Journal of Economic Psychology 61 (2017) 1–14

Contents lists available at ScienceDirect

Journal of Economic Psychology

journal homepage: www.elsevier.com/locate/joep

Nudging electricity consumption using TOU pricing and

feedback: evidence from Irish households q
Valeria Di Cosmo a,⇑, Denis O’Hora b
a
Economic and Social Research Institute and Trinity College, Dublin, Ireland
b
School of Psychology, National University of Ireland Galway, University Road, Galway, Ireland

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

Article history: This paper analyses the electricity usage of 5000 Irish residential consumers in response to
Received 25 July 2015 the introduction of Time Of Use (TOU) tariffs and three different forms of financial feed-
Received in revised form 9 March 2017 back: immediate feedback from in-home displays (IHD), monthly billing and bimonthly
Accepted 9 March 2017
billing. Half-hourly data on consumption collected during the trial indicated that TOU
Available online 16 March 2017
tariffs reduced consumption at peak, with some reductions lasting beyond the end of the
peak period and post-peak spikes in usage were not observed. IHD feedback resulted in
Keywords:
the most reliable reductions and bimonthly billing the least. Moreover, consumers slightly
Demand management
Smart meter trials
increase the electricity usage during the first hours of the night and early in the morning.
Information Households with greater education react to the information associated to the TOU tariffs
TOU tariffs slightly more than the average.
Household behaviour Ó 2017 Elsevier B.V. All rights reserved.

1. Introduction

Balancing electricity demand and supply is difficult and expensive. Traditionally system operators have used supply-side
techniques, such as reserves, to make sure that there is always enough generation to meet a fairly inelastic demand as
demand fluctuates.1
Recently there has greater interest in the potential of demand-side tools to balance energy markets as highlighted by
Feuerriegel and Neumann (2016). In particular, there are considerable advantages in increasing the sensitivity of electricity
demand to its generation costs. First of all, higher electricity generation costs are largely driven by higher demand.2 Princi-
pally, peak demand requires increased electricity generation at times when it is most costly, so the cost savings of reducing peak
demand are greater than proportional to the decrease in demand. A second advantage concerns environmental impact of peak
demand. Power plants that are brought online to supply electricity at peak times tend to run on fossil fuels and be less efficient,
causing greater greenhouse gas emissions per unit of electricity generated. A third advantage concerns energy efficiency.

q
Valeria Di Cosmo acknowledges funding from Science Foundation Ireland, Grant No. 09/SRC/E1780. The opinions, findings and conclusions or
recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Science Foundation Ireland. The authors
thank Niamh Devitt for outstanding research assistance, two anonymous referees, participants of the Conference of the Division of Behaviour Analysis in
Galway, seminars at the ESRI and UCC, and the 8th EMEE Conference for helpful comments. The authors are responsible for any remaining omissions or
errors.
⇑ Corresponding author.
E-mail address: valeria.dicosmo@esri.ie (V.D. Cosmo).
1
Short-term demand is inelastic in part because retail prices have historically been fixed for long periods of time. See for Ireland (Eirgrid, 2009).
2
On average, peak load for the Irish system is about 25% greater than the average day demand.

http://dx.doi.org/10.1016/j.joep.2017.03.005
0167-4870/Ó 2017 Elsevier B.V. All rights reserved.
2 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

Fig. 1. Plot of estimated coefficients at half hourly intervals before, during and after the peak pricing period, by feedback stimulus and tariff for houses in
the Higher Ed category only. See Figure \ref{fig:whole_results} for an explanation of symbols.

Reducing peak demand increases energy efficiency by ensuring more complete usage of current energy resources and
reducing required investments in electricity generation, transmission and distribution in the medium to long term, leading
to savings for the system as a whole. All EU countries must reduce emissions and energy consumption by 2020 to comply
with EU energy efficiency targets (2009/28/EC).3 Reducing peak electricity demand also helps meet this target. Reduction of
peak demand need not require consumers to reduce the use of electricity; they can also redistribute the usage of some appli-
ances (dishwashers, washing machines) among different times of the day and night. Of these alternative times, redirecting elec-
tricity demand from peak times to the night hours presents several benefits, both for consumers and for the system.4
Consumption during the night hours is cheaper than during the day and the peak hours, as shown by Fig. 1.
As shown in Fig. 1, in the hours highlighted in blue, immediately before and after the peak time (highlighted in green), the
wholesale price of electricity is lower than in the peak hours. The basic idea behind the installation of smart meters is that
rational consumers facing real-time prices may postpone or anticipate the use of electricity to benefit from lower electricity
prices. Night-time consumption also takes advantage of increased availability of renewable generation like wind generation,
that comes on stream during the night as highlighted by Finn, OĆonnell, and Fitzpatrick (2013) and Eirgrid (2014). Finally,
increasing the demand elasticity encourages greater competition between generators, as shown by Walsh, Malaguzzi Valeri,
and Di Cosmo (2016).
The current study investigated whether pricing schemes and pricing feedback provided to the consumer change the con-
sumption during peak hours and lead to a shift in consumption from peak hours to other times of day. In examining shifts in
consumption, we were particularly concerned to establish whether short-term redistribution, such as earlier usage just prior
to the peak pricing period or postponement of usage until the end of the peak pricing period would give rise to a new peaks
in the hours immediately before or after the peak pricing period. Moreover, we analysed whether households with different
educational level react to the change in tariffs and information differently from the whole sample.
Previous work have discussed how consumers react to a change of the electricity price mechanism when flat tariffs are
switched to Time of Use (TOU) or continuous pricing schemes. For instance, a number of experiments in the US (see Faruqui
& Sergici, 2010) and in Europe (see Aubin, Fougare, Husson, & Ivaldi, 1995; Filippini, 1995) have investigated whether con-
sumption reacts more to prices when the cost of electricity follows wholesale costs more closely. The experiments in the US
(Faruqui, Harris, & Hledik, 2010) confirm that higher prices during congested hours lead to a slight change in the electricity
consumption, but the magnitude of the effect is not constant across experiments. Moreover, the authors find that customers
may take advantage of lower night time tariffs by shifting their electricity usage to night-time storage heaters and water
heaters on a timer. Faruqui and Sergici (2010) analysed 15 pricing experiments using TOU pricing and found that customers
responded to higher prices during the peak hours by reducing peak hour electricity usage and/or shifting it to less expensive
off-peak periods. On average, TOU rates induced a reduction in peak demand from 3–6% in the 15 pricing experiments. In
some cases, consumers do not significantly change consumption as prices change. There are a number of reasons why this
might occur. Consumers have bounded capacity to reduce or reallocate electricity usage and it is possible that they do not
fully reassess their optimal consumption when the price of electricity increases. Alternatively, the opportunity cost of chang-
ing behaviour may be too high for the consumers (i.e. the time spent monitoring prices constantly is not used for other pro-
ductive activities).
Gabaix and Laibson (2006) found that consumers often express uncertainty on the relationship between inputs and out-
puts, such as how heating a room translates into energy usage or how a use of the washing machine translates into cost.
Other authors such as Allcott (2011), Jessoe and Rapson (2014) and Di Cosmo, Lyons, and Nolan (2014) have investigated

