Identifying Energy-Poor Households with Publicly Available Information: Promising Practices and Lessons Learned from the Athens Urban Area, Greece

Abstract

Energy poverty (EP) has emerged as a significant and multifaceted social challenge, both in Europe and globally, in recent years. Given the constraints of limited public resources, it is imperative that policies addressing EP focus on the most socially vulnerable households. To this end, effective policies necessitate a thorough understanding of the key characteristics defining households affected by this phenomenon. This research aims to contribute to the identification of energy-poor households by examining both established and novel EP indicators, utilizing the Athens urban area as a case study. The study employs the official EP indicator utilized in Greece, alongside three newly developed indicators. These indicators constitute essential components in the formulation of multivariate binary logistic regression models. The models incorporate predictors derived from existing government databases, encompassing household and building characteristics. From a policy perspective, this research introduces novel approaches for addressing EP, as the models and two of the newly introduced indicators demonstrate promising results in identifying energy-poor households. Importantly, this identification can be achieved without the need for primary questionnaire surveys or the compilation of complex information.

1. Introduction

Energy poverty, i.e., the inability to attain a socially and materially necessitated level of energy services [1], has emerged in recent years as one of the biggest, multidimensional social problems in Europe and worldwide [2,3]. In this context, energy poverty is related to the energy performance of dwellings, the household composition, the age and health status of its members, social conditions (single-parent families, the existence of unemployed and retired people, etc.), energy prices, a household’s income, climatic conditions, etc. It should be noted that the phenomenon of energy poverty is rather complex, and all the different aspects that are related to it should be incorporated into the overall context.

The implementation of well-designed energy efficiency measures in buildings, and particularly renovation programs, can help to reduce energy poverty and improve living conditions with significant benefits for health and well-being [4,5,6]. As households suffering most from energy poverty experience more barriers to building retrofits and the available public resources are limited, it is of paramount importance that the implemented policies for tackling energy poverty target the most socially vulnerable households [7,8]. In this direction, the mapping and analysis of the main characteristics of the households that are affected by the energy poverty phenomenon are fundamental for the design of effective policies and measures to combat it.

Within the framework of Article 8 of Directive 2023/1791, a prescribed level of end-use energy savings solely based on the proportional representation of energy poverty (EP) among vulnerable individuals, low-income households, and those residing in social housing has been established. Nevertheless, the imperative lies not only in defining the required energy savings but, more crucially, in accurately identifying energy-poor and vulnerable households with a bottom-up approach. The implemented energy efficiency measures within the context of Article 8 should be checked in regards to the actual involvement of the targeted end-users establishing a specialized monitoring, control, and verification mechanism. This ensures that policy measures aimed at achieving energy-saving targets do not inadvertently harm these vulnerable groups, fostering a just and inclusive energy transition.

The above-mentioned conclusion is confirmed also by the learning guide prepared by the Energy Poverty Advisory Hub for the diagnosis of EP. More specifically, the need to obtain a full overview of EP is highlighted, along with the measurement of the impacts at the local level with a combination of qualitative and quantitative indicators capturing different aspects of the phenomenon. Nevertheless, no further guidance is provided, and it is noted that the definition of an EP indicator, which is representative, measurable, and accurate, constitutes a real challenge.

Building upon the described context, this research seeks to contribute to the identification of energy-poor households by analyzing the existing and new EP indicators, using the Athens urban area as a case study. Specifically, the consensual indicators outlined in Directive 2023/1791, the official EP indicator utilized in Greece and three new EP indicators developed in the context of this study, serve as essential components in formulating multivariate binary logistic regression models. The latter use as explanatory variables the characteristics of households and buildings that are already included in the existing databases used by governmental or local authorities. In other words, these models aim to explore the feasibility of predicting the likelihood of a household experiencing EP based on the readily available data from the relevant governmental authorities.

The rest of the paper is structured as follows: Section 2 reviews previous studies on energy poverty in Greece. Section 3 describes in brief the main characteristics of the study area. Section 4 introduces the methodology and data used to conduct the survey. Section 5 presents the main findings and, finally, Section 6 discusses the main conclusions drawn from the survey and the respective policy recommendations.

