Elicitation of electricity consumption habit formation among new subscribers

Abstract

In this study, our aim is to explore the formation of electricity consumption habits in Cote d'Ivoire, specifically focusing on new subscribers. The growth of residential electricity consumption globally has been influenced by these habits, which can lead to inertia and inefficient energy usage. Using the unique database of the National Electricity Company, we uncover that electricity consumption habits exhibit a high level of persistence, and this persistence tends to strengthen over time. These findings highlight the need to develop policies that encourage electricity efficiency among new subscribers, to combat energy inefficiency and foster positive consumption habits at the household level.

1. Introduction

Forecasting both energy production and consumption has encountered a growing interest in the literature due to the important role of energy for the economy [1]. Most of these studies have focused on the aggregate energy production and consumption in several contexts and using different approaches including time series analysis and machine learning techniques [ [2,3]]. The diversification of approaches, coupled with the large literature on energy forecasting, emphasizes the importance of energy forecasting in facilitating improved planning for power generation [4] on the one hand, and flexibility of energy demand [5] as well as energy conservation and efficiency for climate change mitigation on the other hand.

Electricity consumption is experiencing a substantial and rapid growth worldwide. Data provided by the International Energy Agency (IEA) indicates that total electricity consumption has surged from 6812 TW-hours (TWh) in 1980 to 22,847 TWh in 2019. This remarkable expansion is primarily driven by the industry and residential sectors, which collectively accounted for 68.4 percent of the total electricity consumption in 2019. Over the period from 1980 to 2019, residential electricity consumption has multiplied by 3.5, surpassing the growth rate of 2.8 observed in the industry sector. Consequently, residential electricity consumption has become the second-largest category of electricity consumption, trailing only the industrial sector [6].

This rapid growth of residential electricity consumption is due to (i) population growth combined with rapid urbanisation and increasing access to electricity, (ii) changes in consumption pattern, and (iii) energy inefficiency. First, the World bank reports that the population grew from 6 to 8 billion over the period 2000-19 and the proportion of the population with access to electricity has grown from 78.2 percent to 90 percent over the same period. As a result, the total residential electricity consumption has almost doubled, from 3562 TWh in 2000 to 6072 TWh in 2019. Second, urbanisation plays an import role in increasing electricity consumption [ [7,8]]. Existing literature shows that urban households consume more electricity than rural ones [9], and the urban-rural electricity consumption gap might be as high as 23 percent.

In addition to the effect of population growth and improved access to electricity, several changes in electricity consumption patterns resulted in energy inefficiency. The electricity consumption per capita has nearly doubled from 1980 to 2019 and average electricity consumption per household is increasing by 4.5 percent per year [10], because of rapid proliferation of electronic devices [11] and standby and active appliances [ [10,11]]. Appendix Figure A5.1 illustrates the linkages between the usage of electric appliances and energy inefficiency based on the literature. Energy inefficiency can be due to non-optimal use of appliances or the use of inefficient appliances [12]. The literature suggests that energy inefficiency is more persistent and difficult to curb than energy efficiency [ [13,14]]. Indeed, persistent and anchored electricity consumption habits are likely to annihilate the effectiveness of energy efficiency promotion policies [ [15,16]]. Therefore, to what extent electricity consumption habits are persistent? And how do electricity consumption habits form?

This is the purpose of this paper which aims at analysing the electricity consumption habits of new subscribers in Côte d’Ivoire. The contribution of this paper is twofold. First, it sheds light on the formation of electricity consumption habits among new subscribers, using a unique dataset obtained from the National Electricity Company's consumption monitoring system. Since bad energy habits are more difficult to change, analysing the consumption patterns of new subscribers is particularly important for two reasons. On the one hand, it is well documented in the literature that energy inefficiency is more persistent than energy efficiency [ [13,14]]. In fact, the household level energy inefficiency is mainly structural [17]. Thus, combating these inefficiencies can be effective by promoting the adoption of energy saving appliances through selected initial investments and upgrades; that is easiest for new subscribers [16]. On the other hand, if new subscribers are well oriented toward efficient initial investments, only inefficiency due to bad energy consumption behaviours can be observed. In such a situation, any campaign promoting energy conservation can be effective in reducing energy inefficiencies [17]. This might include initiative like informative bills [18] or any initiative that might shift subscribers from being imperfectly informed or unaware of energy saving. The findings of this paper can be helpful to design well-targeted policies to reduce energy consumption.

Second, we focus on Côte d’Ivoire because it has an emerging and dynamic residential electricity sector that, to the best of our knowledge, has never been the subject of similar study. Therefore, this paper is the first attempt to study electric habit formation in the country. Indeed, the electricity coverage rate, i.e., the proportion of electrified localities, has grown from 34 percent in 2012 to 82 percent in 2021, and the households access rate to electricity has grown from 25 percent to 60 percent over the same period. The electricity demand grew from 3.4 TWh in 2000 to 9.7 TWh in 2021 [19]. Even if the electricity consumption due to the residential electricity sector in Côte d’Ivoire has not grown substantially (36.6 percent of the electricity consumed in 2019 versus 35.9 percent in 2010), the per capita electricity consumption which is 302 kWh in 2019 has been multiplied by 1.6 since 2010 [20].

The dynamic of the residential electricity demand has been enhanced by several policies implemented to increase the access of households to electricity. These include the rural electrification programme (in French, Programme National d’Electrification Rural, PRONER) and the electricity for all program (in French, Programme Electricité Pour Tous, PEPT), both launched in 2014, the former aimed at improving the electricity coverage rate and the latter at facilitating access to electricity for low capabilities households by subsidising the initial cost of electricity access. The government's social program (PS-Gouv) aimed at reducing electricity prices for low-income consumers and implemented in 2019, has further strengthened the Electricity for All initiative. Data provided by the national electricity regulator reveals a significant increase in the number of new subscribers, from 93,527 in 2006 to 202,780 in 2021 [21]. In 2021, approximately 62 percent of the new subscribers were beneficiaries of the PEPT programme.

Our empirical strategy draws on the external habit formation literature to measure the persistence of habit formation among new subscribers in Côte d’Ivoire. Additionally, we account for various factors that can influence residential electricity consumption. These factors include climatic conditions, characteristics of the household and its head, and the prevailing electricity market conditions, all of which are known to be significant determinants of electricity usage. The findings of this study reveal the presence of highly persistent electricity consumption habits among new subscribers, with the level of persistency increasing as the duration of subscription or seniority lengthens. These results remain robust even when alternative proxies of habit formation and estimation techniques are employed, suggesting the reliability and consistency of our findings.

The rest of the paper is organised as follows. Section 2 reviews the literature on electricity consumption habit and its determinants. In Section 3 the empirical strategy and the datasets used in the paper are described. Section 4 presents and provides discussion on the results. Lastly, in Section 5 we conclude and provide the policy recommendations.

2. Related literature

This section is structured in three sub-sections. We start by defining the habit formation and its implication for energy consumption. Next, we present the evolution of habits formation, and then the determinants of energy consumption.

2.1. Habits formation and energy consumption

Consumption habits are long-term automatic pattern of consumption behaviour [16]. Habits are unconscious and provide a certain comfort to consumers. However, habits might be good or bad; and thus, might result in energy efficiency or energy inefficiency accordingly. The literature highlights that habit might account for up to 80 percent of households’ electricity consumption [22] and might be the main factor explaining the increasing residential electricity consumption [15]. In this paper, we examine how persistent is energy consumption among new subscribers.

Household's electricity consumption habits are driven by evolving standards of living [23] that has induced rapid proliferation of electronic devices and appliances [11]. Most of the increase in residential electricity consumption is due to standby and active appliances including air conditioner, television, computer, mobile phones, electric hobs, and showers [ [10,11]], as opposed to freezers or fridges that do not contribute significantly to the increase in residential electricity consumption [10]. Indeed, the daily use of standby or active appliances is part of habit, as switching these appliances off is not complex, requires little involvement, but presents a high degree of constraint for users [16]. Thus, the action of turning off standby appliances might be neglected by consumers, and this might generate inefficiencies. To capture this efficiency, we control for appliances used by households in our empirical analysis.