3
Directive 2009/28/EC of the European Parliament and of the Council (2009) on the promotion of the use of energy from renewable sources and amending
and subsequently repealing directives 2001/77/EC and 2003/30/EC, Official Journal of the European Union.
4
See, as examples, Commission (2015) and Joskow (2012).
 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14 3

the link between the electricity prices and electricity consumption founding that the changes in prices are not sufficient to
reduce electricity use significantly.
The failure of increases in electricity price to effectively change consumer’s behaviour may also stem from how consumers
choose to use electricity. Loewenstein and Thaler (1989) suggest that the decision of how much electricity to use is a classic
intertemporal choice: electricity and its benefits are immediate, but the cost of electricity use is delayed until consumers
receive their bill. A considerable literature has demonstrated that the impact of the expected outcomes of a decision reduces
as a nonlinear function of the delay until those outcomes are expected to occur (Frederick, Loewenstein, & O’Donoghue,
2002; Odum, 2011). The discounting effect is strongest for short delays, with immediate outcomes being considerably more
impactful than outcomes delayed to any extent. For consumers, decisions on when and how much electricity to consume are
difficult, because they value the future benefits of reduced consumption much less than the immediate benefits of electricity
usage. Furthermore, the degree to which individuals discount future rewards and losses is sensitive to the magnitude of
those outcomes (Johnson & Bickel, 2002; Kirby & Maraković, 1996), such that smaller outcomes are discounted more
strongly than larger outcomes. In electricity consumption choices, the future rewards of each decision on its own are quite
small, which encourages steeper discounting of those future rewards and more impulsive behaviour. Consumers may fail to
see the benefit in choosing the future reward and thus, their electricity usage patterns may be resistant to change.
In 2007, the Irish Commission for Energy Regulation (CER) announced a trial smart metering experiment in the Irish
residential electricity market in an effort to investigate the benefits of Time-of-Use (TOU) pricing of residential electricity
and estimate the role that clear and accessible price charged might play in facilitating consumer behaviour change. This
Consumer Behavioural Trial (CBT) employed TOU tariffs in combination with other demand side stimuli (bi-monthly and
monthly billing and in-home displays which show constantly the electricity consumption installed at home) in an effort
to reduce peak demand and overall electricity use in the residential sector. This residential CBT is one of the largest smart
metering behavioural trials conducted internationally to date and thus provides a wealth of information on the impact of
smart metering on electricity consumers (CER/11/080a).
The Irish regulator who conducted the trial summarised the main findings of their statistical analysis in CER (2011). Their
analysis shows the contraction of peak consumption after the trial for all the consumers, with nonlinear patterns related to
the type of stimuli adopted. Di Cosmo et al. (2014) examined data from the Irish CBT smart meter trial to assess the broad
effects of different electricity pricing structures and methods of pricing information provision on household consumption
during three times of the day, peak (from 5PM to 7PM), night (from 11PM to 8AM) and day (all the other times of the
day). The authors found that the reduction of consumption during peak hours was not compensated by the rise of consump-
tion during the night. They did not find any significant difference between the behaviour of households with a tertiary degree
education and the rest of the population. The current study aims to supplement these previous analyses by investigating
consumer behaviour in greater temporal detail. In particular, in light of findings reported by Gabaix and Laibson (2006),
we assessed whether the provision of higher frequency information on electricity use and price helped to resolve consumer
uncertainty and improve conservation and find mixed evidence of these effects in the behaviour of the participants in the
trial.
In the current paper, we investigated the effects of electricity pricing on the half-hourly dynamics of peak consumption.
Data was analysed at 30-min intervals in order to track changes in usage across the day. We were particularly interested in
tracking changes in usage before, during and after the peak interval targeted by TOU tariffs. The trial conducted by the CER
gives us the opportunity to study the response of the consumers to a change in tariffs and stimuli provided in a small market,
characterised by high levels of wind penetration. This makes our study unique, as future studies on demand-response could
be calibrated to the real half-hourly response provided by the Irish consumers.
A central concern of the current study was to examine household’s response to TOU tariffs in order to ascertain whether
households were reducing electricity usage overall or shifting electricity usage from the peak interval. For instance, if house-
holds were to delay making dinner until the end of the peak interval (7PM), then a new peak in electricity usage would be
generated at the end of the targeted peak interval. The rise of new peaks in response to the change in tariffs is a common
assumption in several works (see Faruqui et al., 2010; McKenna & Keane, 2015 among others) and it is important to deter-
mine the magnitude of this shift in order to calibrate future models on real consumer data. This effect could not have been
observed by Di Cosmo et al. (2014) because, in that work, the authors considered only the mean values of day and night, it
was not possible to disentangle usage effects in the hours close to the peak hours from the overall level of usage during the
non-peak hours of the day. This is also different from the effects estimated by similar literature (such as Ito, Ida, & Tanaka,
2015; Jessoe & Rapson, 2014) as these works focus specifically on the reduction of consumption during peak hours.
Second, we investigate whether the consumers are particularly responsive to a certain tariff or stimuli. The half-hourly
specification allows us to test whether different tariffs and stimuli have different effect on the consumption during the crit-
ical peak hours.
Third, examining the data at 30-min intervals allows us to establish which hours would be affected by a shift of consump-
tion from peak hours. Due to the low electricity price during the night, consumers and producers would potentially benefit
from a shift in demand to night-time usage, but planning for such shifts requires further detail on exactly when during the
night consumers replace consumption.
Similarly to Di Cosmo et al. (2014) and Jessoe and Rapson (2014), we found that consumers contracted consumption dur-
ing peak hours. This effect is The main cause of lower peak was a reduction in usage during the peak that was not shifted
from peak hours to other hours. When we examined changes in the electricity usage just prior to and following the peak
4 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

hours, an interesting effect was observed; consumers with in-home display (IHD) installed at home, who had immediate
price feedback, maintained reduced usage past the end of the targeted peak interval, when prices had reduced.5 It is possible
that the change in immediate price feedback was not large enough for households to be sensitive to the change in information
and that households misinterpreted the price signal. It is also possible that since the IHD was a highly salient object in the home
(much more visible than monthly or bimonthly bills), it might have induced a spillover effect of electricity reduction behaviours
beyond the end of the interval. As far as we know, no other studies have identified this effect. Furthermore, this result is quite
interesting because this effect highlights that more information is not always better for the consumer.6
Second, our results did not find significant differences in usage between the different tariffs proposed during the trial.
That is, we found that the households reacted to the increase in the price during peak hours (and the reduction during night
hours) but the magnitude of the price increase did not predict the degree to which usage was reduced.
Third, our findings show a small increase of consumption during the night. Di Cosmo et al. (2014) found a very weak effect
during the night hours, however these authors were considering night hours all together (from 11PM to 8AM). Our analysis
shows that the households increase the electricity consumption during the trial in the first hours of the night (before 1AM)
and in the first hours of the morning (between 530 and 7AM). This suggests that the households may be willing to increase
the night consumption and reduce peak consumption but they should be able to manage that change directly. Future tech-
nological developments may change this result dramatically and allow to a stronger matching between demand and wind
generation.
Finally, this work investigates how the educational level may affect the half-hourly level results. The delay discounting
effect seen in intertemporal choice is weaker for individuals who are better educated and higher earners (Becker &
Chiswick, 1996; Green, Myerson, Lichtman, Rosen, & Fry, 1996; Reimers, Maylor, Stewart, & Chater, 2009). One might expect
interventions designed to highlight future consequences may have a greater impact in households with greater access to
higher education. If, in line with such research, lower socio-economic (SES) groups find it more difficult to change usage
in response to TOU pricing, they would end up paying a larger share of electricity costs and effectively subsidising the con-
sumption of higher SES groups. Since individuals with lower income and less education also have fewer financial resources,
these higher costs constitute a large relative economic burden and might increase fuel poverty as shown by Scott, Lyons,
Keane, McCarthy, and Tol (2008).
Our work finds that households of people with a tertiary degree react to the information provided during the trial slightly
more than the average, reducing their consumption more during high tariffs in the peak hours. Our results differ from Di
Cosmo et al. (2014) as they did not find different behaviour between household with a tertiary degree and the full sample.
However, looking at half-hourly differences in behaviour show that some significant changes in the behaviour of highly edu-
cated people emerge. As a result the policy maker should carefully evaluate the distributional impact of policies finalised to
make the consumers reduce their peak consumption or shift the consumption from the evening to the night.
This paper continues by describing our data used from the consumer behavioural trial in Section 2. Section 3 describes the
methodology used, Section 4 presents the results of our analysis and Section 5 concludes.