2. Previous Studies on Energy Poverty in Greece

Energy poverty is a serious issue in Greece, which is ranked among the countries facing the most severe challenges [9]. The economic crisis that has hit the country for nearly 13 years has resulted in Greece being the fifth European country at risk of poverty or social exclusion, with 26.3% of the population (equivalent to 2,722,000 individuals) being affected [10]. Over the last decade, fuel and energy prices have experienced significant increases, households’ incomes have considerably dropped, while also electricity debts having greatly multiplied. The escalation of energy costs aligns with broader global transformations, including, among others, the COVID-19 pandemic, climate crisis, and refugee crisis. Especially in Greece, the energy sector has undergone significant transformations, i.e., the liberalization of the energy market as part of the European Union’s energy transition, along with the changes in the energy production, transmission and distribution system, functioning under conditions of speculation, which have played a crucial role in the increases in prices. Additionally, the recent geopolitical tensions have deteriorated energy conditions. While the war in Ukraine may not be the primary cause of the increase in energy prices, it could potentially contribute to the persistence of high prices in the long run [11]. Another clear indication of the issue of energy poverty in recent years has been the intensive use of cheap and inappropriate materials as heating fuels, especially in the case of stoves and fireplaces, resulting in serious environmental and health consequences [12]. For instance, during the winter of 2012–2013 in Athens, the levels of particulate matter increased by 30%, and the concentrations of cancer-causing organic compounds in the air multiplied fivefold, leading to heart and respiratory issues among the city’s residents [13,14].

One of the first studies focused on EP in Greece was conducted in 2004 to gather data on the economic, energy, and social characteristics of households in the Athens area [15]. This survey, which included a total of 1110 households, revealed that 1.63% of households experienced energy poverty solely in terms of heating, while 0.35% faced extreme EP. When considering the overall energy needs, including heating and electricity, the percentage of households classified as energy-poor rose to 11.3%, with 2% falling into the category of extreme EP.

Over the past few years, there has been a notable increase in EP research, particularly due to the phenomenon’s rapid deterioration during the economic crisis. During the winter season of 2011–2012 in various regions of Greece, particularly in Athens, there was a notable decline in household energy usage [16]. A survey, gathering data on thermal energy usage from 598 households through a questionnaire, revealed that energy consumption for heating decreased compared to the winter season of 2010–2011 by an average of 15% in absolute terms. This decrease in consumption coincided with a rise in fuel prices and resulted in lower indoor temperatures. Upon analyzing the data based on income categories, it was discovered that the middle-income category (EUR 30,000–40,000) experienced the most significant decline in energy consumption compared to the previous year. This decline amounted to a decrease of 20.9% in absolute terms and a remarkable 72.1% relative to the anticipated consumption.

In their study on energy poverty, Atsalis et al. [17] utilized official data obtained by the Hellenic Statistical Authority (Household Budget Survey). They focused on the objective indicator of 10%, which represents energy costs exceeding 10% of income, and considered two income cases: total disposable income and equivalized income. The findings revealed that, in 2013, approximately 20–25% of Greek households experienced EP, marking a significant increase from the rates of 9–13% in 2008. A further analysis at the regional level indicated that northern Greece was more severely affected by the phenomenon, primarily due to higher thermal requirements.

Furthermore, Papada and Kaliampakos [12] conducted a thorough examination of the energy poverty problem in Greece. Their study, which included a sample of 400 households, covered a wide range of topics, such as living conditions, housing infrastructure, heating systems, subjective perceptions of energy needs, and quality of life. The study revealed that a significant percentage of Greek households, specifically 58%, experienced EP, measured by the 10% indicator (actual expenses compared to disposable household income). Moreover, Papada and Kaliampakos [18] conducted a similar study on the mountainous regions of Greece, which further emphasized the severity of the issue in these areas. The survey, which focused on a representative sample of 400 households living in mountainous regions, found that a significant percentage (73.5%) of these households in Greece experienced energy poverty as a result of the combined factors of higher energy costs and lower incomes.

