and the explanatory variables. The LME
approach is required because of the presence of dummy variables
that represent each household. If the individual errors are
correlated with the bs, the OLS estimator is inconsistent. Hence
3
Specifications of the model include: random coefficients (intercepts and
slopes within household), time specified as a cubic polynomial, correlation of
residuals modeled by AR (1) and heteroscedasticity of residuals modeled by a
power function. AR(1) provided the best fit of all alternatives for modeling the
residuals, as judged by likelihood ratio tests. Inspection of ACF and PACF plots
supported using AR(1) as opposed to higher order serial correlation.
the fixed-effects model (sometimes known as least-squares with
dummy variables) cannot estimate the effects of time-invariant
explanatory variables (Baltagi, 2005, p. 13).
We included a large number of explanatory variables in our
model, each of which had either been found in previous research
to be influential on either total use or time of use, or could
reasonably be supposed to be so. These variables included
characteristics of the household (number of people, income
etc.), the dwelling (age, construction, insulation methods) and
the individual who responded to the household survey (age, sex,
personal values regarding environmental issues etc.).
Table 2 lists the variables from the household survey that best
explain the variation in the monthly average of daily electricity
consumption. There are 20 monthly observations for each of the
332 participant households. Households in the control group are
not included because they did not complete the household
survey. There are two dependent variables: the square root of
daily consumption averaged over the month and the square root
of the proportion of daily consumption consumed during the off-
peak period4. A complicating factor emerged in that mean and
median electricity consumption differ significantly across experi-
mental groups prior to the experiment. However, the distribu-
tions in consumption over-lap, and we expect much of the overall
variation to be explained by variation in house and household
characteristics and season. Because of these pre-existing differ-
ences in consumption between the groups noted above, the
models were estimated separately for data from before and
during the experimental period.
The first seven independent variables listed in Table 2 are fairly
obvious, but a few require explanation. The monthly dummies are,
not surprisingly, highly significant due to the seasonality of elec-
tricity consumption. As expected, average daily consumption
increases with the size of the house and whether water is heated
using electricity. Household characteristics also matter: consump-
tion increases with household size and income and when someone
in the house has special health needs that require electrical
appliance to treat. Not surprisingly, average consumption decreases
when householders spend more weekday time away from home.
Of interest is that the estimated coefficients on several of these
variables change during the experimental period. In particular,
larger estimated coefficients suggest that those with higher
incomes, more people in the house, and those heating water with
electricity responded relatively little to participation in the
experiment.
4
The square root is used as a variance-stabilising transform and was selected
by the Box-Cox procedure. After transformation, inspection of residual plots
confirmed that assumptions of OLS modelling were satisfied.
556 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561