2. Data

The CBT replaced existing mechanical meter readers with smart meters in approximately 5,000 households. The residen-
tial component of the trial involved all these households (customers of Electric Ireland) who were asked to participate in the
trial.7 While participating households self-selected into the trial, which limits generalisation to the overall population, partic-
ipants were randomly assigned to control and treatment groups. Data were collected over the period 14 July 2009 to 31 Decem-
ber 2010, and as the experiment began on 1 January 2010, six months of pre-trial data are available for both the control and
treatment groups. The control group was billed on their normal tariff (bimonthly tariff) and saw no changes to their bill. They
received none of the information stimuli and were requested to continue using their electricity as normal (Commission for
Energy Regulation, 2011a). Benchmark pre-trial data is available for all households (both control and treatment) for the period
14 July 2009–31 December 2009. Treatment groups were exposed to a variety of TOU tariffs and information stimuli (IHD,
bi-monthly and monthly billing). IHD provided the consumers with real-time information on their consumption and the
electricity price, with colours to highlight whether the tariff was high (peak), medium (day) or low (night), similarly to
what described in Jessoe and Rapson (2014).8 More precisely, four TOU tariffs were tested. TOU prices referred to peak
(17:00–18:59 Monday-Friday, excluding public holidays), day (08:00–16:59; 19:00–22:59 Monday-Friday, plus 17:00–18:59
public holidays, Saturday and Sunday) and night (23:00–07:59) periods (based on system demand peaks). A weekend tariff
was also tested (whereby the night rate applied all day Saturday and Sunday, with separate peak, day and night tariffs for week-
days). In comparison with the initial flat-rate tariff, the electricity price associated with peak hour consumption was increased
by a maximum of 166 per cent of its initial value, while the price of electricity during the day and night was decreased by a
maximum of 13 per cent and 37 per cent respectively. The TOU tariffs were designed to be neutral in comparison with the

5
We found that both monthly billing and IHD have a significant effect on consumption reduction during peak hours. Jessoe and Rapson (2014) found a
greater reduction but also the price variation that these authors have in their sample is up to 6 times higher than ours.
6
See also Lunn and Somerville (2015) for a discussion on how too much information may lead consumers to undertake sub-optimal decisions.
7
At the time of recruitment (mid-2008), Electric Ireland customers represented 100 per cent of residential electricity customers in Ireland (CER, 2011).
8
A description of IHD can be found here: http://www.cer.ie/docs/000117/cer13164-presentation-of-energy-usage-information.pdf, pg. 101.
 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14 5

standard flat-rate tariff to ensure that the average participant who did not change their electricity consumption would not be
financially penalised Table 1.
The changes in the peak, day and night prices happen simultaneously under the same tariff. So consumers associated with
tariff A had an increase in the peak price up to 22.7 euro cents per KWh and a decrease in the night price down to 13.62 euro
cents per KWh. Participants to the trial were randomly allocated between treatment and control groups. Moreover, partic-
ipants in the treatment group were randomly allocated between the different type of treatments.9 The randomization process
was carefully made by the statisticians working for the Irish regulator authority, as stated at pg.54 of the report written by CER
(2011). As a result, the statistical tests that can be made in order to test the randomness of the sample are related to the balance
of the groups in the sample. The randomization process itself is made not to achieve balance between groups, but to remove all
the biases that may characterise different groups. As a result, the balance between different groups does not prove that the ran-
domization process has been made correctly. At the same time, lack of balance between groups does not imply that the random-
ization process failed. Tables A1 in the Appendix shows the mean comparisons between tariff and stimuli groups. The
Table shows that different groups are not statistically different from each other. However, the careful explanation provided
by CER (2011) about the randomization process is the best guarantee that the randomization of the sample has been done
correctly.10

3. Methodology

Statistics calculated by CER (2011) using the data from the trial, show that smart metering in combination with
bimonthly billing and energy usage statement affect electricity consumption with a decrease in peak period consumption
of 11.3%. From these initial findings it is not clear that there was new peak formation in energy use as a result of the pre-
dicted price change. In general TOU tariffs and two of the Demand-Side Management stimuli (monthly billing and IHD
installed) were found to reduce overall electricity usage by 2.5% and peak usage by 8.8%. Their statistical analysis, however,
did not assess the causal relation between the change in tariff and the electricity consumption. Moreover, it did not take into
account the other variables (like weather and appliances installed in the house) that potentially affect the consumer’s beha-
viour. We divided the sample into three different subsamples, following the stimuli given to the consumers (bimonthly tariff,
monthly tariff and in home display). We then use half-hourly data to investigate the determinants of the behaviour of the
consumers both during the peak and the boundary hours, close to the change in the tariff. In order to estimate the changes in
electricity consumption before and after the trial we use the difference in difference estimator. For each group, we compare
the mean of consumers in the pre-treatment and treatment period and the mean of consumers in the control and treatment
group, controlling for everything else that has the potential to affect the behaviour of consumers during time. This can be
done by using the random effects estimator for panel data. It is important to highlight that in our sample there are no tariff
differences for consumers in the control group. As the CBT trial was set up as a randomized experiment, people in the control
group statistically have the same characteristics of the consumers in the other group, but they were not subject to a change
in tariff. Moreover, people in the control group were not informed about the change between the ”control” and the ”treat-
ment” period. As people in the control group were facing the same price during the whole period (control plus trial) for them
the diff-in-diff effect of a change in tariff is zero. We include them in our analysis in order to capture symmetric and exoge-
nous shocks that may have happened but that are not under control and are not captured by the regressors included in Eq.
(1) below. All models are estimated using STATA 13, and standard errors are adjusted for clustering at the household level.
qits ¼ a1 Tar A;t þ a2 Tar B;t þ a3 Tar C;t þ a4 TarD;t þ a5 TarW;t þ b1 TarA  Groupi þ b2 Tar B  Groupi þ b3 TarC  Groupi
þ b4 Tar D  Groupi þ b5 TarW  Groupi þ TreatPeriodt þ Controls ð1Þ
in which qits is the half hourly consumption of electricity by household i under stimulus s observed at the half-hour t. We also
include here consumers who were in the control group during the treatment period, to take into account consumers under
stimulus s but not subject to tariff’s change. Tar z;t is the dummy variable indicating that the household was exposed to tariff z
(z ¼ A; B; C; D) during the treatment period. Tar W;t is the dummy variable indicating that the household was exposed to the
weekend tariff during the treatment period. The coefficients therefore represent our difference-in-difference estimates (i.e.,
the effect of the various TOU tariffs on household electricity consumption). Tar z  Groupi is the time-invariant dummy vari-
able indicating that the household is a member of the Tariff z treatment group. These variables should not be significant at
the statistical level, as by construction the consumers were distributed randomly across the groups. Treatp is the dummy
variable indicating the treatment period. This variable indicates the differences between the control and the treatment per-
iod in terms of electricity consumed. We expect that during the trial the consumers reduce their electricity consumption as
they become more aware of their usage, so we expect a negative and significant sign for the Treatp coefficient.
Controls is an additional variable, which includes a dummy variable for public holidays and a set of dummies for each day
of the week. We also include a variable that reflects the heating degree days, sunshine is a variable that reflects sunshine
hours (not included in the night specification), a count variable of the number of appliances owned by the household and
a dummy variable indicating that the household has an electric heating system. We also include in our specification a vari-