Energy inefficiency can be due to non-optimal use of appliances or the use of inefficient appliances [12]. The literature highlights the existence of transient and persistent inefficiencies in household electricity consumption which, if resolved, result in reductions in residential electricity consumption. Persistent or structural inefficiencies refer to systematic inefficiencies that are related to the type of appliances used and to the bad electricity consumption behaviours, while transient inefficiencies refer to inefficiencies due to the adaptation of the subscribers to climatical, environmental and economic issues that are time-varying. In the US, for instance, residential energy consumption could be reduced by 10 percent and 17 percent respectively if persistent and transient inefficiencies are reduced [17]. Similarly, Blasch et al. [24] find that eliminating transient and persistent inefficiencies in Swiss households’ energy consumption could lead to energy savings of a 11 percent and 22 percent respectively. Persistent inefficiencies can be mitigated by the systematisation of the use of energy saving appliances [ [[25], [26], [27]]]. Indeed, using more efficient appliances can save up to 13 percent of energy [28] and helps reduce peaks electricity consumption [29]. But the “long pay-back period” for efficient appliance could discourage people from investing in more energy-efficient appliances [30]. The usage and efficiency of appliances will not be captured in this analysis due to data limitation.

2.2. Dynamics of energy consumption habits

Electricity consumption habits might evolve due (i) to unexpected changes in households' living environment or (ii) policy interventions designed to reduce residential electricity consumption. As far as changes in household's environment are concerned, it has been shown that electricity consumption pattern may change due to exogenous chocs, such as the Covid-19 pandemic which exacerbated peaks and troughs in residential electricity consumption [4]. More generally, a change at individual level – such as relocation or retirement – can modify electricity consumption pattern [16]. Data limitation does not allow us to capture this change in the conditions of the consumers.

Pertaining to policy interventions for reducing households' electricity consumption, several approaches can be used. The more documented approach are the time-of-use scheme and the real time pricing; even if the real time pricing is practically less used due to the high cost of its monitoring [31]. Recent literature on the effectiveness of the time-of-use pricing highlights that this scheme can mitigate the effect of shocks that result in an increase in residential electricity consumption [4]. As for real time pricing, it is shown that electricity consumption can reduce up to 3 percent in hours with high prices [32]. Electricity price in Cote d’Ivoire can vary with the type of subscription (i.e., moderate versus normal), hence this study will not capture real time pricing. However, we do control for electricity price.

Other approaches for encouraging electricity savings include billing strategy. Households pay attention to information of the energy bill they received. When the bill is more informative, it results in up to 10 percent energy saving [18]. This is consistent with the inattention behaviour in electrical appliances purchasing or use, i.e., even if the households are well informed of the effects of the appliances they use on the energy bill, recalling this information on their bills on a regular base or providing feedback might change their consumption pattern [ [33,34]]. Furthermore, it has been shown that any methods aimed at raising consumers’ awareness of their consumption patterns could be effective in encouraging electricity savings. These may include (i) information and communications technology tools [5], such as data-driven user recommender systems [23], (ii) advanced metering infrastructures [35], and (iii) community- or individual-level enhancement of environmental and climate-change awareness [36].

2.3. Determinants of energy consumption

Energy consumption habits is formed by several individual actions [23]. These individual actions are affected by external factors related to consumers' socioeconomics conditions, climatic conditions, and energy market conditions. Pertaining to the consumers' socioeconomics conditions, factors such as income, education, household's size, and structure are mainly analysed. Most of the literature suggests a positive association between household income and electricity consumption, regardless of the type of household [ [7,8,11,17,37]]. However, Blasch et al. [24] find that the effect of income on electricity consumption becomes insignificant after controlling for residential and household characteristics, confirming the findings of [38]. In terms of household education level, the literature indicates a general reduction in electricity consumption with higher levels of education [ [8,24]]. Yet, this negative effect is not significant for low electricity users [8]. Regarding household size, multiple studies including [17,24,37], and [8] support a positive effect of household size on electricity consumption. Nevertheless, the effects of the number of elderly members in the household on electricity consumption show some variation. Huang [8] and Kavousian et al. [38] suggest a negative effect, while Blasch et al. [24] find a positive effect, mainly observed among high electricity users.

Climatic condition also plays important role in electricity consumption. Thermal comfort exhibits strong correlation with households’ electricity consumption [39]. It is worth noting that the impact of temperature on residential electricity consumption remains consistent throughout the period of measurement, as highlighted by Kang and Reiner [40], while also being contingent on the specific climate of the study area. This effect is non-linear. Indeed, research conducted by Kang and Reiner [40] suggest that in colder countries, the effect tends to be negative, indicating that electricity consumption increases as temperature drops. Conversely, studies like [41] indicate that in hotter countries, the effect can be positive, implying that electricity usage increases with rising temperatures. This discrepancy can be attributed to the presence of a threshold temperature level. Below or above this threshold, individuals may require the use of cooling or heating appliances to maintain a comfortable indoor temperature, thus contributing to higher electricity consumption. Lastly, it is shown that there is a rural-urban discrepancy in the effect of variation in temperature on electricity consumption [42], with rural electricity consumption being more volatile [43].

The market condition, primarily captured by electricity prices, plays a significant role in residential electricity consumption. There is a consensus among various studies, including [17,24], and [11], regarding the impact of electricity prices on residential electricity consumption. It is widely observed that residential electricity demand exhibits a strongly significant and negative price elasticity. This implies that changes in electricity prices can have a substantial influence on household electricity usage, making price an effective tool for promoting electricity savings. However, it is important to consider that households may react differently to price changes due to the imperfections in the available information. Well-informed households, as highlighted by Allcott and Greenstone [34], tend to be more responsive to fluctuations in energy prices compared to those with limited information. This variation in households’ responsiveness to price changes underscores the role of information dissemination and consumer awareness in shaping their behavioural responses to energy price adjustments. Consistent with this literature, we control for socio-economic factors, climatic and market conditions.

3. Empirical strategies

This section describes the theoretical model, the econometric specification and data used in this paper.

3.1. Theoretical model

The theoretical model is drawn from Alessie and Lusardi [44]. These authors consider a household maximizing the following utility function with habit formation:

$$\begin{array}{c}{\max }_{\left\{c_t\right\}}{E}_t\sum _{\tau =t}^{\infty }{\left(1+\rho \right)}^{t-\tau }{u}_{\tau }\left(c_{\tau }-\gamma c_{\tau -1}\right)\end{array}$$ (1)

subject to an inter-temporal budget constraint:

$$\begin{array}{c}{A}_t=\left(1+r\right)A_{t-1}+y_t-c_t\end{array}$$ (2)

where $c_t$ indicates consumption in period $t$, $\gamma$ is the habit formation parameter, $r$ denotes the real interest rate, $\rho$ is the rate of time preference, $A_t$ is household wealth (e.g., financial asset), $y_t$ is non-capital income. The initial conditions $A_{-1}$ and $c_{-1}$ are given.

Denote $c_t^{*}=c_t-\gamma c_{t-1}$, the problem of the household can be rewritten as follows:

$$\begin{array}{c}{\max }_{\left\{c_t^{*}\right\}}{E}_t\sum _{\tau =t}^{\infty }{\left(1+\rho \right)}^{t-\tau }{u}_{\tau }\left(c_{\tau }^{*}\right)\end{array}$$ (1’)

under the constraint:

$$\begin{array}{c}{c}_t^{*}=\left(1+r\right)A_{t-1}-A_t+y_t-\gamma c_t\end{array}$$ (2’)

Derivations with respect to $c_t^{*}$ and $A_t$ yield

$$\{\begin{array}{c}\begin{array}{c}{u}_t^{\prime }\left(c_t^{*}\right)={\lambda }_t\end{array}\\ \begin{array}{c}{\lambda }_t=\frac{1+r}{1+\rho }{E}_t{\lambda }_{t+1}\end{array}\end{array}$$ (3ab)

where ${\lambda }_t$ is the Lagrange multiplier associated with equation (2’).