A study conducted by Palmos Analysis [19] shed light on the significant problem of energy poverty in the Thessaloniki Urban Complex (TUC). The findings revealed that 62% of households in the TUC experienced EP, meaning that they spent more than 10% of their annual income on heating and electricity. Furthermore, 44% of respondents reported having to switch their heating system, usually from an oil burner to a natural gas system, due to the high cost. A significant portion of households (65%) reported having to reduce other expenditures in order to balance the rising expenses of heating, with 32% specifically affecting their food and supermarket budgets. Also, Boemi et al. [20] conducted a study in Western and Central Macedonia, where they analyzed 762 questionnaires to assess energy vulnerability in relation to building and socioeconomic characteristics, and Ntaintasis et al. [21] examined energy poverty within the Attica region using a combination of subjective, objective, and composite indicators. The data collected from 451 households indicated that the estimation of energy poverty can produce different results depending on the indicators used.

More recently, a study conducted by Spiliotis et al. [22] in Athens revealed a strong correlation between energy consumption and economic and weather conditions. In a subsequent study conducted by Papada and Kaliampakos [23], a new index developed called the “Degree of Coverage of Energy Needs” revealed that 45% of Greek households squeeze their energy needs, i.e., consume less energy than they theoretically need, 38.25% waste energy, while only 16.75% of households cover adequately their energy needs.

At the national level, Lyra et al. [24] used microdata from the EU-SILC survey and discovered that 40% of households in Greece experience energy poverty. By employing logistic regression models, the researchers identified various structural factors, such as the dwelling type, the location of residence, household income, and educational level, as the primary determinants of energy poverty in Greece. Kalfountzou et al. [25] conducted an integrated statistical analysis of the indicators of energy poverty for the case of Greece using binary logistic regression models to predict certain indicators (namely “10% rule”, M/2, and 2M) on the basis of socioeconomic factors. Finally, Halkos and Kostakis [26] found that approximately 9–10% of households appeared to be consistently affected by energy poverty, with factors such as education, dwelling characteristics, employment status, migration background, and income level influencing the likelihood of experiencing and overcoming this issue.

Furthermore, surveys conducted on Greek buildings have revealed a significant amount of energy waste compared to the rest of Europe. Specifically, during the period of 2011–2012, it was observed that Greek households showed the highest energy consumption in Europe, surpassing that of Spain by approximately 30% and that of Portugal by double. Surprisingly, even countries with colder climates, such as Belgium and the Nordic countries, had a significantly lower energy consumption [27]. According to Santamouris et al. [16], the primary reason for this waste can be attributed to the prolonged inaction of the Greek government in implementing laws for thermal insulation in buildings. It was only in 2010 that basic regulations (KENAK) were introduced into Greek legislation, aiming to align the country’s buildings with European standards. As a result, a large number of Greek people live in energy inefficient homes in terms of heat losses.

3. Study Area Main Characteristics

The Athens urban area (AUA), also known as the “Athens–Piraeus Urban Complex”, forms the core and center of Greater Athens and stretches across the Attica Basin over an area of 412 km² in Attica, the highest-populated region in Greece. The Athens urban area consists of 40 municipalities, 35 of which are located within four regional units of the former Athens Prefecture (North Athens, West Athens, Central Athens, and South Athens) and 5 municipalities are located within the regional unit of the former Piraeus Prefecture.

The climate of the area is mild. The average annual temperature over the last 30 years (1991–2020) was 18.5 °C, the total annual precipitation is roughly 433 mm, and the average humidity is 61% [28]. The average heating degree days (HDDs) and cooling degree days (CDDs) for the study area for the period of 2017–2022 were 1036 and 572, respectively [29].

With a population over three million, the AUA is the largest urban conglomeration in Greece, with a high population density. The population of the study area is ageing [30]. In line with this fact and compared to the last population census, the share of one-person households (34.8% in total) increased by around 37% and that of two-person households (27.4% in total) by 4%, while three-person (18.7% in total), four-person (14.6% in total), and five or more-person (4.6% in total) households decreased by 2.3%, 10.7%, and 9.9%, respectively [30].