9
For a complete description of the trial see CER (2011).
10
We thank one anonymous referee for this clarification.
6 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

able that interacts the heating degree days with the dummy for electric heating; this variable should control for high elec-
tricity consumption during the winter of 2010, in which the temperatures in Ireland were exceptionally low, as well as the
differential response to TOU tariffs among households with different heating types.
Finally, on the 20th of July 2010 and on the 6th of December 2010 due to technical problems smart meters stop register-
ing the electricity consumption of a subsample of 80 households. In order to assess the impact of this technical problem we
run different robustness checks: first we run the regressions interpolating consumption between the 19th and the 21st of
July and the 5th and 7th of December respectively; second we drop those days from our sample and, finally, we drop those
days only from the subsample of affected consumers. The results in all of the cases didn’t vary significantly from each
other.11

4. Results

Analysing the consumer behaviour on half-hourly basis will allow us to answer different questions.
First, are the consumers willing to reduce their consumption during peak hours? And if so, is there any particular hour in
which this reduction is more effective? Moreover, is there a particular stimulus or tariff particularly effective in making the
consumers achieving significant changes in consumption?
Second, is it possible to observe a significant shift of consumption patterns to the first hours of the night, time in which
the wind generation is higher? Or are the consumers creating a new peak just after 7PM? The following sections deal with
these questions and offer some answers.

4.1. Peak

In this Section we will focus on the behaviour of the consumers during the peak hours (from 5PM to 7PM), which are the
hours in which consumption is the highest, accordingly with Fig. 1. We will investigate whether the consumers reduce their
demand of electricity and if they are interested in shifting their consumption before and after those hours, generating a new
peak around 8PM. Table 2 and Fig. 2 below show the results of the consumer’s reaction to the change in tariffs and stimuli
during the trial. As highlighted by Di Cosmo et al. (2014) during the peak hours (from 5PM to 7PM), consumers across all
three stimuli (bi-monthly bill, monthly bill and IHD) significantly modify their behaviour during the treatment period,
reducing their consumption with respect to the control period. However, Di Cosmo et al. (2014) do not investigate the effects
of a tariff change on half-hourly basis. This is particularly important for two reasons. First, to check whether the consumers
react constantly during the hours affected by the change in tariff. It’s important to assess the consumer’s reaction on half-
hourly basis in order to establish whether the price signal is perceived correctly in all the four half-hours affected by the tariff
change, or if changes in the behaviour occur after the first half-hour. Second, our analysis allows us to assess whether during
the boundary half-hours (i.e. the half-hours which immediately precede or follow the change in tariff) there are changes in
electricity consumption. This is important because loads may be shifted easily from the peak hours to the period immedi-
ately after, generating a new demand peak which potentially undermines the results achieved by the tariff increase during
the peak hours Wickham (2009).
The following Figure shows the estimated coefficients by feedback stimulus and tariff. The full results are reported in
Table 2 in the next page. In the Figure below the coefficients which are statistically significant are represented by full dots.
The coefficients which are statistically significant measure the difference in the consumer’s electricity consumption between
the control and the treatment period. For example, the first column of Fig. 2 shows the estimated coefficient for consumers
associated to tariff A, which is the tariff with lowest variation from the flat tariff the consumers faced in the control period.
The first line represents consumers with bi-monthly bill. The second line, consumers with monthly bill and the third line,
consumers with IHD. All the coefficients estimated for this group of consumers are statistically different from zero, but peo-
ple with IHD installed in their home were more likely to change their consumption after the introduction of the TOU tariffs
than consumers facing other stimuli.12
Looking at the half hourly data allows us to draw some interesting conclusions. Fig. 2 above shows that the more respon-
sive households also keep their consumption lower than during the control period at 730PM, which is after the end of the
peak tariff. For example, usage remind significantly lower than baseline for consumers with IHD and tariff A until 8PM. For all
households, though usage increased at the end of the period it remained close to or below baseline levels, meaning that there
was no post-peak rebound effect due to short-term postponement of usage and no new peak in consumption. The continued
contraction of consumption at 7.30 and 8PM may have different explanations. Consumers may have reduced usage that was
unnecessary and this effect would have persisted past the end of the peak if such unnecessary usage spanned the hours from
6 to 8 pm. Alternatively, consumers may have adjusted slowly to the change back to the day tariff applied after 7PM. The
time signal might not have been clear across households and if higher usage was renewed sporadically across households
within the post-peak period then overall usage would remain low. Interestingly, the one group whose usage remained sig-
nificantly lower was in receipt of immediate price feedback through the IHD. Nevertheless, even if households were aware of

11
Complete results are available from the authors upon request.
12
Also this result is coherent with the findings by Di Cosmo et al. (2014).
 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14 7

4pm to 8pm (Whole Sample)

sig
A B C D

0.05 FALSE

TRUE
Coefficient

0.00
feedback
−0.05
Bimonthly

−0.10 Monthly

IHD
16:30
17:00
17:30
18:00
18:30
19:00
19:30

16:30
17:00
17:30
18:00
18:30
19:00
19:30

16:30
17:00
17:30
18:00
18:30
19:00
19:30

16:30
17:00
17:30
18:00
18:30
19:00
19:30
Time

Fig. 2. Electricity price in Ireland, average by hour (2009–2010), euro/MWh.

the timing of change to the lower tariff, behaviours developed to reduce usage during the peak period may have persisted
past the end of the period simply through behavioural resistance to change. In addition, many household tasks induce peri-
ods of electricity usage so, if a reduced usage strategy was employed for these tasks, the subsequent reductions may last
beyond the end of the peak interval.
Consumers, on the whole, did not engage in short-term postponement of electricity usage. In the next section, we con-
sider the evidence for a consumption shift from peak to night hours.
Table 2 provides the tariff coefficients obtained in the models of usage for each 30 min interval. In these models, the coef-
ficient of the treatment period dummy was always significant and negative, as expected, indicating lower usage during the
treatment period than during the control period. The effects of the group dummy, Tar z  Groupi in Eq. (1), were consistently
non-significant, indicating that the tariff-groups were not statistically different from each other, which was also expected.13
Comparisons across different feedback stimuli were not possible, but Wald tests were conducted to compare the magnitude of
the coefficients associated to the prices and few significant differences were observed.14 Environmental control variables had
consistent effects across models. The coefficients of the sunshine dummy was significant and negative, meaning that electricity
usage decrease during sunshine hours, which is quite reasonable considering the electricity usage for lighting and dummies for
electric heating, appliances and bank holidays were also significant and positive.
The foregoing results indicate that, on the most part, consumers responded in an all or nothing fashion to tariffs. In other
words, the smallest increases in the peak tariff from the flat tariff were sufficient to induce consumers to reduce their elec-
tricity use during peak hours. Evidence of the impact of larger tariff increases on consumer usage was observed only between
6 and 7PM and only when on a monthly bill or with an IHD installed in their houses. However, these results may significantly
change once new technologies that effectively promote the electrification of the system are introduced in the market, and
bigger savings may be achieved by postponing the use of some appliances during the night.
Analyzing the patterns by the stimuli provided, in the IHD group the relationship between the costs of the TOU tariff and
observed reductions was not as strong as for the people in the monthly-bill group. In particular, comparing people facing the
highest tariff (tariff D) in these two groups lead to some interesting conclusions. As shown also by Table 2, in all the half-
hours people with the highest tariff and monthly-bill reduce their consumption more than people facing the same tariff
but with IHD installed. There are two straightforward reasons why this might have occurred. First, IHD feedback (given in
euro KW/hour) might have reduced behaviour of consumers on the most punitive tariff to such a degree that it reached a
floor of electricity reduction, a local minimum due to the limited elasticity of electricity demand. Such a possibility seems
the most parsimonious explanation of the observed consumption patterns and is supported by the fact that consumption
of consumers on Tariff A who received IHD had reductions similar to the lowest reductions observed in the study. A second
possibility concerns the information provided by the IHD when compared to that provided by monthly billing. For a small
enough unit of time (e.g., 1 min), there is very little difference between the cost of electricity across tariffs, so households
might not have been sensitive to these small differences as they were observed in small rolling increments. Conversely, with
monthly billing, differences in the cost of peak electricity would have been bigger and more clearly observable. Billing at a
delay of one month exhibited a clear relationship between cost magnitude and consumption reduction.15 Finally, in order to
compare the KWh associated to the decrease of the consumption during the peak hours for the different tariff groups, we mul-
tiplied the estimated coefficients by the average consumption of the control group. During peak hours, consumers with

