Smart meter devices and the effect of feedback on residential electricity consumption: Evidence from a natural experiment in Northern Ireland

TL;DR: In this article, the authors estimate the effect of real-time usage information on residential electricity consumption in Northern Ireland using a unique set of data and exploiting a large-scale natural experiment.

About: This article is published in Energy Economics.The article was published on 2013-03-01 and is currently open access. It has received 248 citations till now. The article focuses on the topics: Smart meter & Consumption (economics).

Summary

1. Introduction and Motivation

• Residential buildings account for a large share of the world’s energy consumption, and offer a natural target for policies that seek to reduce CO2 emissions from (fossil-fuel) power generation, dependence on imported fuels, and vulnerability to supply shocks.
• Many observers argue that providing better information and feedback on consumption helps improve energy conservation and energy efficiency in the residential sector— by itself or when combined with other traditional policy tools such as economic incentives, pricing and regulation.
• Moreover, because this plan requires prepayment, it suggests that households on it will be monitoring their usage.
• The authors examine this matter using data from 18 waves of Northern Ireland’s Continuous Household Survey (from 1990 to 2009), which they merge with price and plan information from the electricity utility, and weather data.
• Section 4 presents the model and the empirical approach.

2. Previous Literature

• Imperfect information and uncertainty about the price of electricity have received much attention in the energy economics literature.
• AC C EP TE D M AN U SC R IP T ACCEPTED MANUSCRIPT 6 Hartman (1988) finds that audits do decrease energy usage, but that failure to account for self-selection grossly overstates the impact of the audit program.
• Smart meters are two-way wireless communication devices that i) measure, store and transmit usage data to the utility at regular intervals, allowing it to monitor usage and bill the customer for it remotely without having to physically read the meter, and ii) can be used to convey real-time tariff changes, supply-wide conditions and peak-load information to the consumer (Darby, 2010).
• Fischer (2008) selects 26 projects from various countries over 1987-2006 where feedback about electricity consumption was provided to residential customers, and concludes that overall feedback does reduce usage by 1-20% percent, with “usual” savings in the 5-12% range.

3. Background on Utilities and Pricing Schemes in Northern Ireland

• The authors study period is 1990-2009, and during this time Northern Ireland Electricity (NIE) was the electric monopoly for the residential sector in all of Northern Ireland.
• If, at the time of issuing a new bill, there is less than 10% balance on their card (or less than £10), they receive a discount.
• The keypad meters combine prepayment with an interactive display that allows consumers to easily monitor their electric usage and cost.
• As of November 2010, households on the keypad accounted for 34% of the NIE residential customer base, direct debit monthly plans for 26%, direct debit quarterly for 4.7%, budget accounts for 0.2% and EasySavers for 6.3%.

A. Theoretical Motivation

• The authors are interested in modeling the response to information that a typical prepayment customer will experience after the introduction of the keypad device.
• The authors argue that inattention is unavoidable: For many consumers, the 13 Personal communication from Gerry Forde, NIE, 15 December 2010.
• Faruqui et al. (2010) report that worldwide over 5 million customers use power on a pre-payment basis, with the bulk of pre-payment users in the UK, New Zealand and South Africa.
• As a result, little monitoring of usage occurs, and consumers imperfectly observe their electricity usage.
• This model, however, does not provide unambiguous predictions as to whether an exogenous change in information increases or decreases monitoring and electricity consumption.

B. The Experiment and the Treatment

• If the assignment to the treatment and control groups is not random, the right-hand side of (1) contains an additional term, namely the selection bias, which is equal to .
• As mentioned, in this paper, the authors exploit the fact that in April 2002, NIE introduced a new metering device—the keypad—that allows customers to track consumption in real time, and a new pricing structure for its prepayment plan.
• Since the price depends on the plan, customers select into their plan, and plan choice may be correlated with energy use patterns, there is potential for selection bias.