The two remaining independent variables are a bit unusual. Table 3

The first indicates that, all else constant, households who profess Summary of results of modelling total and proportional usage.
relatively strong environmental values on average consume less
Dependent Period Comparison GoF GoF Group
electricity. It seems that in this case values translate to behavior (B) (W) effect?
in a significant way. Of interest is that the magnitude of the
coefficient falls during the experiment; those with environmental Average daily consumption Before Control 0.59 0.91 0.781
values consume less, but not quite as much less relative to the Info 0.67 0.90 0.315
During Control 0.52 0.82 0.853
others in the experiment. The conservation effects we report later Info 0.60 0.82 0.396
suggest that it is the others that are changing their behavior to be Proportion of consumption Before Control 0.06 0.78 0.062
more like those that express strong environmental values. off-peak Info 0.24 0.79 0.127
Finally, the pricing plan chosen by the household (i.e., the During Control 0.05 0.64 0.045
Info 0.21 0.64 0.187
tradeoff between the fixed daily charge and the charge per kW h
consumed) correlates strongly with consumption. This seems
unsurprising because we would expect that households choose
their plan based in part on their consumption patterns. That this 6. Seasonal variation in peak and off-peak consumption
comes out in the statistics, however, indicates that households are
indeed sensitive to the differences in the pricing plans. The apparent lack of response to TOU pricing reported in the
The right-hand side of the table indicates that surprisingly few of previous section is surprising, but the aggregation may hide a
the variables collected in the survey explain the cross-sectional seasonal response. In addition, the results reported above may be
variation in the proportion of the daily consumption consumed off- influenced by unobserved house or household characteristics. We
peak. Not surprisingly, those who spend less time at home tend to can treat the problem of unobserved variables by comparing daily
consume a higher proportion their energy off-peak. Of potential consumption for each household from the year before the
interest is that off-peak water heating is significant only during the experiment with consumption during the experiment. That is,
experiment. Some households may have been able to restrict water we conduct a differences-in-differences analysis: we estimate the
heating to take advantage of lower off-peak prices. average differences across each experimental and control group in
Table 3 provides goodness of fit measures for the models the year-to-year differences in each household’s electricity con-
above in addition to other models of interest, and tests for sumption during peak and off-peak periods. Specifically we
differences across in groups. Two goodness-of-fit (GoF) measures estimate the following equation using OLS:
are distinguished for each model. The first, GoF (B), measures the
proportion of cross-sectional variation explained by the model, X
4 X
4
lnðQ i Þ ¼ a þ bg dGg þ j dTOU þ rg dTOU  dGg þ g temp þ e
i.e., the differences across households averaged over time. The g¼1 g¼1
average daily consumption models fit fairly well for a cross- ð1Þ
sectional analysis and are as good as other similar estimates, for
example the study of domestic Californian consumption reported where Qi is the quantity of electricity consumed each day during
by NETL (2009). Note that the fit for the control sample is the peak or off-peak period by household i. (Taking the natural log
relatively low as we lack explanatory variables taken from the of consumption generates estimates of the coefficients that are
household survey. Not surprisingly, the fit of the models explain- interpretable as percentage changes.) a is a constant term that is
ing cross-sectional variation in the proportion of consumption the average daily peak or off-peak consumption of the households
off-peak is smaller. in the control group. dGg is a dummy variable that equals 1 if the
The second goodness-of-fit measure, GoF (W), measures the household is in one of the experimental groups. For example, if
proportion of time series variation explained by the model, i.e., g¼1, then the household is in the information-only group. The
the variation within each household from month to month. In this coefficients bg measure the difference in the average daily con-
case the goodness of fit measures are relatively high because most sumption of the households in each experimental group from that
of the time series variation is seasonal and is easily captured by of the control group. dTOU is a ‘dummy’ variable that equals 1 if
the time variable. the day is in 2009, i.e., during the TOU pricing period. The
The right-hand column labeled ‘‘Group Effect’’ lists the estimated coefficient j measures the average percentage differ-
p-values for the differential effect of being in one of the groups ence in control-group consumption in 2009, under TOU pricing,
exposed to TOU pricing (i.e., in one of the low, medium, or high relative to that in the same period in 2008. We expect j to be
price differential groups). The row labeled ‘Control’ compares all positive in winter because the winter of 2009 was colder than the
the experimental groups against the control group and the row winter of 2008. dTOU  dGg interacts dTOU with a dummy that
labeled ‘info’ compares the groups with a pricing treatment equals 1 if the household is in each of the four sample groups.
against the information only group. The effects should not be Thus r1, r2, r3, and r4 measure the average percent changes from
significant before the experimental period, and this is achieved 2008 in consumption by the households in each experimental
with the possible exception of the comparison to the control group relative to that of the households in the control group.
group for the proportion of off-on peak usage (p¼.062). Surpris- These are the estimates of interest. temp is the maximum for peak
ingly, we would expect to see more significant differences or minimum for off-peak daily temperature, so the estimate of g
between groups during the experiment, and that is not the measures the overall average percentage effect on consumption of
generally the case, though the significant of the difference a one degree change in temperature. e is the ‘error term’. It
between the control group and the experimental groups becomes represents all of the variation in household consumption not
slightly more significant during the experimental period. It should explained by the variation in temperature, the group, or the year.
be borne in mind that the models explaining the variation in To detect seasonal effects we run a series of regressions on six
control group consumption contain fewer explanatory variables weeks of weekday observations of each household’s consumption.
(mainly just month and pricing plan). The first regression uses data from late summer when vacation
The interpretation is that there is no significant annual impact season has ended. Each subsequent regression starts one day
of TOU pricing on either average electricity consumed or on the later, with the last of the 170 regressions centered on a day in
average proportion of daily consumption consumed off-peak. mid-winter (early July). We then plot each of the four coefficients
 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561 557