13
Thanks to a referee suggestion we also estimate the model with fixed effect. The results on the tariff dummies were not significantly different from the
random effects presented here, and they are available from the authors upon request.
14
Coefficients of tariff A and tariff D were statistically different at 6PM and at 7PM for monthly and IHD stimuli respectively.
15
Data during control period have been collected from July to December 2009. As a result, the first 7 months of the trial period (January-July 2010) do not
have the corresponding control period observations. Therefore, we are not able to check the coefficients for consumers with monthly bills for the first month of
the trial and check whether they started to respond immediately or with a monthly lag.
8 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

Table 1
Control and treatment period tariffs (euro/cents per kWh, including VAT).

Tariff Pre-treatment period Post-treatment period % change

Peak, day and night Peak Day Night Peak Day Night
Control 16.24 16 16 16 1.5 1.5 1.5
Tariff A 16.24 22.7 15.89 13.62 39.8 2.2 16.1
Tariff B 16.24 29.51 15.32 12.46 81.7 5.6 23.1
Tariff C 16.24 36.32 14.76 11.35 123.7 9.1 30.1
Tariff D 16.24 43.13 14.19 10.22 165.6 12.6 37.1
Tariff W/E 16.24 33.03 14.45 11.35 103.4 11 30.1

Data source: CER (2011).

bimonthly bills reduce their average consumption by 28.2 KWh, consumers with monthly bill reduce their average consumption
by 36.3 KWh and consumers with IHD reduced their average consumption by 38.3 KWh.

4.2. Day and night

4.2.1. Day
The effects of TOU tariffs and financial feedback were in the main limited to peak hours. During the majority of day time,
there were no significant consumption changes. That said, some changes in behaviour were observed the first part of the day,
from 10 to 12 AM. Consumers with monthly billing and IHD reduced consumption for certain tariffs relative to baseline (see
Table 3).
It is not clear why reductions were observed in the Monthly and IHD groups at this time. One possibility is that home
practices in the Monthly group that reduced usage during the peak period were also employed outside of peak times. There
is some evidence to support this. First, coefficients throughout the day indicated reduced usage for the Monthly group
(though not always significantly so) relative to pre-intervention usage. Second, the dominant strategy for all groups was
one of reduced usage rather than displaced usage, and reduction practices may be employed consistently during the day
if there was sufficient over-usage during the pre-intervention phase. Finally, as evidenced in the night-time hours, there
was some evidence of displacement of usage in all three groups - that is, consumers used more electricity during low cost
night-time hours than during the pre-intervention phase. This might have facilitated an all-day reduction strategy by con-
sumers in the Monthly group. That is, by shifting much of the high consumption activities to the night-time period, con-
sumers could more easily engage in a straightforward reduction strategy during daytime hours.
During the other times of the day, however, people under all the three stimuli do not react to change in tariffs. This result
is aligned with the findings of Di Cosmo et al. (2014) which highlight that the tariffs applied during the day in the trial period
are almost the same as the tariffs applied during the control period as shown by Table 1. However, these authors were focus-
ing on the day as the time of the day between 8AM to 5PM and again between 8PM and 11PM, without disentangling the
different patterns of consumption. Our model allows us to focus on specific hours. Focusing on the period between 10AM and
12AM shown in Table 3 it is possible to see that people reduce their consumption also during the first peak of the day which
occurs before 12AM. Moreover, focusing on the period after that and before the evening peak (from 12 to 5PM), our results
show that there are almost no significant changes between treatment and control.16 Finally, during the final hours of the day
(between 8PM and 10PM) no major changes were observed. Some consumers with IHD slightly reduced consumption at 8PM,
confirming the trend of over-reaction observed during peak hours, but the effect disappeared by 9PM. As discussed in the fol-
lowing paragraphs, our analysis highlights that during 2009 and 2010 consumers were willing to change their consumption
once they were able to controlling the appliances directly. If consumers were at home, they were also able to shift the use
of same appliances or reduce the consumption of electricity. First, the small amount of appliances that could be easily post-
poned (i.e. dishwashers, washing-machines etc.) probably were not enough to justify a great shift during different time periods.
Second, the lack of smart appliances (i.e. appliances that could be controlled remotely from the phone, timers etc.) probably
made the consumers change their consumption with respect to the control period only when they were able to do it directly,
that is, when they were at home.

4.2.2. Night
As shown in Fig. 1, the electricity prices are low after 11PM until the first hours of the morning. The TOU tariffs used in the
CBT trial mimic the price trend, being lower in the night than in the day and peak hours. As a result, in order to benefit from
the low prices, consumers should shift their consumption from the peak and daily hours to the night hours. Night results are
also aligned with the results by Di Cosmo et al. (2014), as consumers did not consistently change their behaviour during
night time. Sporadic significant increases in usage were identified however in all three groups. As mentioned previously,
it is possible that some displacement of usage into low-cost night-time hours occurred. Displacement can be difficult to cap-
ture however, since it is not synchronised across users. If displaced usage was spread across the night-time intervals then

16
Complete results are available from the authors upon request.
 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14 9

Table 2
Estimation results – peak hours, (kWh/day). All the other results will be available from the authors upon request.

Hour Stimulus A B C D
1700 Bimonthly 0.021 0.019 0.009 0.001
(0.016) (0.030) (0.017) (0.02)
Monthly 0.003 0.015 0.011 0.008
(0.018) (0.024) (0.018) (0.022)
IHD 0.021 0.015 0.001 0.003
(0.015) (0.03) (0.016) (0.026)
1800 Bimonthly 0.059⁄⁄⁄ 0.069⁄⁄ 0.035⁄⁄ 0.058⁄⁄
(0.018) (0.032) (0.017) (0.022)
Monthly 0.039⁄⁄ 0.06⁄⁄ 0.063⁄⁄⁄ 0.097⁄⁄⁄
(0.018) (0.025) (0.019) (0.26)
IHD 0.077⁄⁄⁄ 0.082⁄⁄ 0.066⁄⁄⁄ 0.071⁄⁄
(0.015) (0.035) (0.018) (0.029)
1900 Bimonthly 0.036⁄⁄ 0.081⁄⁄ 0.029⁄ 0.077⁄⁄⁄
(0.016) (0.03) (0.017) (0.022)
Monthly 0.054⁄⁄ 0.079⁄⁄⁄ 0.072⁄⁄⁄ 0.108⁄⁄⁄
(0.018) (0.023) (0.017) (0.03)
IHD 0.073⁄⁄⁄ 0.053⁄⁄ 0.072⁄⁄⁄ 0.107⁄⁄⁄
(0.015) (0.030) (0.016) (0.03)
1930 Bimonthly 0.006 0.041 0.004 0.02
(0.105) (0.029) (0.015) (0.019)
Monthly 0.026 0.0384⁄ 0.032⁄⁄ 0.029
(0.016) (0.022) (0.016) (0.022)
IHD 0.039⁄⁄ 0.039 0.024 0.047⁄⁄
(0.014) (0.03) (0.015) (0.024)
2000 Bimonthly 0.00019 0.0373 0.00697 0.0154
(0.015) (0.030) (0.015) (0.020)
Monthly 0.025 0.0069 0.015 0.010
(0.016) (0.021) (0.017) (0.024)
IHD 0.030⁄⁄ 0.021 0.008 0.178
(0.014) (0.029) (0.015) (0.024)
⁄
p < 0.1.
⁄⁄
p < 0.05.
⁄⁄⁄
p < 0.001.