Combining (3a) and (3b) yields

$$\begin{array}{c}{u}_t^{\prime }\left(c_t^{*}\right)=\frac{1+r}{1+\rho }{E}_t{u}_t^{\prime }\left(c_{t+1}^{*}\right)\end{array}$$ (4)

This paper assumes that the interest rate is equal to the rate of time preference ($r=\rho$). Therefore, $c_t^{*}$ is constant:

$$\begin{array}{c}{u}_t^{\prime }\left(c_t^{*}\right)=E_t{u}_t^{\prime }\left(c_{t+1}^{*}\right)\Rightarrow c_t^{*}={E_{t}c}_{t+1}^{*}\end{array}$$ (5)

After a few algebraic transformations, equation (2’) can be rewritten as

$$\begin{array}{c}\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{c}_{\tau }^{*}=-\gamma c_{t-1}+\left(1-\frac{\gamma }{1+r}\right)\left(\left(1+r\right)A_{t-1}+E_t\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{y}_{\tau }\right)\end{array}$$ (6)

Given that $c_t^{*}$ is constant – i.e., $c_{\tau }^{*}=c_t^{*}\forall \tau \ge t$ – equation (6) is equivalent to:

$$\begin{array}{c}\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{c}_t^{*}=-\gamma c_{t-1}+\left(1-\frac{\gamma }{1+r}\right)\left(\left(1+r\right)A_{t-1}+E_t\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{y}_{\tau }\right)\end{array}$$ (6’)

The left-hand side of equation (6’) can be rewritten as follows:

$$\begin{array}{c}\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{c}_t^{*}=\frac{1+r}{r}{c}_t^{*}\end{array}$$

Recall that $c_t^{*}=c_t-\gamma c_{t-1}$ and using (6’), we get:

$$\begin{array}{c}\frac{1+r}{r}\left(c_t-\gamma c_{t-1}\right)=-\gamma c_{t-1}+\left(1-\frac{\gamma }{1+r}\right)\left(\left(1+r\right)A_{t-1}+E_t\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{y}_{\tau }\right)\end{array}$$ (7)

Therefore,

$$\begin{array}{c}{c}_t=\gamma \left(1-\frac{r}{1+r}\right)c_{t-1}+\frac{r}{1+r}\left(1-\frac{\gamma }{1+r}\right)\left(\left(1+r\right)A_{t-1}+E_t\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{y}_{\tau }\right)\end{array}$$ (8)

Finally,

$$\begin{array}{c}{c}_t=\frac{\gamma }{1+r}{c}_{t-1}+\left(1-\frac{\gamma }{1+r}\right)Y_{pt}\end{array}$$ (9)

where

$$\begin{array}{c}{Y}_{pt}=\frac{r}{1+r}\left(\left(1+r\right)A_{t-1}+E_t\sum _{\tau =t}^{\infty }{\left(1+r\right)}^{t-\tau }{y}_{\tau }\right)\end{array}$$ (10)

We can rewrite Equation (10) as follows:

$$\begin{array}{c}{Y}_{pt}=\frac{1}{1+r}\left(E_t{Y}_{p,t+1}+r{c}_t\right)\end{array}$$ (11)

Using Equations (9), (11), we show that current consumption is entirely determined by past consumptions as follows:

$$\begin{array}{c}{c}_t-c_{t-1}=\gamma \left(c_{t-1}-c_{t-2}\right)\end{array}$$ (12)

In our dataset, the consumption of each individual at time $t$ is not available, but only the aggregate consumption over a certain period ($\sum _{\tau \le T}{c}_{\tau }$). Therefore, the end of period Equation (12) is defined as:

$$\begin{array}{c}{c}_T=\gamma c_{T-1}+\alpha \end{array}$$ (13)

Where $\alpha =c_1-{\gamma c}_0$.

3.2. Econometric specification and variables

The purpose of the analysis is to estimate the parameter $\gamma$ for energy consumption in Côte d’Ivoire. Our baseline econometric model is derived from Equation (13) above and is written as follows:

$$\begin{array}{c}{c}_{it}=\gamma c_{i,t-1}+{\alpha }_i+{\epsilon }_{it}\end{array}$$ (14)

where $c_{it}$ is the electricity consumption of individual $i$ from his/her subscription until period $t$, ${\alpha }_i$ is individual-specific effects and ${\epsilon }_{it}$ is the error term.

Individual-specific effect is model as ${\alpha }_i={\alpha }_0+X_i\beta$ to account for factors ($X_i$) that affect electricity consumption. These factors are presented in Table 1. Hence, the full model is:

$$\begin{array}{c}{c}_{it}={\alpha }_0+\gamma c_{i,t-1}+X_i\beta +{\epsilon }_{it}\end{array}$$ (15)

Table 1.

Variables’ definition and sources.

Variables	Description	Source
Aggregate consumption	Total electricity consumption per subscriber over her period of access (in kilowatt hour)	CIE's administrative data
Electricity price	Department level average electricity price computed over the subscriber's period of consumption	CIE's administrative data
Subscribed power	Electricity power (in kVA) subscribed	CIE's administrative data
Subscriber type	Dummy variable for whether the subscriber is a normal domestic subscriber or moderate one (who benefit from the government's social plan)	CIE's administrative data
Type of electric meter	Dummy variable for whether the subscriber use a mono phased electric meter or tri phased one	CIE's administrative data
Precipitation	Department level average monthly precipitation (in mm/day) over the subscriber's period of consumption	MERRA-2
Temperature	Department level average monthly temperature (in °C) over the subscriber's period of consumption	MERRA-2
Per capita income	Department level average per capita income (in x1,000,000 XOF)	LSMS 2008
Fridge ownership	Department level proportion of households with at least one Fridge	LSMS 2008
Freezer ownership	Department level proportion of households with at least one Freezer	LSMS 2008
Air conditioner ownership	Department level proportion of households with at least one Air conditioner	LSMS 2008
Fan ownership	Department level proportion of households with at least one Fan	LSMS 2008
Household's size	Department level average household size	LSMS 2008
Number of elderlies	Department level average number of elderlies in the household	LSMS 2008
Number of workers	Department level average number of workers in household	LSMS 2008
Household's head years of education	Department level average household head's year of education	LSMS 2008
Age of household head	Department level average household head's age	LSMS 2008
Female headed household	Department level proportion of female headed household	LSMS 2008
Urbanisation rate	Department level urbanisation rate	LSMS 2008

Notes: This table presents the variables used in this paper, their definitions, and the sources of raw data. CIE stands for Compagnie Ivoirienne d’Electricité, that is the National Electricity Company. MERRA is the Modern-Era Retrospective analysis for Research and Applications database, and LSMS is the Living Standard Measurement Survey.

It is worth noting that the past consumption of subscribers is not directly observed. Therefore, habit formation is proxied by using the behaviour of neighbouring subscribers, consistent with the literature on external habits [45]. In this approach, a “neighbour” for an individual subscriber with $t$ months of seniority is identified as a cohort of subscribers with similar characteristics, such as electric meter type, subscriber type, and subscribed power, who have been subscribers for $t-1$ months. The average consumption of the cohort is then used as a proxy for habit formation. It is crucial to acknowledge that using the average subscriber as a proxy for habit formation may be affected by extreme values in the distribution. To address this concern and ensure the robustness of the results, alternative measures such as the median and percentiles (40 and 60) of the cohort's consumption are also used as proxies for habit formation. This allows for a comprehensive assessment of the effect of habit formation on electricity consumption while mitigating the potential influence of outliers.