15.00%

10.00%

5.00%

0.00%

21-Jan
28-Jan
4-Feb
11-Feb
18-Feb
25-Feb

6-May
13-May
20-May
27-May
3-Jun
10-Jun
17-Jun
24-Jun
1-Jul
8-Jul
4-Mar
11-Mar
18-Mar
25-Mar
1-Apr
8-Apr
15-Apr
22-Apr
29-Apr
-5.00%

-10.00%

-15.00%

-20.00%

-25.00%

-30.00%
Info-only Low Medium High
-35.00%

Fig. 1. Moving six-week average % difference from control in weekday peak consumption.

on the dTOU  dGg interaction variable for every regression. Thus,
each of the plots shown consists of a series of 170 six-week
moving-average estimates of the percent difference in each
experimental group’s year-to-year consumption relative to that
of the control group.5
Figs. 1 and 2 show the results of most interest. Fig. 1 shows
how the weekday peak-period consumption of each of the
experimental groups differs from that of the control group from
summer to winter. The left side of the graph, which covers
February into May, indicates little systematic response to the
TOU pricing experiment in late summer and autumn.6 The
standard errors vary slightly around 3%, so differences between
group averages are largely insignificant.7
The abrupt drop in all of the curves in May corresponds to the
onset of winter in the southern hemisphere and shows a sig-
nificant conservation response to participation in the experiment.
Higher electric bills apparently encouraged significant conserva-
tion in winter. There is surprisingly no statistical difference across
groups in the extent of conservation. Indeed, the group facing the
highest peak prices appears to have conserved the least, though
the difference from the information-only group is not statistically
significant.
Fig. 2 similarly shows relative changes in year-on-year weekday
off-peak consumption relative to control. On average, each of the
experimental groups again conserves relative to control. Standard
errors again vary around 3%, so off-peak conservation in summer and
autumn is border-line significant. Again, conservation grows in
percentage terms in winter; larger winter-time bills apparently
encourage more effort at conservation, averaging about 15% relative
to Control. In contrast to peak-period consumption, and more in line
with expectations, the groups that experience lower off-peak prices
respond by conserving less than those in the Information-only group.
The year-on-year responses of individual households in
each group not surprisingly vary from the averages shown in
the figures above. We would expect this variation to depend
systematically on house and household characteristics, at least in
winter.
5
The estimated coefficients and standard errors for the first six weeks
regression and the last weeks regression are reported in the appendix.
6
The previous summer months also showed no difference to control and have
not been plotted for the sake of clarity in the figure.
7
Showing 95% confidence intervals clutters the diagrams excessively.
There are several house characteristics that might influence
the percentage winter-time response in consumption to a change
in price. In this relatively short-run setting, we expect the
percentage response to higher prices to be smaller in the presence
of house characteristics that encourage higher overall consump-
tion. For example, we would expect a lower percentage response
to higher prices in larger houses and in houses with electric water
heating. Another key characteristic of New Zealand houses is that
very few were built with any thermal insulation before insulation
requirements were introduced in 1978. Ceiling insulation has
been retro-fitted in most of these houses, but they remain
relatively poorly insulated. Electricity consumption in these
houses tends to be higher in winter, so we expect a smaller
percentage response to higher winter prices. Another unusual
characteristic of New Zealand houses is that central heating is
very rare, 2.5% in our sample. Indeed, most households heat with
relatively small resistance heaters. More recently, more efficient
heat pumps have been become popular, with about 11% of our
sample heating with a heat pump. We would expect the greater
efficiency of these heat pumps to discourage a strong response to
higher peak prices.
There also are several common household characteristics that
seem likely to affect response to higher peak prices. We expect
higher income households to respond less to higher prices
because most of the household services that require electricity
to produce are probably income normal. A larger number of
people in the household implies a lower per-capita income, and
should perhaps have the opposite effect. Households who spend
more time away from home on weekdays seem also likely to
respond less to higher peak prices, in part because they may have
more opportunity to adjust their work schedule to avoid peak
times at home, and in part because they consume relatively little
during peak (note that household size and hours away from home
are positively correlated with income).
We expect that many of these house and household character-
istics are likely to lead to the opposite effects off-peak, i.e., to a larger
response to lower prices. For example, higher income households
seem less likely to conserve in response to high peak prices and
more likely to take advantage of low off-peak prices. And we expect
a lower percentage response to price the lower the outdoor
temperature (in part because overall consumption is higher).
We estimate the following equation to obtain the average
effects of one unit changes in house and household characteristics
558 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561