In absolute terms, the gross domestic product (GDP) of the AUA was EUR 65.95 billion in 2020, accounting for about 40% of the whole Greek economic output. The GDP per capita was more than EUR 21,500 or 140% of the national average in the same year, and the unemployment rate stood at 14.2% (national unemployment rate: 16.3%). Significant income inequalities are also observed within the Athens urban area regions. The richest regions are Central and North Athens and the poorest is West Athens, whereas South Athens and Piraeus region stand in the middle. Compared to Central Athens, which is the richest region in the study area, the GDP per capita in North Athens is around 82%, in South Athens and Piraeus about 59%, and in West Athens only 35%.

The latest Greek Housing Census was conducted from July to October 2021, but the results have not yet been released. Therefore, the description of the energy performance characteristics of the buildings is based on the 2011 Greek Housing Census and the statistical results of the Energy Performance of Buildings Certificates, which are presented on an annual and quarterly basis for the Hellenic Territory by the Ministry of Environment and Energy (https://bpes.ypeka.gr/?page_id=21&stat=222 (accessed on 30 October 2023)). The total number of residences is around 1,662,500. About 62% of the houses were built before the implementation of thermal requirements and energy-related building codes (before 1980). The area with the oldest houses is Central Athens (around 75% of the houses were built before 1981), followed by the Piraeus region (about 62% of the houses were built before 1981). Around 12.4% of dwellings are less than 50 m², 39.3% are between 50 and 79 m², 32.5% are between 80 and 109 m², and the rest (i.e., 15.8%) are more than 110 m². More than 71% of dwellings are classified in the three worst energy classes (E, F, and G), about 25% in the middle energy classes (C and D), and about 4% in the highest energy classes (A+ to B). About 62% of the primary energy consumption is used for heating, 21.8% for domestic hot water (DHW), 16.2% for cooling, and less than 0.01% for lighting. Moreover, only 0.02% of the primary energy consumption is produced by RES.

4. Materials and Methods

The analysis in the study area regarding the conditions of EP was based on data from Eurostat’s EU Statistics on Income and Living Conditions (EU SILC) and Household Budget Survey (HBS) datasets, which were retrieved from the Hellenic Statistical Authority. The EU SILC and HBS survey microdata (at the household level) were provided for the years of 2017–2021. From the dataset, the observations selected were those that referred to region EL30 (variable DB040) and degree of urbanization 1 (variable DB100—cities/densely populated area).

Regarding the EU SILC dataset, the analysis was carried out using the following EP indicators:

• EP 1—Leaking roof, damp walls/floors/foundation, or rot in window frames or floor (variable HH040) (available up to 2020).
• EP 2—Inability to keep home adequately warm (variable HH050).
• EP 3—Arrears on utility bills (once and twice or more) (variable HS021).
• EP 4—Weighted Composite Index: 0.25 × HH040 + 0.5 × HH050 + 0.25 × HS021 (once and twice or more) (suggested by [31] (available up to 2020)).
• EP 5—Any form of EP (i.e., the household is defined as experiencing EP if any of the three main indicators HH040 or HH050 or HS021 apply) (available up to 2020).

EP indicators 1, 2, and 3 are the most widely used and commonly acknowledged consensual-based indicators. The rest of the indicators are used to explore the depth of EP (EP4), or a “worst-case” scenario where energy-poor households are those facing any form of EP (EP5). Moreover, the at-risk-of-poverty rate (households with an income of less than 60% of the median national equivalized income after social transfers) was considered.