The model is estimated using an Ordinary Least Squares (OLS) approach. Additionally, three estimations are performed to analyse various aspects of habit formation.

• (i)To analyse the heterogeneity of the habit formation across climatic zones, an OLS approach is used, allowing for a comparison of the effects of habit formation in different geographic regions with distinct climatic conditions.
• (ii)To explore the heterogeneity of the habit formation among subscribers with different consumption levels, a quantile regression approach is employed. This enables the analysis of habit formation at different points of the consumption distribution, providing insights into how habits may vary across different consumption levels.
• (iii)To investigate the dynamics of the habit formation as subscriber's seniority increases over time, using a rolling regression approach is used. This approach allows for the estimation of habit persistence and changes in habits as subscribers progress through different years of seniority.

By employing these additional estimation techniques, the study aims to gain a deeper understanding of habit formation dynamics, explore potential heterogeneity in habit formation across climatic zones and consumption levels, and capture the evolving nature of habits as subscribers' seniority increases.

3.3. Data

Table 1 present the description of the variables used herein. The dataset used for the analysis includes information on a total of 621,828 subscribers in Côte d’Ivoire with subscription date ranging from January 2006 to December 2011. The total electricity consumed (in kilowatt) and the total electricity bill paid up to December 2011 are reported. The dataset also provides information on the subscriber type, power at the date of the subscription, and type of employment of the subscriber. Since the paper aims at analysing the habit formation of new subscribers in the residential sector, subscribers classified as firms and administrations are excluded from the analysis. As a result, the sample is reduced to a subsample of 380,830 subscribers. Furthermore, subscribers with only one month of seniority (7950 subscribers) are excluded due to the requirement of one-period lag value included in the specified model. After removing missing values, the final sample used for the analysis consists of 372,449 new subscribers.

In addition to the subscriber data, the study also incorporates monthly weather data for each locality in Côte d'Ivoire. Precipitation and temperature at 2 m are obtained from the Modern-Era Retrospective analysis for Research and Applications (MERRA-2) database, provided by the NASA, over the period 2006 to 2011. For each subscriber, we calculate the average precipitation and average temperature over the period during which they have access to electricity. This information allows for the examination of the potential influence of weather conditions on electricity consumption habits. Moreover, electricity price data are obtained from the National Electricity Company (In French, Compagnie Ivoirienne d’Electricité “CIE”). These prices represent the annual average prices per department computed as the total value of sales divided by the total quantity of electricity sold. Using this data, we calculate the average electricity price for each subscriber over the period electricity consumption. This enables the analysis of the relationship between electricity prices and consumption habits.

Other socioeconomic variables include average per capita income, urbanization rate, average age of household's head, average year of education of household's head, proportion of female headed households, number of unemployed per household, proportion of households with fan, air conditioner, refrigerator, and freezer, average household size, average number of elderly (aged 60 or above) per household computed at the departmental level using the 2008 national living standard measurement survey (LSMS) data. By incorporating these socioeconomic variables, the study aims to examine their potential impact on electricity consumption habits and further understand the socioeconomic factors associated with consumption patterns.

4. Results and discussion

4.1. Sample description

The analysis on the overall sample (see Table 2) shows that new subscribers used electricity continuously for an average of 32 months during the observation period (2006–2011), with an average electricity consumption of 181.89 Kwh per month. The average subscribed power is 1.92 kVA with significant heterogeneity. Only 40.67 percent of new subscribers are normal subscribers. This implies that about 60 percent of new subscribers are social subscribers who benefit from subsidised electricity prices if their monthly electricity consumption is not greater than 100 Kwh. Almost all new subscribers (97.94 percent) use single phase electric meter.

Table 2.

Subscriber characteristics and consumption.

Variable	Mean/proportion	Std Dev	Min	Max
Total electricity consumption per subscriber over her period of access (in kilowatt hour)	4689.28	9131.06	0	1,022,410
Monthly electricity consumption per subscriber (in kilowatt hour)	181.89	375.64	0	30,070.88
Number of months as subscriber	32.13	18.28	2	71
Subscribed power (in kVA)	1.92	2.37	1.1	39.6
Subscriber as normal household	0.4067	0.4912	0	1
Subscriber with single phase electric meter	0.9794	0.1420	0	1

Source: CIE's administrative data on subscriber consumption

At the departmental level (see Table 3), the average urbanisation rate is 36.26 percent with considerable variability. The average household size per department varies between 3.37 and 8.38, with a relatively low presence of elderly (less than one) individuals within households. Household heads are relatively young (42.88 years old) and mostly men (only 18.44 percent of households are headed by women) with a low level of formal education (4.07 years of education on average). Regarding the use of household appliances, only 7.83 percent of household have a fridge, 4.14 percent have a freezer, 1.23 percent have an air conditioner, and 31.27 percent have a fan. The annual average per capita income at the departmental level is XOF 1.17 million (€1783.65) with a high degree of variability, and the average number of workers in households per department varies between 1.44 and 4.41.

Table 3.

Equipment use and household characteristics at department level.

Indicator	Mean/proportion	Std Error	Min	Max
Proportion of households with at least one Fridge	7.83	1.27	0.00	62.72
Proportion of households with at least one Freezer	4.14	0.52	0.00	19.25
Proportion of households with at least one Air conditioner	1.23	0.4	0.00	22.23
Proportion of households with at least one Fan	31.27	3.06	1.54	88.49
Average household size	4.89	0.11	3.37	8.38
Average number of elderlies in the household	0.22	0.01	0.07	0.55
Urbanisation rate	36.26	3.62	0.00	100
Average household head's year of education	4.07	0.26	0.64	11
Average household head's age	42.88	0.29	37.78	49.81
Proportion of female headed household	18.44	0.91	1.85	35.12
Average number of workers in household	2.35	0.08	1.44	4.41
Average per capita income (in x1,000,000 XOF)	1.17	0.18	0.12	7.36
Average monthly precipitation over 2006–2011 (in mm/day)	3.67	0.06	2.7	4.85
Average monthly temperature over 2006–2011 (in °C)	26.11	0.07	24.39	27.15
Average annual electricity price over 2006–2011 (in XOF/Kwh)	75.84	0.91	53.49	97.71

Source: Department level statistics produced using the Living Standard Measurement Survey (LSMS 2008), data from MERRA-2 and from the national electricity company

When analysing climate-related variables, we find that there is low variability across departments. The average monthly temperature is 26.11 °C, with variations ranging from 24.39 °C to 27.17 °C among departments. Similarly, the average monthly precipitation is 3.67 mm per day and varies between 2.7 and 4.85 mm per day. The annual average electricity price per department varies between XOF 53.49 (€0.08) and XOF 97.71 (€0.15) per Kwh. Overall, the climate-related variables show relatively consistent temperature and precipitation patterns across departments, while electricity prices exhibit notable variation among different departments.

4.2. Main estimation results

Table 4 displays the estimated parameters of the model specified in Equation (15). The model is estimated following the methodology described in Section 3.2 that relies on an ordinary least square approach. In total, six models are estimated, with each model including different blocks of explanatory variables, in order to check whether our habit parameter is robust to the addition of different explanatory variables. The first model serves as the baseline model without any control variables, while the sixth model represents the full model including all the specified explanatory variables. In all the model, regional fixed effects are included and are found to be significant. This means that regional variations have a significant impact on electricity consumption habits among new subscribers. Furthermore, all six models are globally significant, indicating that the overall model fit is statistically significant. The R-squared values hover around 0.13 across the different models.

Table 4.

Estimates of habit formation.