Table 3
Average of significant coefficients, 10AM-12AM.

Bimonthly Monthly IHD

A NS 0.02872 0.04043
B NS 0.04527 NS
C NS 0.03257 NS
D NS 0.04456 NS

Complete results based on the half-hourly analysis are available upon request from the Authors. NS = not significant at 5% level.

this would be reflected in small non-significant increases. In fact, the coefficients for the vast majority of intervals were pos-
itive for all tariffs and feedback groups, indicating higher usage than during the pre-intervention period. However, small
increase in electricity consumption are significantly different from zero before 1AM and after 7AM in the morning for some
of the consumers in the trial, which is consistent with the reduction of the tariff during night hours, as shown by Table 4.
Focusing on the first hours of the day, in which rise in consumption will be beneficial in order to match wind generation
in the island, there are no significant change between treatment and control group. There are some little changes between
treatment and control groups at 5.30 and 7 in the morning, probably because people are able to postpone some appliances
like washing-machines and dishwasher just before going to work. However, the coefficients are slightly significant at the 5%
level, and cannot really compensate the reduction of consumption during peak hours. These results are particularly interest-
ing because they highlight that people generally prefer to reduce consumption instead of postponing appliances during other
times of the day. In order to determine how many KWh were consumed during the night and compare this number with the
reduction in KWh of electricity consumed during the peak hours we multiply the estimated coefficients by the average con-
sumption of the control group. Consumers with bimonthly bill increase their average consumption during the night hours by
7.9 KWh. Consumers with monthly bill increase their average consumption by 1.5 KWh and consumers with IHD installed in
their houses increase their consumption by 2.1 KWh. As a result, the rising in consumption of electricity during night hours
did not compensate the consumption decrease during the peak hours. This result has interesting implications. First, our
results supplement the findings of Di Cosmo et al. (2014). Instead of considering the night hours (from 23 to 7AM)
10 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

Table 4
Estimated coefficients, night, selected hours.

Bimonthly Monthly IHD

1130PM
C 0.040⁄⁄ 0.031⁄
(0.016) (0.015)
12PM
C 0.046⁄⁄ 0.033⁄
(0.015) (0.016)
D 0.064⁄⁄
(0.028)
1230PM
C 0.045⁄⁄
(0.016)
D 0.063⁄
(0.028)
1AM
C 0.038⁄
(0.016)
5.30AM
B 0.054⁄
(0.028)
7AM
C 0.04⁄⁄
(0.019)

Robust standard errors in parenthesis, only statistically significant coefficients are shown. p < 0:001.
⁄
p < 0.1.
⁄⁄
p < 0.05.

generically, we are able to observe how the consumer’s behaviour change during the different hours of the night. From this
analysis, it is clear that consumers find it easier to shift the usage of electric appliances to the first hours of the night (before
1AM) and in the last hours, those prior to the start of the working day.
Second, the evidence against nocturnal consumption, despite the low prices, may be due to lack of available technologies
that facilitate this choice (i.e. timer or specific ”night” program etc.). Further studies should be done once smart appliances
will be available to understand if those appliances significantly impact on consumers choices.
Finally, policy makers should be careful in assuming that price signal will be enough to make consumers shift from day to
night consumption, in order to accommodate cheap wind generation in the Irish context. Our results show that there is just a
little activity during the first hours of the night and the of the morning, but during the night there are no significant changes
in consumption patterns. However, the (partial) electrification of heating and transport, associated to smart technologies can
dramatically change our results.

4.3. Education

In order to check the robustness of our results we ran the same regression described in the previous chapter controlling
by education level achieved by the head of household, as other variables such as income or occupational status were not
available.17 Education is a well known proxy for income and social status, Becker and Chiswick (1996). As a result, we examine
here whether these variables also affect the behavioural response of consumers. We define as ”Highly educated” (HE) house-
holds, those in which the head of the household achieved at least the tertiary education (degree in university). ”Less educated”
(LE) are all the other households in the sample. This distinction allowed us to identify interesting differences in half hourly
behaviour during peak hours in these groups, that are shown by Table 5 below and summarised by the following Figure. Results
for day and night do not change significantly including the education dummy, and are not reported here.18 However, peak
results indicated that HE households are more sensitive to the increased peak tariffs, as they tend to reduce the consumption
of electricity in those hours more than the rest of the sample. One important consequence of this outcome is that HE households
had the potential to save more money than LE households following the introduction of the peak tariffs.
As for the other results of our analysis, treatment dummy is significant and negative, and group dummy is not significant.
As the results for peak hours, sunshine is negative and significant where the other dummies (electric heating, appliances,
bank holiday) are positive and significant. Table 5 also shows that there are not statistically differences between people
who achieved tertiary education and the full sample. However, after 7PM education dummy becomes significant at the

17
See Di Cosmo et al. (2014).
18
Complete results are available from the authors upon request.
 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14 11

Table 5
Estimation results – education, peak hours, (kWh/day).

Hour Stimuli A B C D Edu dummy

1700 Bimonthly 0.0186 0.03003 0.006394 0.00231 NS
(0.017) (0.03) (0.017) (0.02)
Monthly 0.00157 0.01036 0.01047 0.00357 NS
(0.018) (0.025) (0.018) (0.023)
IHD 0.02019 0.010775 0.00053 0.00415 NS
(0.016) (0.033) (0.017) (0.027)
1800 Bimonthly 0.0577⁄⁄ 0.08052⁄⁄ 0.0392⁄⁄ 0.0559⁄⁄ NS
(0.019) (0.033) (0.018) (0.023)
Monthly 0.0387⁄⁄ 0.0627⁄⁄ 0.067⁄⁄⁄ 0.0986⁄⁄⁄ NS
(0.019) (0.026) (0.019) (0.027)
IHD 0.077⁄⁄⁄ 0.0837⁄⁄ 0.066⁄⁄⁄ 0.071⁄⁄ NS
(0.016) (0.036) (0.018) (0.029)
1900 Bimonthly 0.038⁄⁄ 0.0888⁄⁄ 0.0321⁄ 0.07688⁄⁄⁄ S⁄⁄
(0.016) (0.031) (0.017) (0.023)
Monthly 0.0522⁄⁄ 0.07809⁄⁄⁄ 0.07171⁄⁄⁄ 0.11068⁄⁄⁄ S⁄⁄
(0.019) (0.024) (0.018) (0.028)
IHD 0.0742⁄⁄ 0.0556⁄ 0.074⁄⁄⁄ 0.1065⁄⁄⁄ S⁄⁄
(0.015) (0.031) (0.016) (0.030)
1930 Bimonthly 0.00791 0.04788 0.00677 0.01837 S⁄⁄
(0.015) (0.029) (0.016) (0.020)
Monthly 0.02183 0.03551 0.03087⁄ 0.02776 S⁄⁄
(0.017) (0.023) (0.017) (0.023)
IHD 0.03997⁄⁄ 0.0353 0.0289⁄ 0.04676⁄ S⁄⁄
(0.015) (0.03) (0.015) (0.024)
2000 Bimonthly 0.0120 0.0442 0.0048 0.0111 S⁄⁄
(0.015) (0.030) (0.016) (0.02)
Monthly 0.202 0.0008 0.013 0.0077 S⁄⁄
(0.016) (0.022) (0.017) (0.025)
IHD 0.0286⁄⁄ 0.016 0.011 0.0153 S⁄⁄
(0.0144) (0.030) (0.015) (0.025)

NS = not significant, S = significant.