5.00%

0.00%

13-May
20-May
27-May
11-Feb
18-Feb
25-Feb
21-Jan
28-Jan

11-Mar
18-Mar
25-Mar

10-Jun
17-Jun
24-Jun
15-Apr
22-Apr
29-Apr
6-May
4-Feb

4-Mar

3-Jun
1-Apr
8-Apr

1-Jul
8-Jul
-5.00%

-10.00%

-15.00%

-20.00%

-25.00%

Info-only Low Medium High

-30.00%

Fig. 2. Moving six-week average difference from control in weekday off-peak consumption.

on peak and off-peak price elasticities:

Table 4
lnQ i ¼ a þ blnPi þ j lnPi  X i þ g dHi þ e ð2Þ Effects on elasticities of each one unit change in house and house characteristics.

in this case, Qi is weekday peak (or off-peak) consumption by Price elasticity Substitution
household i who faces price, Pi. Xi is a vector of house and elasticity
household characteristics. The term dHi is a dummy variable for
Peak Off peak
each household, which produced ‘fixed-effects’ estimates. The
estimates of j on the price/household characteristic interaction Ln Price (peak, off-peak, or ratio)  0.371  0.0902  0.453
terms are reported in Table 4. (0.074) (0.077) (0.048)
The average effects of the same house and household char- Ln Price  floor area (10 m2) 0.0801 0.0127 0.0114
(0.014) (0.0034) (0.0019)
acteristics on the substitution elasticity are estimated using the
Ln Price  built pre-1980 0.586  0.381  0.00954
following (similar) equation: (0.093) (0.066) (0.038)
Ln Price  heat pump 0.0289  0.156  0.0832
Qp Pp Pp
ln ¼ a þ bln þ jln  X þ g dHi þ e ð3Þ (0.0288) (0.047) (0.026)
Q op Pop P op Ln Price  electric hot water 0.261 0.0386 0.146
(0.047) (0.036) (0.020)
Again, the estimates of j are reported in the table. The average Ln Price  household income 0.0311  0.0183 0.0115
price and substitution elasticities reported in the context of ($10,000s) (0.0064) (0.0046) (0.0026)
Figs. 1 and 2 were estimated using similar equations, but with Ln price  hours away from home  0.0345 0.0222  0.00553
the interaction terms omitted. (0.0057) (0.0041) (0.0023)
Ln price  household size  0.0297  0.0462  0.0448
The data consist of observations on consumption of the house- (0.016) (0.013) (0.0073)
holds in the experimental groups on non-holiday week days in Ln price  temperature  0.00327  0.00709 0.0117
June and July (winter) in both 2008 and 2009. The control group (0.00042) (0.00038) (0.0024)
households are excluded because they did not complete the N 17,775 18,201 17,063
household survey, and we therefore lack observations on house
and household characteristics. Note that the estimates are similar
if we restrict the sample to observations in June and July of 2009,
during the TOU experiment. The substitution elasticities in the right-hand column of
Standard errors appear in parentheses, bold indicates signifi- Table 4 are broadly consistent with the price elasticities.
cance at the 5% level. Numbers of observations vary due to
occasional errors in recording of data.
The left-hand column of numbers shows estimates of the 7. Response to the peak/off-peak boundary
influence on peak price elasticity of house and household char-
acteristics. All of the coefficients have signs and magnitudes In this section we take advantage of the half-hour readings to
consistent with expectations, though those on heat pump and look more closely at changes in the pattern of consumption across
household size are insignificant. The middle column shows the 7 pm peak/off-peak time boundary. Half-hour residential
similar estimates for off-peak. As expected, many of the signs consumption peaks at about 6:30 pm. TOU pricing should, in
are opposite those for peak; the households least responsive to theory, encourage a drop in consumption prior to and a rise in
higher prices tend to be the most responsive to lower prices. This consumption after 7 pm.
seems consistent in that these households are relatively keen to To investigate, we look at electricity consumption at half-hour
enjoy the household services produced with electricity. intervals from 4:00 pm to 10:30 pm on twenty consecutive
As expected, the household response to price falls as tempera- weekdays in July 2008 (prior to TOU pricing) and the correspond-