Log aggregate consumption	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Lag of log aggregate consumption	0.8892*** (0.0533)	0.7489*** (0.0419)	0.736*** (0.0408)	0.7506*** (0.0391)	0.7462*** (0.0448)	0.742*** (0.0439)
Seniority (number of months)	0.0024* (0.0013)	0.0058*** (0.001)	0.0052*** (0.0009)	0.0047*** (0.0013)	0.0043*** (0.0012)	0.0047*** (0.0011)
Subscribed power		0.0283*** (0.0103)	0.0272*** (0.0094)	0.0255*** (0.0089)	0.0246*** (0.008)	0.0254*** (0.008)
Subscriber type (ref = normal domestic subscriber)
Moderate domestic subscriber		−0.0549 (0.0639)	−0.4683*** (0.1264)	−0.4651*** (0.1228)	−0.7812*** (0.217)	−0.8237*** (0.2102)
Type of electric meter (ref = single phased)
Tri phased		−0.0168 (0.0986)	−0.0252 (0.0946)	−0.0219 (0.0924)	0.0074 (0.0888)	0.01 (0.0866)
Electricity price			−0.0206*** (0.0055)	−0.0205*** (0.0051)	−0.0346*** (0.0084)	−0.0367*** (0.0081)
Precipitation				−0.0583 (0.0483)	−0.0428 (0.0502)	−0.0084 (0.0321)
Temperature				−0.009 (0.0769)	0.0015 (0.052)	−0.0249 (0.0507)
Per capita income					0.0055 (0.0315)	0.0091 (0.0256)
Fridge ownership					−0.0275** (0.0118)	−0.0303*** (0.0094)
Freezer ownership					0.0054 (0.0091)	0.0159* (0.0082)
Air conditioner ownership					0.0361 (0.0254)	0.0547** (0.0207)
Fan ownership					0.0097** (0.0044)	0.002 (0.003)
Household's size						0.126* (0.0661)
Number of elderlies per household						−1.2703*** (0.4532)
Number of workers per household						0.0899 (0.0781)
Household's head years of education						−0.0318* (0.0173)
Age of household head						−0.0105 (0.0174)
Female headed household						−0.0046 (0.0051)
Urbanisation rate						0.0037** (0.0018)
Intercept	0.5844 (0.3569)	1.5228*** (0.2609)	3.404*** (0.6619)	3.7762* (2.1674)	4.5124*** (1.3144)	5.7544*** (1.4468)
Dummy for region	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	372,449	372,449	372,449	372,449	372,449	372,449
R2	0.1306	0.1314	0.1324	0.1325	0.1354	0.1364
F-stat	2796.54	2449.99	2367.64	2186.97	1881.07	1548.11
P-value	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001

Cluster (clustered at department level) robust standard errors in parenthesis. *** significance at 1 percent level, ** significance at 5 percent level, * significance at 10 percent level.

The habit formation parameter (denoted by $\gamma$ in Equation (7)) is significant in all models and is closer to 1 (the coefficient is 0.742 in the full model and 0.889 in the baseline model). It means that electricity consumption habit among new subscribers is highly persistent. This result is consistent with previews literature which states that habits, measured by persistent daily routines and baseline patterns, account for up to 80 percent of electricity consumption [22]. Highly persistent habit might be harmful since it can be the main purpose of increase in households' electricity consumption [15] and it is difficult to curb. Further discussions on the formation of electricity consumption habits are provided in Section 4.3 by considering the role of subscribers’ seniority (subsection 4.3.1), climatic conditions (subsection 4.3.2) and the type of consumers according to their level of electricity consumption (subsection 4.3.3).

The characteristics of the subscription have the expected sign and provide valuable insights into their effects on electricity consumption habits among new subscribers. Firstly, a one kVA increase in the subscribed power leads to a 2.5 percent increase in aggregate electricity consumption. This positive relationship suggests that higher subscribed power levels are associated with higher levels of electricity consumption, indicating that subscribers with greater power requirements tend to consume more electricity. Secondly, moderate domestic subscribers consume approximately 56.12 percent1 less electricity than normal domestic subscribers with similar characteristics. This negative relationship indicates that moderate domestic subscribers, who presumably have lower electricity consumption needs or preferences, exhibit significantly lower levels of electricity usage compared to their counterparts. However, the type of electric meter has no significant effect on the aggregate electricity consumption.

Consistent with the literature, electricity consumption has a negative price elasticity (e.g. [17]). A one XOF increased in electricity price results in a 3.67 percent reduction in the aggregate electricity consumption. This negative relationship suggests that higher electricity prices are associated with lower levels of electricity consumption among new subscribers.

However, the analysis reveals that environmental variables, such as average precipitation and average temperature, do not have a significant effect on electricity consumption among new subscribers. This implies that variations in these environmental factors do not significantly influence the overall electricity consumption patterns observed.

The analysis indicates that income, while positively related to electricity consumption, is not found to be statistically significant in explaining consumption among new subscribers. Consistently with Blasch et al. [24], this suggests that income may not have a significant direct effect on electricity consumption after controlling for other income-related household characteristics. In contrast, the ownership of specific household appliances emerges as important determinant of electricity consumption. The estimates highlight that owning a freezer or an air conditioner is positively associated with aggregate electricity consumption, indicating that households with these appliances tend to consume more electricity. However, the ownership of fan does not show a significant association with electricity consumption. Surprisingly, the ownership of fridge is negatively associated with aggregate electricity consumption. This result aligns with previous findings suggesting that the presence of active appliances may not necessarily lead to an increase in electricity consumption [10].

The analysis reveals important associations between household characteristics and electricity consumption habits among new subscribers. An increase in household size is positively associated with electricity consumption. This suggests that larger households tend to consume more electricity compared to smaller households. While the number of workers in the household shows a positive association with electricity consumption, this relationship is not statistically significant. This suggests that the presence of more workers in the household may contribute to higher electricity consumption, but the effect is not robust enough to be statistically significant. The estimates indicate a negative association between the number of elderlies in the household and electricity consumption. This result is also corroborated with the negative effect of household's head age on electricity consumption, even if this effect is not significant. These results implies that older household heads may exhibit lower electricity consumption habits, but further investigation is needed to confirm the significance of this relationship. The estimates also show that more educated household heads have lower electricity consumption. This suggests that higher levels of education may be associated with greater energy efficiency or awareness of energy-saving practices. In addition, although not statistically significant, the estimates show a negative association between female-headed households and electricity consumption. This suggests that, on average, female-headed households may exhibit lower electricity consumption, although further analysis is required to establish the significance of this relationship.

The urban-rural discrepancy in electricity consumption is found to be significant. The estimates show that a one percentage point increase in urbanization rate leads to 0.37 percent increase in electricity consumption. While the magnitude of this effect may be lower compared to some findings in the literature which can reach up to 23 percent [ [7,8]], this result highlights the presence of inefficiencies and potential poor consumption habits in urban areas. The result suggests that urban areas may exhibit higher electricity consumption and potentially inefficient energy usage patterns compared to rural areas. This finding underscores the need to address and promote energy efficiency measures and better consumption habits in urban areas to mitigate these inefficiencies and promote sustainable electricity consumption practices.

4.3. Robustness check and heterogeneity analysis

To ascertain the robustness of our results, we perform additional regression by using alternative external habits proxies. Specifically, the median, 40^th percentile and 60^th percentile of the cohort's consumption are used as alternative measures of habit formation. The results are reported in Appendix A1, Table 5. The findings from these alternative specifications confirm the presence of highly persistent habit formation in electricity consumption among new subscribers in Côte d'Ivoire. This consistency across different proxies strengthens the evidence for the persistence of consumption habits and underscores the robustness of the study's main findings.

We also analyse the heterogeneity of the electricity habit formation through (i) the analysis of its dynamics over time, (ii) the analysis of its variations across climatic zones, and (iii) the analysis of its variations by type of subscriber measured by electricity consumption level.