D. Electricity Demand

• The authors begin with the demand equation: (2) Subscripts i, j and t denote the household, area where the household resides, and wave of the CHS surveys, respectively.
• The authors allow for possible correlation between and , which makes electricity usage and the choice of plan endogenous.
• Bourguignon et al. (2007) compare the performance of the Dubin-McFadden correction term in (6) with a simplified version that imposes the constraint that the α coefficients sum to zero, and with the selection correction procedures developed by Lee (1983) and Dahl (2002).
• The authors control for unobserved heterogeneity by including ward-specific intercepts (the in equation (2)), under the assumption that the households and/or the dwellings in a ward are similar.

E. The Effect of Usage Information on Usage

• The question at the heart of this paper is whether providing feedback about consumption of electricity makes consumers change their usage levels.
• In April 2002 NIE replaced the powercard plan with the keypad plan, which substituted the old meter with a more advanced device that displayed real-time information.
• Formally, (7) ijtikk km imt imtimt mijtitititjijt eP P PP INCpE ˆln ˆ1 ˆlnˆ lnlnln 210 δDγx where D is a vector of dummies for the electricity scheme the household is on, and vector captures the effect that the type of plan has on electricity, above and beyond that of the price associated with that plan.
• In Belfast County borough district, for example, there are 52 wards.
• An effect different from zero suggests less-than-perfect information , which the meter helps correct.

F. The Choice of Independent Variables

• Vector x in equations (2), (6) and (7) is comprised of variables that the authors expect to influence to the demand of electricity directly (e.g., house size, etc.) or via the cost of monitoring.
• It also includes the number of years the household has been living in this home to proxy for the household’s familiarity with the energy efficiency of this dwelling and the vintage of heating and electrical equipment.
• Education and other household characteristics may also serve as proxies for the cost of monitoring electricity usage.
• It is also likely that individuals may choose a plan over another based on word of mouth or this plan’s popularity with neighbors and friends.

5. The Data

• Attention is restricted to those households that presumably have a reasonable degree of control over the use of energy at their premises.
• In subsequent regressions, the authors further exclude households that rented their dwelling from the Housing Executive (i.e., public or assisted housing, which account for 21.77% of the original sample) or from a housing association (a private charity that provides low-cost housing: 2.40% of the original sample), which results in a sample of 34,779 observations.
• 20 We also computed 19 Data from the UK Department of Energy and Climate Change show that the usage of electricity in homes with Economy7 plans and electric storage heaters is 50% larger than its counterpart in homes with standard meters.the authors.the authors.
• The authors therefore constructed daily average temperatures, and hence heating degree days for i) all of the Northern Ireland monitoring stations, ii) only the Belfast locations, and iii) only the Dublin airport monitor.
• Descriptive statistics for electricity consumption and prices are displayed in table 8. Prices are all deflated to 2009 constant British i), but virtually identical results are obtained when the authors use the others.

A. Electricity Demand

• The authors first order of business to estimate the electricity demand (6), using the two-step approach described in section 4.
• The coefficients on the missing income and top-coded dummies are positive and significant, suggesting that households that do not report their income might be wealthy or otherwise have significantly larger electricity consumption than those that do.
• Housing characteristics are likewise associated with energy consumption.

C. Robustness Checks

• The authors also re-run models (D)-(F) with the Lee and Dahl selection correction terms instead of the unrestricted Dubin-McFadden approach.
• Households with very low levels of usage (due to preferences for conservation or low income) are unlikely to be able to reduce consumption even further, and may choose to improve comfort, once they realize that they are on track relative to their usual bills.
• The coefficients are similar to their counterparts in table 11, and the “average treatment effect” of the introduction of the keypad ranges from a 10.45% reduction in electricity usage (specification (A)) to a 18% reduction (specification (F)).
• Panel (B) of table 13 checks the effect of model specification decisions.
• When the selection correction terms are not included in the model, which means that no allowance is made for the endogeneity of price and plan choice, the effect of the keypad is a little less pronounced.