ture rises and overall expenditure on electricity decreases. ing weekdays in July 2009 (during TOU pricing). Our estimating
 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561
8%
6%
4%
2%
0%
4:00
4:30
5:00
5:30
6:00
6:30
7:00
7:30
8:00
8:30
9:00
9:30
10:00
10:30
-2%
-4%
-6%
-8%
Fig. 3. Average year-on-year % change in daily evening-peak consumption.
equation measures the average percentage difference in con-
sumption during TOU pricing in 2009 from that in 2008 in each
half hour of the evening peak. We expect these differentials to be
negative prior to the 7:00 pm price change (households use less
during the high-priced peak period), and positive after (house-
holds take advantage of lower prices off peak).
The results are shown in Fig. 3. The figure indicates precisely
the expected pattern of load shifting in response to TOU pricing:
consumption was lower in 2009 relative to 2008 prior to and
higher after 7:00 o’clock (standard errors are about 3%). Relevant
is that that all three groups with a price differential respond
similarly. The magnitude of the price differential did not affect the
size of the shift. Again, the absolute magnitude of the price
response is modest, but meaningful; consumption increased by
an average of about 6% relative to the 6:30 peak.
We also conducted a similar analysis of half-hour data around
the morning off-peak/on-peak price boundary. Here the pattern of
changes around the morning boundary is less tidy, but there is a
significant 7% drop in consumption between 7:00 and 7:30, again
indicating a significant and meaningful response to higher peak
prices.
8. Conclusions and implications
To summarize, Mercury Energy recruited 400 households to
participate in an experiment in time-of-use pricing. The sample
households opted-in to the experiment. The demographic char-
acteristics of the sample households are not nationally represen-
tative: they are on average older, wealthier, and live in better
insulated houses. Each sample household received a list of
energy-conservation tips and a chart in their monthly bill show-
ing daily peak and off-peak consumption. Peak to off-peak price
ratios varied from 1.0 (no differential) to about 3.5, with the
average of the peak and off-peak prices equal to the non-TOU
price (so average peak price ranged from 18 to 28b/kW h and off-
peak price ranged from 8 to 18b/kW h). The control group
consists of households chosen at random from the same neigh-
borhood who had no knowledge of the experiment. The experi-
ment ran for one year, beginning in mid-winter.
With respect to effect of the experimental conditions on
electricity consumption, analysis of the data indicates:
 Variation in average daily electricity consumption is explained
by house and household characteristics, as expected, though
we largely fail to explain variation in the proportion of
electricity consumed off-peak. Of interest is that households
who profess environmental values on average consume less
electricity, other characteristics held constant.
559
 TOU pricing appears to have no statistically significant effect
on average daily electricity consumption or, surprisingly, on
on-peak consumption when averaged over the entire year.
 Significant peak-period conservation in winter across all experi-
mental groups relative to the control group. Winter peak con-
servation averages in the range of 10–15%. This is consistent with
existing reports of seasonality in response rates.
 No significant variation in peak conservation with price across
experimental groups; peak price elasticity of demand is very
low.
 Similar levels of off-peak conservation, but off-peak conserva-
tion does vary with price; households on average took advan-
tage of lower off-peak prices.
 Variation in response around the peak/off-peak boundary,
with a relatively large average increase in consumption off-
peak.
 Response to price varies with the floor area of the house,
household income, household size, the amount of time the
household spends away from home and whether the house-