4.3.1. Dynamics over time

We examine the dynamics of habit formation as subscribers' seniority increases. This is done through a recursive rolling regression, with a fixed starting date and a 12-month as window. The sample is split based on subscribers’ seniority and the habit formation coefficient is analysed as seniority increases. Fig. 1 shows the dynamics of the habit formation coefficient as seniority increases. In addition to the habit formation coefficient, the effects of the other controls are displayed for six selected models, for subscribers with zero to one year of seniority, zero to two years of seniority, and so one. Results of these estimates are presented Table 6 in appendix A2. The estimates show that the persistence increases with the seniority of subscribers. The dynamics observed in Fig. 1 show that the persistence in habit formation grows until 40 months of seniority and remains stable around 0.75 thereafter.

Fig. 1.

Dynamics of the habit formation parameter Note: Lower bound and upper bound of the confidence interval are obtained as $\hat{\gamma }\pm 2*SE$, where $SE$ is the clustered robust standard error of the estimate $\hat{\gamma }$.

The estimates reveal an interesting finding regarding the relationship between price elasticity and seniority in electricity consumption among new subscribers. The estimates show that the price elasticity decreases as the seniority of subscribers increases. Specifically, the effect of a one XOF increase in electricity price on consumption reduction is found to be higher for subscribers with at most one year of seniority, resulting in a 7.17 percent reduction in electricity consumption. However, the effect reduces to 3.67 percent when seniority reaches six years. This result implies that the effectiveness of a tariffication policy, aimed at influencing consumption behaviour through price signal, diminished as subscribers gain more seniority. In other words, the impact of price changes on consumption reduction becomes less pronounced as subscribers become more accustomed to their consumption patterns over time.

Furthermore, we find that the urban-rural discrepancy only becomes significant after two years of seniority. This finding suggests that consumption patterns and the urban-rural disparity in electricity usage may take some time to develop and become apparent among new subscribers.

The results of this additional analysis provide insights into the dynamics and variations of habit formation in electricity consumption among new subscribers. Understanding these patterns and variations contributes to a deeper understanding of how habits develop and evolve over time. Moreover, the findings highlight the importance of considering the effects of seniority and the timing of policy interventions when designing strategies to promote energy efficiency and address urban-rural discrepancies in electricity consumption habits.

4.3.2. Variations across climatic zones

The study employs a full heterogeneous model approach to analyse the variations in electricity consumption across different climatic zones in Côte d'Ivoire. The analysis considers three climatic zones: Centre and North regions (hottest regions), West regions (very humid regions), and South regions (coldest regions). The estimates, presented in Table 7 in Appendix A3, reveal significant differences in electricity consumption habits among these climatic zones. We find that consumption habits among new subscribers in the South regions tend to be more persistent compared to the other regions: the habit parameter is 0.75 in South regions compared to 0.58 in the west regions. We also find that the price elasticity is higher in West regions (−5.02 percent) than in other regions, with the lowest price elasticity observed in the South regions (−3.88 percent). This latter result aligns with the findings on persistence, as regions with lower persistence in electricity consumption habits tend to have higher price elasticity. The effect of temperature on electricity consumption shows significant variation across climatic zones. In the West regions (the most humid regions), the magnitude of the effect of temperature is more than twice as high (+18.6 percent) compared to the Centre and North regions (the hottest regions), where the effect of temperature is not significant. This suggests that transient inefficiency, potentially related to the need for cooling in humid regions, is a major driver of the impact of temperature on electricity consumption habits among new subscribers.

These findings highlight the importance of considering climatic variations in analysing electricity consumption habits. The study demonstrates significant differences in persistence, price elasticity, and the effect of temperature across different climatic zones, indicating the influence of local climate conditions on consumption patterns. Overall, this analysis enhances our understanding of how climatic variations contribute to heterogeneity in electricity consumption habits and underscores the need for tailored policies and interventions to promote energy efficiency and address specific challenges in different climatic zones.

4.3.3. Variations of consumption habits

The analysis of the variations of consumption habits with respect to the level of consumption of the subscriber utilises quantile regression technique. To effectively account for the regional fixed effects included in the model, the quantile regression is estimated using a method of moment with multiple fixed effects [ [46,47]]. Three quantiles are considered: the 20^th percentile (low electricity utilizers); the 50^th percentile (normal or medium electricity utilizers), and the 80^th percentile (high electricity utilizers). The estimates, presented in Table 8 in Appendix A4, show that the persistence of electricity consumption habit declines slightly from low utilizers (+0.81) to high utilizers (+0.63), suggesting that regardless of whether subscribers have low or high levels of electricity usage, consumption habits tend to persist over time. In addition, the persistence of electricity consumption habit for low utilizers is around 30 percent as high compared to high utilizers. These findings imply that even if the persistence of electricity consumption habit is very high in general, it is moderated among high utilizers, indicating that low consumption habits are more persistent that high consumption habit.

Moreover, as expected, the estimates show that electricity consumption price elasticity for low utilizers is more than 3 times higher (−0.0489) than that of high utilizers (−0.0178). This implies that price changes have a greater impact on reducing electricity consumption among low utilizers compared to high utilizers. We also find that the urban-rural is not significant for low electricity utilizers, but significant for normal and high utilizers. This result implies that low utilizers exhibit similar consumption patterns, regardless of their residence areas. Finally, our estimates of the effects of education on electricity consumption align with the literature with education having no significant effect on electricity consumption for low utilizers [8] but a negative effect on electricity consumption for other categories of utilizers [ [8,24]].

5. Conclusion and policy implications

In this study, we aimed to investigate the formation of electricity consumption habits among new subscribers in Côte d'Ivoire. To achieve this, we combined datasets from the National Electricity Company, the Living Standard Measurement Survey, and the NASA's MERRA-2 database. Our analysis specifically focused on new subscribers with a maximum of 6 years of seniority and examined the presence of external habits in their electricity consumption patterns.

The key findings of our study reveal highly persistent habits in electricity consumption among new subscribers, with the persistence of these habits increasing over time. We also identified heterogeneity in habit formation across different climatic zones and according to the consumption level of the subscriber. Furthermore, our results demonstrated robustness when using alternative proxies of external habits, supporting the reliability of our findings.

These findings have important implications for energy efficiency policies and interventions. The results highlight the importance of targeting and addressing electricity consumption habits at an early stage of subscription since habit formation becomes more stable after approximately 40 months of seniority. This emphasizes the need to develop policies that promote energy efficiency among new customers to reduce energy waste. One potential approach to encourage energy efficiency among new subscribers is to provide incentives for investing in efficient appliances. By incentivizing the adoption of energy-saving technologies, we can contribute to the development of more sustainable consumption habits. Additionally, offering coaching or sensitization activities designed for new subscribers can help them monitor and manage their electricity consumption effectively. This strategy can leverage on education, as the paper finds that energy consumption decreases with the level of education.

Price seems also to be a powerful tool to promote energy savings, given its negative effect on energy consumption. Based on this result, the study suggests that a tariffication policy, designed to promote energy efficiency and consumption reduction, should be primarily targeted at new subscribers. By focusing on new subscribers and implementing effective pricing mechanisms during the early stages of their subscription, policymakers can better enforce energy efficiency practices and shape consumption habits at the household level. Nevertheless, one should also bear in mind that one size does not fit all, as the increase in price has a greater effect on low energy consumers.

Overall, our study provides valuable insights into the formation of electricity consumption habits among new subscribers in Côte d'Ivoire. The findings emphasize the importance of early intervention, policy development, and consumer education to promote energy efficiency and reduce energy waste at the household level.

The main limitation of this paper lies in the data. We do not have access to more recent data to extend the time span of the study. The paper could also be extended if we could benefit from more detailed data that might allow an in-depth analysis of the role of each type of electric appliance used and disentangle the bad habits from the good.

Data availability statement

The authors have no permission to share data.

Funding

This research received no grant from any funding agency.

Disclaimer

The views expressed in this paper are those of the authors and do not represent the views of the International Monetary Fund (IMF), its Executive Board, or its Management. All disclaimers apply.

CRediT authorship contribution statement

Richard K. Moussa: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Data curation, Conceptualization. Désiré K. Kanga: Writing – review & editing, Writing – original draft, Methodology, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendices.