⁄
p < 0.1.
⁄⁄
p < 0.05.
⁄⁄⁄
p < 0.001.

5% level. As a result, there are some statistically significant differences between HE households and the rest of the group at
the end of the peak period. This results is shown also in the following Figure:
Differences due to education and/or income have important policy implications. Our results highlight that HE households
with higher frequency information (monthly-bills or IHD) are more elastic in responding to changes in tariffs than the sam-
ple average in some specific hours of the day. This finding highlights the additional information that can be gleaned by inves-
tigating the daily dynamics of household responses to electricity tariffs. Previous work on this dataset by Di Cosmo et al.
(2014) did not identify these effects because peak hours were considered as a block (see Fig. 3).
The CBT trial was set up in such a way that participants were fully compensated for losses associated to the change in
tariffs; however, if the policy makers are keen to introduce TOU tariffs they should be aware that educated people are likely
to save more money during peak hours than the average of the sample.
Second, as education may be considered a proxy for the income level, HE households may consume more electricity than
LE households in non-essential activities. Comparing the data on electricity consumption of the highly educated consumers
with the less educated, lead to interesting results: the average per-capita electricity consumption during the whole period
(control and treatment) is 1.66 KWh per hour for HE households and 1.50 KWh for LE households. The same results for the
treatment period only are lower for both groups (1.61 KWh for highly and 1.46 for less) but highly educated reduce their
consumption more than the less educated during the trial. Looking at the percentiles of the consumption distribution
emerges that people in the highly educated group consume more than the less educated (see Table 6).
Thus, it is possible to assume that highly educated households contracted their electricity consumption more than less
educated ones because the electricity consumption is associated with non-essential activities. Thanks to the information
provided by monthly and IHD installed, it was possible for people in this group to monitor their consumption and reduce
it accordingly to the new tariffs.
Finally, Table 5 shows that HE households in the IHD group keep electricity consumption low even after the end of the
peak tariff at 7PM. This result is similar to what we found for the full sample. As mentioned previously, if non-essential usage
was reduced in these houses, then effects on non-essential usage might have generalised beyond the end of the peak interval.
Such an over-reaction might be more likely occur when household were adjusting to the novel electricity prices and calibrat-
ing their usage in accordance with it.
12 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

4pm to 8pm (Higher Ed Houses)

sig
A B C D
0.05 FALSE

TRUE
Coefficient
0.00
feedback
−0.05
Bimonthly

−0.10 Monthly

IHD
16:30
17:00
17:30
18:00
18:30
19:00
19:30

16:30
17:00
17:30
18:00
18:30
19:00
19:30

16:30
17:00
17:30
18:00
18:30
19:00
19:30

16:30
17:00
17:30
18:00
18:30
19:00
19:30
Time

Fig. 3. Plot of estimated coefficients at half hourly intervals before, during and after the peak pricing period, by feedback stimulus and tariff for the whole
sample. Vertical grey dashed lines indicate the beginning and end of the peak pricing period. Coefficients less than zero (below the horizontal dashed line)
indicate lower usage during that half hour period. Solid circles indicate reductions that were statistically significantly different from baseline usage and
open circles denote usage that was not statistically different. Plots were developed using ggplot2 \cite{ggplot2}.

Table 6
Percentiles of consumption, low and high education.

LE HE
1% 0.012 0.015
5% 0.048 0.052
10% 0.085 0.087
25% 0.187 0.192
50% 0.426 0.462
75% 0.97 1.102
90% 1.892 2.094
95% 2.556 2.781
99% 4.031 4.245

In order to understand whether this over-reaction disappeared after the first months of the trial or if it were stable until
the end of the trial, we ran the model just for the final months of the trial (from September to December). Our results confirm
that HE households with IHD installed at home reduced their consumption after 7PM with respect to the control period, but
in December this effect weakens (it becomes statistically different from 0 at the 10% level). Unfortunately, we cannot predict
how consumers change behaviour in the long run, as the trial ended in December 2010. However, this result may indicate
that there are changes in the behaviour through the end of the trial. The current study investigated the effects of electricity
pricing on the daily dynamics of electricity usage. Further investigation is required to track the dynamics of electricity usage
from day to day or month to month to highlight how households adjust to novel pricing structures. Longer studies are
required to provide detail on the medium to long term stability of pricing effects on electricity usage.

5. Conclusions

The current paper examined the half-hourly behaviour of consumers exposed to a change in the electricity tariff and to
billing information based on data collected during the trial conducted by the Irish Electricity and Gas Regulation Authority
(CER) between 2009 and 2010 to estimate the reaction to tariff changes by different groups of consumers. We focused on the
change in consumption during the three schemes applied in the trial (day, peak and night) and during the half-hours imme-
diately before and after these tariffs, focusing on the peak (from 5PM to 7PM) and the night (from 11PM to 7AM) hours. In
this large-scale investigation of electricity consumption, TOU pricing and financial feedback influenced the degree to which
consumers reduced usage. Consumption reduced during the peak period and then rebounded to pre-peak levels after the
period. For households with IHD, reduced consumption persisted for one half hour post-peak.
In the main, the data suggest that households reduced consumption rather than shifting consumption from peak. In par-
ticular, one strategy that was expected was that households might simply delay consumption until the end of the peak per-
iod. If such a pattern was common across households, post-peak spikes in usages above baseline would have been recorded,
but they were not. It is possible that some consumption was shifted from peak to other times of day, but that it was not done
in a coordinated fashion and thus was not visible in the overall numbers. There were some half-hourly periods across the day
that had significantly increased consumption, but these periods did not conform to any obvious pattern of usage and these
periods had relatively low baseline usage which might have exaggerated these effects.
The choice of not shifting all the consumption from the peak to the night and then benefit from low tariffs can be ascribed
to different factors. First, some reduction of electricity usage during peak hours may simply reflect misuse of the appliances.
 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14 13

Higher prices make consumers more aware about the cost of electricity and induce them to care more about the appliances
used during peak hours and to behave more carefully. Second, the lack of appropriate technologies (such as integrated
timers, mobile appliances etc.) possibly made difficult for the consumers to switch on some appliances during the night.
Finally, consumer preferences also may have played a role in the lack of shifting. Consumers may have perceived that imme-
diate access to heat, light and entertainment compete and often outweigh delayed financial costs. In line with this interpre-
tation, consumers provided with financial information at the greatest delay (the bi-monthly billing group) were least
sensitive to TOU tariffs. In addition, consumers provided with immediate financial information through in-home displays
showed the greatest reductions in usage and reductions persisted beyond the end of the peak period.
The observation than households reduced consumption rather than shifting electricity usage to another time of day may
indicate that some peak electricity usage might not be moveable. Social and work practices determine the timing of certain
activities that require electricity usage. The most obvious example of this is dinner time, which, for many households, will
have to after children complete school and adults complete work. Other households will synchronise dinner time with these
limited households, because evening social activities will be timed to facilitate the greatest number. Since some activities
that require electricity cannot be moved, there is less usage that can be shifted from peak times to other times in the
day. However, it is worth noting that our social and work practices (e.g., email usage) are facilitated and updated by tech-
nological advances (Shove, 2004), as are our expectations of comfort (e.g., air conditioning), so technological advances have
the potential to alleviate or exacerbate the problems of scheduling electricity usage efficiently.
There are potential policy implication when one considers the response of highly educated households (i.e. with tertiary
degree) to price changes compared to the response of lower educated households. First, the education level of the head of the
household has an effect on response to tariffs or feedback for consumers in the monthly-bill and the IHD groups. The changes
in electricity consumption between the highly educated households and the average sample are not huge, consequently, the
introduction of TOU pricing would not be expected to unfairly target low income or low education households. However, our
findings show that better educated households may have greater capability to contract their consumption during peak hours
more than less educated households, saving more money after the change in tariffs. Our results also show that both in the
trial and in the control period, HE households consume (on per-capita average) more electricity than less educated people.
Education may be considered a proxy of the income level, and the electricity consumption for highly educated consumers
may be associated with activities that are easier to reduce or postpone.
Second, the impact of IHD on electricity usage was such that even relatively minor increases in peak pricing gave rise to
considerable reductions in usage. Given that more punitive tariffs affect behaviour through greater monetary costs on the
customer, there is a potential for significant monetary losses by households that fail to respond to peak pricing. By employing
IHD, smaller costs with smaller exposure to these negative side effects can have greater impact on consumer behaviour.