hold heats water with electricity.
Of interest is the extent of winter-time conservation. Overall
demand peaks in winter, so a drop of 10–15% could significantly
decrease stress on the supply system. This conservation could
result from the rather modest and low-cost information provided
to participant households, or it might be the result of a so called
‘Hawthorne’ effect, a response that occurs simply as a reaction to
participating in the experiment. While we are in the usual
situation of not being able to give a definitive answer to this
dilemma we do know that recent reevaluations of Hawthorne
effects show that they while they seem unpredictable they are
usually quite small (Merrett, 2007). Also, the effects we detect are
occurring nearly a year after the onset of the experiment, a lag
long enough to expect Hawthorne effects to have dissipated.
From a policy perspective, the information conventionally pro-
vided to households with regard to electricity prices, costs and
consumption seems remarkably limited. The typical monthly bill
provides essentially no information about daily consumption pat-
terns. All households know is that they consume more in winter (as
evidenced by the larger bill). The additional information provided to
these participant households costs little to supply (given half-hour
meter readings): a simple bar chart showing peak and off-peak
consumption each day of the month. Yet this modest amount of
extremely low-cost information appears to have encouraged sig-
nificant conservation during the months of most stress on the
supply system. This seems likely to represent ‘low-hanging fruit’;
households who have not given much thought to electricity con-
sumption that does not demand a large part of their household
budget apparently give it some thought.
A surprise is that substantial percentage variation in peak
prices (from 18b to 28b/kW h) has no effect, on average, across
participant groups. This appears to suggest that TOU pricing is
ineffective as a demand management instrument. To understand
consider the situation of a typical sample household in winter. To
keep the arithmetic simple, but still relevant, assume that prior to
the experiment an individual household might have paid $200 for
energy (excluding the fixed daily charge) per winter month at
20b/kW h and consumed 600 kW h during peak ($120 total) and
400 off-peak ($80). Next assume that participation in the experi-
ment combined with better information encouraged peak con-
servation of 10% (the low-hanging fruit). If peak price during the
experiment is 30b/kW h, then the peak bill rises to $162/month.
If the household takes advantage of the 10b/kW h off-peak price
by not conserving at all (or, more likely, by conserving on low-
value consumption – the low-hanging fruit – and enjoying
additional high-value consumption), then the off-peak bill falls
560 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561
to $40/month, for a total of $202/month, essentially no change in
the monthly bill.
While our results indicate very modest responses to our TOU
pricing experiment we do not believe that it is the end of the
story. Our experiment was relatively short and those who were
inclined to economize did what they could regardless of the price
differential. Existing evidence indicates that price elasticities
grow over time. If the household had committed to this pricing
plan, over time one would expect to see alterations in both
behavior – such as changes in work schedules – and investments
in technology – such as more energy-efficient appliances
equipped with timers – that help the household avoid relatively
high peak prices. As noted earlier, we cannot and would not want
to try to generalize our findings to the country as a whole. As a
demand management technique, time-of-use pricing could be
introduced as a market led product that could be developed to
appeal to a segment of the population, or it could be introduced
on a compulsory basis. The political complexion of New Zealand