Appendix A1. Estimates with alternative habit formation proxies

Table 5.

Robustness check (full model with median, p40, and p60 instead of mean).

Log aggregate consumption	Model 1: lag consumption at median	Model 2: lag consumption at p40	Model 3: lag consumption at p60
Lag of log aggregate consumption	0.7585*** (0.0443)	0.685*** (0.0442)	0.7976*** (0.0441)
Seniority (number of months)	0.003*** (0.0011)	0.0052*** (0.0011)	0.0018* (0.0011)
Subscribed power	0.03*** (0.0067)	0.0342*** (0.0075)	0.0276*** (0.0064)
Subscriber type (ref = normal domestic subscriber)
Moderate domestic subscriber	−0.7132*** (0.2139)	−0.8216*** (0.2174)	−0.6288*** (0.2145)
Type of electric meter (ref = single phased)
Tri phased	−0.1463** (0.0726)	−0.0496 (0.0805)	−0.191*** (0.0705)
Electricity price	−0.0355*** (0.0082)	−0.0367*** (0.0084)	−0.0342*** (0.0081)
Precipitation	0.0289 (0.0346)	0.0333 (0.0351)	0.028 (0.0342)
Temperature	0.0214 (0.0526)	0.026 (0.0514)	0.0178 (0.0513)
Per capita income	0.0055 (0.0248)	0.0071 (0.0254)	0.005 (0.0247)
Fridge ownership	−0.0309*** (0.0091)	−0.0316*** (0.0093)	−0.0309*** (0.0091)
Freezer ownership	0.0168** (0.008)	0.0172** (0.0082)	0.0167** (0.008)
Air conditioner ownership	0.0563*** (0.0204)	0.0571*** (0.0206)	0.0563*** (0.0203)
Fan ownership	0.002 (0.0029)	0.0021 (0.003)	0.002 (0.0029)
Household's size	0.1373* (0.0686)	0.1377* (0.0695)	0.1342* (0.0681)
Number of elderlies per household	−1.3125*** (0.4588)	−1.3329*** (0.4661)	−1.3231*** (0.4589)
Number of workers per household	0.0774 (0.0788)	0.0793 (0.0795)	0.0811 (0.0786)
Household's head years of education	−0.0306* (0.0168)	−0.0306* (0.0171)	−0.0311* (0.0167)
Age of household head	−0.0099 (0.017)	−0.0091 (0.0173)	−0.0095 (0.0169)
Female headed household	−0.0045 (0.0051)	−0.0043 (0.0051)	−0.0044 (0.005)
Urbanisation rate	0.0036** (0.0017)	0.0036** (0.0017)	0.0036** (0.0017)
Intercept	3.8601** (1.4722)	4.5199*** (1.4622)	3.3573** (1.48)
Dummy for region	Yes	Yes	Yes
Obs.	372,449	372,449	372,449
R2	0.1366	0.1358	0.1368
F-stat	1588.93	1572.21	1594.76
P-value	<0.001	<0.001	<0.001

Cluster (clustered at department level) robust standard errors in parenthesis. *** significance at 1 percent level, ** significance at 5 percent level, * significance at 10 percent level.

Appendix A2. Rolling regression estimates

Table 6.

Dynamics in persistence with increase in seniority.

Log aggregate consumption	Model 1 : 0–1 year of seniority	Model 2 : 0–2 years of seniority	Model 3 : 0–3 years of seniority	Model 4 : 0–4 years of seniority	Model 5: 0–5 years of seniority	Model 6: 0–6 years of seniority
Lag of log aggregate consumption	0.6005*** (0.0361)	0.6423*** (0.0595)	0.7336*** (0.0554)	0.7733*** (0.0621)	0.7514*** (0.0551)	0.742*** (0.0439)
Seniority (number of months)	0.0163*** (0.0038)	0.0167*** (0.0035)	0.0083*** (0.0022)	0.0045* (0.0026)	0.0053*** (0.0019)	0.0047*** (0.0011)
Subscribed power	0.0563*** (0.0103)	0.0423*** (0.0094)	0.0295*** (0.01)	0.024** (0.0105)	0.0261*** (0.009)	0.0254*** (0.008)
Subscriber type (ref = normal domestic subscriber)
Moderate domestic subscriber	−1.8145*** (0.6692)	−1.3561*** (0.4588)	−1.0679*** (0.347)	−0.8723*** (0.3206)	−0.8429*** (0.2593)	−0.8237*** (0.2102)
Type of electric meter (ref = single phased)
Tri phased	−0.0506 (0.1665)	−0.0363 (0.0939)	−0.0205 (0.1023)	0.0065 (0.0988)	0.009 (0.0878)	0.01 (0.0866)
Electricity price	−0.0717*** (0.0252)	−0.0535*** (0.018)	−0.0447*** (0.0136)	−0.0388*** (0.0126)	−0.0375*** (0.0101)	−0.0367*** (0.0081)
Precipitation	−0.0415 (0.0342)	−0.0467 (0.031)	−0.0428 (0.0308)	−0.0195 (0.0347)	−0.0182 (0.0312)	−0.0084 (0.0321)
Temperature	0.051 (0.0497)	−0.0443 (0.0404)	−0.0536 (0.0427)	−0.0409 (0.0515)	−0.0466 (0.0452)	−0.0249 (0.0507)
Per capita income	−0.0102 (0.0359)	0.0048 (0.0269)	0.0111 (0.0255)	0.0211 (0.0262)	0.0189 (0.0272)	0.0091 (0.0256)
Fridge ownership	−0.0081 (0.0151)	−0.0197* (0.0116)	−0.0272*** (0.0092)	−0.0304*** (0.0094)	−0.0312*** (0.0096)	−0.0303*** (0.0094)
Freezer ownership	−0.0096 (0.0135)	0.016 (0.0115)	0.0225** (0.0103)	0.0228** (0.0093)	0.0196** (0.0089)	0.0159* (0.0082)
Air conditioner ownership	0.0192 (0.0282)	0.0252 (0.0227)	0.0386** (0.0184)	0.0489** (0.0198)	0.0516** (0.0205)	0.0547** (0.0207)
Fan ownership	0.0013 (0.0039)	−0.0001 (0.0033)	−0.0002 (0.0028)	0.0011 (0.0029)	0.0015 (0.003)	0.002 (0.003)
Household's size	0.0274 (0.0913)	0.061 (0.0727)	0.0722 (0.0638)	0.0697 (0.0635)	0.0833 (0.064)	0.126* (0.0661)
Number of elderlies per household	−0.4824 (0.4959)	−0.5459 (0.4567)	−1.0003** (0.4024)	−1.208*** (0.4384)	−1.2588*** (0.4576)	−1.2703*** (0.4532)
Number of workers per household	0.1324 (0.0928)	0.1047 (0.088)	0.1006 (0.0777)	0.1097 (0.0811)	0.1243 (0.0816)	0.0899 (0.0781)
Household's head years of education	−0.0244 (0.0318)	−0.0059 (0.0295)	−0.0091 (0.0213)	−0.0271 (0.0186)	−0.0292 (0.018)	−0.0318* (0.0173)
Age of household head	−0.0026 (0.0248)	−0.0188 (0.0204)	−0.0058 (0.0185)	−0.0078 (0.0194)	−0.0072 (0.0188)	−0.0105 (0.0174)
Female headed household	−0.0249*** (0.0077)	−0.0156** (0.0063)	−0.0121** (0.0053)	−0.0102** (0.0051)	−0.0078 (0.0051)	−0.0046 (0.0051)
Urbanisation rate	0.0032 (0.0022)	0.003 (0.0022)	0.0041** (0.0018)	0.004** (0.0018)	0.0042** (0.0019)	0.0037** (0.0018)
Intercept	8.3141** (3.2687)	9.1483*** (1.8387)	7.3867*** (1.8866)	6.3425*** (2.2282)	6.3653*** (1.6323)	5.7544*** (1.4468)
Dummy for region	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	58,908	141,841	221,881	300,702	337,461	372,449
R2	0.0951	0.1148	0.1175	0.1184	0.1311	0.1364
F-stat	162.86	484.03	777.19	1062.73	1339.77	1548.11
P-value	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001

Cluster (clustered at department level) robust standard errors in parenthesis. *** significance at 1 percent level, ** significance at 5 percent level, * significance at 10 percent level.