The expenditure-based EP indicators, which were calculated using HBS data, were the following:

• The 2M indicator, which identifies as energy-poor the households whose share of energy expenditure in income is more than twice the national median. To calculate this indicator, both energy expenditures and income were equivalized to consider the differences in a household’s size and composition. The energy cost variable (HE045) and the disposable income variable (HH099) were divided by the equivalent energy cost and the equivalent household size (HB062), respectively. It was then examined whether the ratio of equivalized energy expenditure to equivalized disposable income (equivalized HE045/equivalized HH099) exceeded the double national median ratio in the current year. If the ratio surpassed twice the national median in the current year, the household was classified as energy-poor. Subsequently, the percentage of energy-poor households was computed using this criterion.
• The M/2 indicator, which identifies as energy-poor the households whose absolute energy expenditure is below half the national median or, in other words, abnormally low. As in the previous case, the energy expenditure was equivalized by the equivalent energy cost. The equivalized energy cost variable (HE045) was assessed to determine if it fell below half of the national median for the current year. If it was found to be lower than half of the national median, the household was categorized as energy-poor. Ultimately, the percentage of energy-poor households was computed based on this criterion.
• The official national energy poverty index (NEPI), according to which a household is classified as energy-poor if the following two conditions apply simultaneously: (i) the annual cost of the total final energy consumed by the household is lower than 80% of the expenditures theoretically required to cover the minimum final energy consumption of this household, and (ii) the total equivalized income of the household, which is influenced by the household’s size and composition and calculated using the modified OECD equivalence scale, is lower than 60% of the median equivalized income of all households in Greece, according to the definition of relative poverty.

The structure of the NEPI incorporates the key dimensions of the EP problem, namely the discrepancy between the consumed and required energy to ensure adequate internal thermal conditions in homes as well as households’ income. However, a key point of criticism of the NEPI is the ambiguity in defining the minimum required energy consumption that is used in developing condition (i) of the adopted definition. In addition, the identification of energy-poor households requires a complex calculation process and, particularly, the calculation of the minimum required energy consumption of the residence, which obviously depends on its characteristics, the level of thermal insulation, the climatic conditions, etc.

Aiming to overcome these problems, two new energy poverty indicators were formulated in the context of this analysis, as follows:

• The modified NEPI, which has the same structure as the NEPI and differs only in terms of condition (i), where in order to classify a household as energy-poor, the annual cost of the household’s energy consumption must be lower than 60% of the expenditures associated with the theoretically required energy consumption of its dwelling as determined by the national Regulation of Energy Performance of Buildings (KENAK).
• The modified LIHC (low income high cost) index that classifies as energy-poor a household with an equivalized residual income lower than 60% of the equivalized median income of all Greek households. The index is calculated by deducting the 60% of equivalized estimated energy costs required to ensure adequate energy services in its dwelling (based on KENAK theoretically required energy consumption) from the equivalized total income of the household. The 60% value of the theoretically required energy consumption is used to account for the real energy consumption, which was found to be 60% of the theoretically estimated needs in Greece. Energy costs are divided by the equivalization factors used in the UK LIHC indicator, in order to produce equivalized energy costs.

Furthermore, recognizing the need for an EP indicator that is simple enough to be useful and robust enough to be meaningful, the application of a simplified form of the LILEE (low income low energy efficiency) indicator (“modified LILEE”) was examined. Based on the modified LILEE, a household is considered energy-poor if its income is below 60% of the median income and if it resides in a low-energy class home. For the case study, homes built before 1980, i.e., before the implementation of the first insulation regulation were considered as “low-energy class”. The modified LILEE allows for the identification of EP households in the context of the “worst first” principle while avoiding extensive calculations. More specifically, for estimating this indicator using the HBS data, the following process was followed:

• The three first age classes of DS018 (year of construction) variable, i.e., houses built before 1980 in the case of Greece, were selected.
• The equivalized income was estimated using the HH099 (net income) and HB062 (equivalent size—modified OECD scale) variables.
• Ten different income classes were created using the median national income per year, as follows:

  –

  Income class 1—below 20% of the national median income.

  –

  Income class 2—between 20% and 40% of the national median income.

  –

  Income class 3—between 40% and 60% of the national median income.

  –

  Income class 4—between 60% and 80% of the national median income.

  –

  Income class 5—between 80% and 100% of the national median income.