currently suggests that the former is more realistic than the latter
and would avoid issues around the social consequences of forcing
change upon consumers who may lack the ability to respond to
time-varying prices. In that context generalisabilty to the total
population is not an issue and our sample is likely to be
representative of the segment that is the prime target for time-
of-use pricing options by retailers.
Of interest is the large variance across the sample in when
households consume electricity. We had a singular lack of success
in explaining this compared to overall use. When averaged our
sample did yield the recognized pattern of diurnal peaks in the
morning and evening with the expected seasonal change from
summer to winter. However, examining individual households
reveals hardly a single dwelling that conforms closely to the
aggregated pattern. There is seemingly little uniformity in the
way people live, even in a fairly modern and homogeneous
Auckland suburb. This is a feature that reinforces the point made
in the previous paragraph that as a policy instrument TOU pricing
could be politically difficult to implement in a uniform way across
a whole population, and certainly research is needed to better
understand the influences of electricity usage by time of day.
Appendix 1. Sample of results from estimation of Eq. (1)
Table A1 shows the results of the last of the 170 regressions
used to build Figs. 1 and 2. The left-hand column of numbers
consists of the results from estimating Eq. (1) using observations
on non-holiday weekday peak consumption in the six weeks to
the end of July in 2008 (before the experiment) and 2009 (during
the experiment). The right-hand column reports the same thing
using observations on off-peak consumption.
The estimates for the first four variables in the left-hand column
represent the average percent difference in each experimental group
average consumption from control in 2008. Average consumption
levels clearly varied across the groups: households in the info-only
and low-price differential groups consumed more, on average, and
households in the low and medium-price differential groups con-
sumed less, on average, than households in the control group. These
differences may suggest non-random allocation of households to
groups, or it may simply reflect an unusual outcome of a random
process. Retailer staff allocated households to groups, reportedly
using a random process (Excel random number generator). As a
practical matter, the distributions in daily consumption and in house
and householder characteristics across groups largely over-lap. The
differences-in-differences approach treats issues with variation in
characteristics across groups.
Table A1
Sample six-week average % difference from control in weekday consumption.
Peak Off-peak
Info-only 0.201 (0.0226) 0.111 (0.0232)
Low 0.518 (0.0225) 0.418 (0.0231)
Medium  0.122 (0.0221)  0.0616 (0.0227)
m  0.135 (0.0222)  0.230 (0.0228)
dTOU 0.131 (0.0251) 0.120 (0.0257)
dTOU  info-only  0.140 (0.0316)  0.139 (0.0323)
dTOU  medium  0.0951  (0.0308)  0.111 (0.0316)
dTOU  low  0.124 (0.0314)  0.0888 (0.0322)
dTOU  high  0.105 (0.0310)  0.0677 (0.0318)
Temp at 7 pm/am  0.00938 (0.00258)  0.0222 (0.00263)
Constant 2.474 (0.0355) 2.604 (0.0364)
N 16,962 16,989
The estimates on the variable dTOU represent the average
percent difference in consumption by households in the control
group in 2009 relative to 2008. Consumption rose significantly in
2009 due to lower average winter temperatures.
The coefficients on the four interaction terms are of most
interest. They represent the average percent difference from
control in the average year-on-year difference consumption. All
of the coefficients are negative and significant, indicating con-
servation during the experiment relative to control. The extent of
conservation during the peak period does not vary significantly
across the three groups; the groups facing higher peak prices do
not conserve more than those facing lower peak prices. The extent