7. Discussion and Conclusion

• The authors investigation suggests that households do respond to the provision of information by using less electricity, even accounting for type of home, heat, household characteristics and possible selection of households into pre-payment plans.
• From the utilities’ perspective, these systems reduce the costs of metering, billing, and investigating outages.

Figures

Table 13. Robustness checks. All regressions are based on the specifications of table 11, unless otherwise indicated.

Table 11. Electricity Demand: Effect of Price, Income, House and Household Characteristics. Dep. Var.: ln kWh per quarter.

Table 2. Discounts offered to specific electricity plans in Northern Ireland.

Table 1. Northern Ireland Electricity Tariffs.

Table 11. Electricity Demand: Effect of Price, Income, House and Household Characteristics. Dep. Var.: ln kWh per quarter.

Figure 1. Log electricity consumption: Average for prepayment (treatment) and control households (not adjusted for household characteristics) by year.

Table 10. Electricity Demand: Effect of Price, Income, House and Household Characteristics. Dep. Var.: ln kWh per quarter. T statistics in parentheses. CHANGE FOR D-F

Table 3. Composition of the sample by year.

Frequently asked questions

Q1. What are the contributions in "Smart meter devices and the effect of feedback on residential electricity consumption: evidence from a natural experiment in northern ireland" ?

In this paper, the effect of real-time usage information on residential electricity consumption in Northern Ireland has been evaluated using a large-scale natural experiment. 

Q2. What are the future works mentioned in the paper "Smart meter devices and the effect of feedback on residential electricity consumption: evidence from a natural experiment in northern ireland" ?

32 Assuming no changes in the operating costs in the future, total per-unit costs for the life of keypad devices ( assumed to be 10 years ) are £62-73. There are other environmental benefits, such as reductions in the emissions of conventional air pollutants associated with power generation, and energy security benefits associated with reduced energy usage, which the authors do not attempt to estimate in this paper, but that are likely to be sizeable and deserve future research. It is higher than the actual prices of Certified Emissions Reductions ( CER ) that can be bought and sold on the European AC C EP TE D M AN U SC R IP T ACCEPTED MANUSCRIPT 35 Perusal of the results reported in tables 11 and 12 suggests that most of their estimated reductions are cost-effective, and that their most conservative estimate of the effect of the keypad barely misses the £ 25/tonne mark. 

Q3. What is the definition of fuel poverty?

Government estimates suggest that about one-third of the households in Northern Ireland are “fuel poor,” with fuel poverty being defined when 10% or more of the household income is spent on all household fuel use (DSDNI, 2006). 

Q4. How do the authors avoid losing the observations from those households that do not report income?

To avoid losing the observations from those households that do not report income, the authors create a companion missing income dummy, and recode income to zero when not reported. 

Q5. What did Owen and Ward (2007) report about the powercard system?

Owen and Ward (2007) report that the powercard system required visiting the customer’s home to change tariffs, was vulnerable to theft and fraud, and required tracking down and reconciling usage and billing. 

Q6. What is the way to enhance or manipulate the feedback provided by regular utility bills?

54 One way to enhance or manipulate the feedback provided by regular utility bills is to augment it with “social norms” contents. 

Q7. What are the only two North American instances of pre-pay pricing plans?

The only two North America instances of pre-pay pricing plans reported in Faruqui et al. are offered by Woodstock Hydro in Ontario, Canada and the Salt River Project utility in Arizona. 

Q8. What are the main aspects of the feedback provision that are successful at reducing usage?

She identifies aspects of the feedback provision that were most successful at reducing usage, such as breakdown by appliance, computerized and very frequent feedback—which are made possible by advanced metering—and sufficiently long project duration. 

Q9. How many households were randomized to receive information about their energy usage?

In a randomized field experiment involving 80,000 households in Minnesota, information about the energy usage of neighbors and visual cues about doing “better” or “worse” in electricity usage relative to similar neighboring homes has been found to reduce energy consumption by 1.9% relative to the baseline (Allcott, 2008).