Appendix A3. Spatial heterogeneity analysis

Table 7.

Estimates of models by geographic zones.

Log aggregate consumption	Model 1: Centre and North regions	Model 2: West regions	Model 3: South regions
Lag of log aggregate consumption	0.5976*** (0.0824)	0.5794*** (0.0437)	0.7475*** (0.0555)
Seniority (number of months)	0.0062** (0.0028)	0.0038 (0.0023)	0.0044*** (0.0012)
Subscribed power	0.0261 (0.017)	0.0325** (0.0153)	0.0238*** (0.0084)
Subscriber type (ref = normal domestic subscriber)
Moderate domestic subscriber	−0.9362*** (0.1617)	−0.8736*** (0.1484)	−0.916*** (0.2176)
Type of electric meter (ref = single phased)
Tri phased	−0.2257 (0.1785)	−0.3161** (0.1296)	0.0545 (0.0714)
Electricity price	−0.0448*** (0.007)	−0.0502*** (0.0116)	−0.0388*** (0.0079)
Precipitation	0.1321*** (0.0352)	0.081 (0.064)	0.0057 (0.0297)
Temperature	0.0783 (0.1023)	0.186* (0.1027)	0.061 (0.0618)
Per capita income	0.0234 (0.0283)	−0.7361*** (0.1118)	0.0625* (0.0312)
Fridge ownership	−0.2343*** (0.0456)	−0.0245 (0.0347)	−0.0432*** (0.0087)
Freezer ownership	−0.1031*** (0.0212)	−0.04 (0.0244)	0.0396*** (0.0108)
Air conditioner ownership	0.4108*** (0.0875)	1.9059*** (0.2905)	0.0688*** (0.0181)
Fan ownership	0.0338*** (0.0076)	0.092*** (0.0107)	0.0031 (0.0031)
Household's size	−0.2091*** (0.0611)	0.4735*** (0.0997)	0.2627** (0.1124)
Number of elderlies per household	3.2113*** (0.7662)	−7.7406*** (1.8949)	−0.4595 (0.4367)
Number of workers per household	0.1867*** (0.0493)	−0.8964*** (0.1143)	−0.1743 (0.1037)
Household's head years of education	0.272*** (0.0599)	−0.1445** (0.0519)	−0.0878*** (0.0281)
Age of household head	−0.0328 (0.0212)	0.368*** (0.0437)	−0.0507* (0.0264)
Female headed household	−0.0602*** (0.0081)	−0.1182*** (0.0165)	0.0127 (0.01)
Urbanisation rate	0.0093* (0.0047)	−0.0375*** (0.0088)	0.0022 (0.0015)
Intercept	6.0024** (2.7441)	−9.6723*** (3.2632)	5.0412** (1.8391)
Dummy for region	Yes	Yes	Yes
Obs.	44.000	54,341	274,108
R2	0.1127	0.1159	0.1422
F-stat	206.86	284.86	1892.65
P-value	<0.001	<0.001	<0.001

Cluster (clustered at department level) robust standard errors in parenthesis. *** significance at 1 percent level, ** significance at 5 percent level, * significance at 10 percent level.

Appendix A4. Quantile regression estimates

Table 8.

Estimates of consumption habits by consumption level using quantile regression.

Log aggregate consumption	Location	scale	20th percentile	50th percentile	80th percentile
Lag of log aggregate consumption	0.742*** (0.0435)	−0.1101*** (0.0323)	0.8144*** (0.0623)	0.7051*** (0.0342)	0.6307*** (0.0251)
Seniority (number of months)	0.0047*** (0.0011)	0.0041*** (0.0008)	0.002 (0.0015)	0.0061*** (0.0009)	0.0089*** (0.0008)
Subscribed power	0.0254*** (0.0079)	0.0048 (0.0049)	0.0222** (0.0108)	0.027*** (0.0066)	0.0302*** (0.0045)
Subscriber type (ref = normal domestic subscriber)
Moderate domestic subscriber	−0.8237*** (0.2085)	0.1788 (0.1723)	−0.9413*** (0.3161)	−0.7638*** (0.1542)	−0.643*** (0.0775)
Type of electric meter (ref = single phased)
Tri phased	0.01 (0.0859)	0.3071*** (0.0474)	−0.1919 (0.117)	0.1129 (0.0753)	0.3205*** (0.0427)
Electricity price	−0.0367*** (0.0081)	0.0186*** (0.0064)	−0.0489*** (0.0118)	−0.0304*** (0.0061)	−0.0178*** (0.004)
Precipitation	−0.0084 (0.0319)	−0.0384* (0.022)	0.0169 (0.0422)	−0.0213 (0.0278)	−0.0472* (0.0252)
Temperature	−0.0249 (0.0502)	0.0811** (0.0412)	−0.0783 (0.0733)	0.0022 (0.0413)	0.0571* (0.0332)
Per capita income	0.0091 (0.0254)	0.027** (0.0122)	−0.0087 (0.0289)	0.0182 (0.0244)	0.0364 (0.0246)
Fridge ownership	−0.0303*** (0.0093)	0.0026 (0.0052)	−0.032*** (0.0115)	−0.0294*** (0.0085)	−0.0276*** (0.0078)
Freezer ownership	0.0159** (0.0081)	0.0052 (0.0067)	0.0125 (0.0105)	0.0177** (0.0076)	0.0212** (0.0086)
Air conditioner ownership	0.0547*** (0.0206)	−0.0039 (0.0107)	0.0572** (0.026)	0.0533*** (0.0182)	0.0507*** (0.0151)
Fan ownership	0.002 (0.003)	−0.0011 (0.002)	0.0027 (0.0037)	0.0017 (0.0028)	0.001 (0.0028)
Household's size	0.126* (0.0656)	−0.0852** (0.0373)	0.1821** (0.0821)	0.0975 (0.0594)	0.0399 (0.054)
Number of elderlies per household	−1.2703*** (0.4496)	0.0181 (0.1806)	−1.2822** (0.5097)	−1.2642*** (0.4285)	−1.252*** (0.4105)
Number of workers per household	0.0899 (0.0775)	0.0881 (0.0564)	0.0319 (0.1059)	0.1194* (0.0666)	0.179*** (0.0584)
Household's head years of education	−0.0318* (0.0172)	−0.0059 (0.0134)	−0.0279 (0.0228)	−0.0338** (0.0155)	−0.0377** (0.016)
Age of household head	−0.0105 (0.0172)	0.0046 (0.0124)	−0.0135 (0.0211)	−0.0089 (0.0165)	−0.0058 (0.0181)
Female headed household	−0.0046 (0.0051)	0.0058 (0.0041)	−0.0084 (0.007)	−0.0027 (0.0044)	0.0012 (0.0043)
Urbanisation rate	0.0037** (0.0018)	0.0004 (0.0012)	0.0034 (0.0023)	0.0038** (0.0016)	0.0041*** (0.0015)
Intercept	5.7275*** (1.4147)	−1.9325 (1.1734)	6.9985*** (1.9599)	5.0802*** (1.2421)	3.7735*** (1.2367)
Dummy for region	Yes	Yes	Yes	Yes	Yes

Cluster (clustered at department level) robust standard errors in parenthesis. *** significance at 1 percent level, ** significance at 5 percent level, * significance at 10 percent level.

Appendix A5. Usage of electric appliances and link with habits and inefficiencies

Fig. A5.1.

Path from electric appliances to energy inefficiencies.