of conservation does vary off-peak; lower prices encourage less
conservation, as might be expected.
The coefficients on temperature are negative, as expected:
consumption decreases as temperature increases.
The number of observations differs slightly due to omissions of
observations flagged by the retailer as unreliable (due to technical
problems).
The dependent variable is the natural log of weekday peak
or off-peak electricity consumption. Standard errors appear in
parentheses. Observations come from the six weeks prior
to the end of July (i.e., correspond to the last data point in
Figs. 1 and 2).
References
Aigner, D., 1985. The Residential Electricity Time-of-Use Pricing Experiments:
What Have We Learned? in Social Experimentation, Jerry A. Hausman and
David A. Wise, Eds. The University of Chicago Press and National Bureau of
Economic Research, 11–54. Retrieved from /http://www.nber.org/chapters/
c8372S.
Archibald, R.B., Finifter, D.H., Moody, C.E., 1982. Seasonal variation in residential
electricity time-of-use pricing experiments: evidence from survey data.
Applied Economics 14 (2), 167–181.
Baladi, S.M., Herriges, J., 1998. Residential response to voluntary time-of-use
electricity rates. Resource and Energy Economics 20 (3), 225–244.
Baltagi, B.H., 2005. Econometric Analysis of Panel Data, 3rd Ed John Wiley & Sons,
Chichester.
Bernstein, M., Griffin, J., 2006. Regional Differences in the Price-Elasticity of
Demand for Energy. National Renewable Energy Laboratory. Retrieved from
/http://www.nrel.gov/docs/fy06osti/39512.pdfS.
Caves, D., Christensen, L., 1984. Consistency of residential consumer response in
time-of-use pricing experiments. Journal of Econometrics 26 (1-2), 179–203.
Caves, D., Heririges, J., Kuester, K., 1989. Load shifting under voluntary residential
time-of-use rates. The Energy Journal 10 (4), 83–99.
Dunlap, R.E., Van Liere, K.D., Mertig, A.G., Jones, R.E., 2000. Measuring endorse-
ment of the new environmental paradigm: a revised NEP scale. Journal of
Social Issues 456 (3) 425-422.
EPRI (2008) Price Elasticity of Demand for Electricity: A Primer and Synthesis,
Electric Electricity Research Institute, January. Retrieved from /http://my.epri.
com/portal/server.pt?Abstract_id=000000000001016264S.
Faruqui, A., George, S.S., 2002. The value of dynamic pricing in mass markets. The
Electricity Journal 15 (6), 45–55, July.
 P. Thorsnes et al. / Energy Policy 49 (2012) 552–561
Faruqui, A., Hledik, R., Tsoukalis, J., 2009a. The power of dynamic pricing. The
Electricity Journal 22 (3), 42–56, April.
Faruqui, A., Sergici, S. 2009b. Hosehold Response to Dynamic Pricing of
Electricity—A Survey of Experimental Evidence. The Brattle Group. Retrieved
from /http://www.science.smith.edu/  jcardell/Readings/uGrid/House%20De
mandResp%20Experience.pdfS.
Filippini, M., 2011. Short- and long-run time-of-use price elasticities in Swiss
residential electricity demand. Energy Policy 39 (10), 5811–5817.
Herter, K., McAuliffe, P., Rosenfeld, A., 2006. An exploratory analysis of California
residential customer response to critical peak pricing of electricity. Energy 32
(1), 25–34.
Herter, K., 2007. Residential implementation of critical-peak pricing of electricity.
Energy Policy 35 (4), 2121–2130.
561
Lillard, L.A., Aigner, D.J., 1984. Time-of-day electricity consumption response to
temperature and the ownership of air conditioning appliances. Journal of
Business and Economic Statistics 2 (1), 40–53.
Matsukawa, I., 2001. Household response to optional peak-load pricing of
electricity. Journal of Regulatory Economics 20 (3), 249–261.
Merrett, F., 2007. Reflections on the Hawthorne effect. Educational Psychology 26
(1), 143–146.
NETL (2009) An Investigation into California’s Residential Electricity Consumption,
D DOE/NETL, 2009/1371. Downloaded from /http://www.netl.doe.gov/ener
gy-analyses/pubs/CA_US_elec_consumption_FINAL_July-29-2009.pdfS on 18th
September 2011.
Reiss, P.C., White, M.W., 2008. What changes energy consumption? prices and
public pressure. The RAND Journal of Economics 38 (3), 636–663.