Appendix A. Statistics of the control and the treatment groups

Table A1.

Table A1
Mean comparison of exogenous variables, by hours.

Control Bimonthly Monthly IHD

Internet 0.47 0.51 0.52 0.51
Washing machine 0.81 0.82 0.83 0.80
Tumble dryer 0.57 0.55 0.58 0.54
Dish washer 0.53 0.55 0.58 0.51
Freezer 0.40 0.43 0.43 0.40
TV small 0.53 0.54 0.56 0.53
TV big 0.66 0.70 0.71 0.69
Desktop computer 0.36 0.40 0.40 0.38
Laptop 0.42 0.44 0.44 0.45
Game consolle 0.25 0.28 0.27 0.28
Cooker 0.58 0.57 0.57 0.51
Shower 0.69 0.69 0.71 0.67
Electric heat 0.06 0.05 0.07 0.05
Water heating 0.47 0.48 0.48 0.46
Age1825 0.00 0.01 0.00 0.01
Age2635 0.08 0.08 0.07 0.08
Age3645 0.15 0.17 0.18 0.17
Age4655 0.19 0.19 0.20 0.20
Age5665 0.17 0.18 0.20 0.18
Age65þ 0.23 0.19 0.19 0.17
High educated 0.52 0.50 0.49 0.47
Less educated 0.29 0.31 0.33 0.32
14 V.D. Cosmo, D. O’Hora / Journal of Economic Psychology 61 (2017) 1–14

References

Allcott, H. (2011). Consumers’ perceptions and misperceptions of energy costs. American Economic Review, 101(3), 98–104.
Aubin, C., Fougare, D., Husson, E., & Ivaldi, M. (1995). Real-time pricing of electricity for residential customers: Econometric analysis of an experiment.
Journal of Applied Econometrics, 10(S1), S171–S191.
Becker, G. S., & Chiswick, B. R. (1996). Education and the distribution of earnings. American Economic Review, 56(1/2), 358–369.
CER, 2011. Electricity smart metering customer behaviour trials (cbt) findings report. Technical report, CER.
Commission, E., 2015. Communication from the commission to the european parliament, the council, the european economic and social committee, the
committee of the regions and the european investment bank. a framework strategy for a resilient energy union with a forward-looking climate change
policy. Technical report, European Commission.
Di Cosmo, V., Lyons, S., & Nolan, A. (2014). Estimating the impact of time-of-use pricing on irish electricity demand. The Energy Journal, 35(2).
Eirgrid, 2009. Short term active response. Technical report, Irish System Operator.
Eirgrid, 2014. Annual renewable energy constraint and curtailment report. Technical report, EIRGRID.
Faruqui, A., Harris, D., & Hledik, R. (2010). Unlocking the 53 billion savings from smart meters in the eu: How increasing the adoption of dynamic tariffs
could make or break the eu’s smart grid investment. Energy Policy, 38(10), 6222–6231. The socio-economic transition towards a hydrogen economy -
findings from European research, with regular papers.
Faruqui, A., & Sergici, S. (2010). Household response to dynamic pricing of electricity: A survey of 15 experiments. Journal of Regulatory Economics, 38(2),
193–225.
Feuerriegel, S., & Neumann, D. (2016). Integration scenarios of demand response into electricity markets: Load shifting, financial savings and policy
implications. Energy Policy, 96, 231–240.
Filippini, M. (1995). Swiss residential demand for electricity by time-of-use: An application of the almost ideal demand system. The Energy Journal, 16,
27–39.
Finn, P., OĆonnell, M., & Fitzpatrick, C. (2013). Demand side management of a domestic dishwasher: Wind energy gains, financial savings and peak-time
load reduction. Applied Energy, 101, 678–685. Sustainable Development of Energy, Water and Environment Systems.
Frederick, S., Loewenstein, G., & O’Donoghue, T. (2002). Time discounting and time preference: A critical review. The Journal of Economic Literature, XL(June),
351–401.
Gabaix, X., & Laibson, D. (2006). Shrouded attributes, consumer myopia, and information suppression in competitive markets. Quarterly Journal of Economics,
121(2), 505–540.
Green, L., Myerson, J., Lichtman, D., Rosen, S., & Fry, A. (1996). Temporal discounting in choice between delayed rewards: The role of age and income.
Psychology and Aging, 11(1), 79.
Jessoe, K., & Rapson, D. (2014). Knowledge is (less) power: Experimental evidence from residential energy use. American Economic Review, 104(4),
1417–1438.
Johnson, M. W., & Bickel, W. K. (2002). Within-subject comparison of real and hypothetical money rewards in delay discounting. Journal of the Experimental
Analysis of Behavior, 77(2), 129–146.
Joskow, P. (2012). Creating a smarter us electricity grid. Journal of Economic Perspectives, 26(1), 29–48.
Ito, K., Ida, T., Tanaka, M., 2015. The Persistence of Moral Suasion and Economic Incentives: Field Experimental Evidence from Energy Demand.
Kirby, K. N., & Maraković, N. N. (1996). Delay-discounting probabilistic rewards: Rates decrease as amounts increase. Psychonomic Bulletin & Review, 3(1),
100–104.
Loewenstein, G., & Thaler, R. H. (1989). Anomalies: Intertemporal choice. The Journal of Economic Perspectives, 181–193.
Lunn, P., Somerville, J., 2015. Surplus identification with non-linear returns. ESRI Working Paper.
McKenna, K., & Keane, A. (2015). Residential load modeling of price-based demand response for network impact studies. IEEE Transactions on Smart Grid, PP
(99), pp. 1-1.
Odum, A. L. (2011). Delay discounting: I’m a k, you’re a k. Journal of the Experimental Analysis of Behavior, 96(3), 427–439.
Reimers, S., Maylor, E. A., Stewart, N., & Chater, N. (2009). Associations between a one-shot delay discounting measure and age, income, education and real-
world impulsive behavior. Personality and Individual Differences, 47(8), 973–978.
Scott, S., Lyons, S., Keane, C., McCarthy, D., Tol, R.S.J., 2008. Fuel poverty in ireland: Extent, affected groups and policy issues. Technical report, ESRI.
Shove, E. (2004). Efficiency and consumption: Technology and practice. Energy & Environment, 15(6), 1053–1065.
Walsh, D., Malaguzzi Valeri, L., Di Cosmo, V., 2016. Strategic bidding, wind ownership and regulation in a decentralised electricity market. MPRA Paper
71502, University Library of Munich, Germany.
Wickham, H. (2009). ggplot2: Elegant graphics for data analysis. New York: Springer.