  –

  Income class 6—between 100% and 120% of the national median income.

  –

  Income class 7—between 120% and 140% of the national median income.

  –

  Income class 8—between 140% and 160% of the national median income.

  –

  Income class 9—between 160% and 180% of the national median income.

  –

  Income class 10—over 180% of the national median income.

• The three first income classes, i.e., households with an income lower than 60% of the national median income, were selected. These households have incomes below the poverty threshold and live in the worst-performing buildings.

The official EP indicator (NEPI) and the new EP indicators developed, i.e., the modified NEPI, the modified LIHC, and the modified LILEE, were further explored using binary logistic regression with the commonly acknowledged EP drivers related to income, energy cost, and energy efficiency.

Binary logistic regression enables the estimation of the likelihood of a specific outcome in a binary categorical variable (such as energy-poor or non-energy-poor). This prediction is based on the values derived from independent variables. The method employs the odds ratio, calculated as the ratio of the probability of the desired outcome to the probability of the opposite outcome. In other words:

$$odds ratio=\frac{p}{1-p }$$

where p represents the probability of the desired contingency occurring. The accounting regression model is expressed in the following form:

$$\mathit{ln}\left(odds ratio\right)={\beta }_0+{\beta }_1{{\rm X}}_{1i}+{\beta }_2{{\rm X}}_{2i}+\dots +{\beta }_k{{\rm X}}_{ki}+{\epsilon }_{i }$$

where β₀, β₁, … , β_k are the coefficients awaiting estimation, k is the number of independent variables, and ε_i is the standard error of the i-th observation.

Therefore, if Y represents the binary categorical variable, where a value of 0 indicates non-energy-poor households and a value of 1 indicates energy-poor households, and p is the probability of Y taking the value 0, the accounting regression equation is formulated as:

$$\mathit{logit}\left(p\right)=\mathit{ln}\left(\frac{p}{1-p}\right)={\beta }_0+{\beta }_1{{\rm X}}_{1i}+{\beta }_2{{\rm X}}_{2i}+\dots +{\beta }_k{{\rm X}}_{ki}$$

The binary models were examined with model fit statistics, including the goodness of fit and pseudo R², as well as classification metrics, namely sensitivity, specificity, and overall accuracy. Sensitivity, also referred to as the true positive rate, denotes in this case the percentage of energy-poor households accurately identified as such, whereas specificity, known as the true negative rate, represents the percentage of non-energy-poor households classified correctly. The objective is to maximize both sensitivity and specificity; however, as the threshold probability becomes lower, the true positive rate is enhanced, but the true negative rate concurrently diminishes. The overall accuracy is the fraction of predicted values that are successful. These three metrics are calculated by the confusion matrix (Figure 1) using the following equations:

$$\mathrm{S}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y}=\frac{TP}{(TP+FN)}$$

$$\mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}=\frac{TN}{(FP+TN)}$$

$$\mathrm{A}\mathrm{c}\mathrm{c}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{y}=\frac{TP+TN}{(TP+FP+FN+TN)}$$

Finally, in order to test the stability of the logistic models over time, the Wald test was implemented [32], which examines the equality of the coefficients across the pooled (i.e., all years) and the annual models.

5. Results

5.1. EP Level Considering the Consensual Indicators

As mentioned above, according to Article 8, para. 3 of Directive (EU) 2023/1791 (EED recast), the assessment of EP by Member States considers the following indicators:

(a)

The inability to keep the home adequately warm;

(b)

Arrears on utility bills;

(c)

Total population living in a dwelling with a leaking roof, damp walls, floors or foundation, or rot in window frames or floor;

(d)

At-risk-of-poverty rate.

Moreover, under the same legislative provision, in cases where the proportion of households experiencing EP is not evaluated in Member States’ energy and climate plan, the average calculated share of the specified indicators is employed to ascertain the distribution of the necessary cumulative energy savings among individuals impacted by EP, vulnerable customers, residents of low-income households, and individuals residing in social housing. The proportion of households in EP over the period of 2017–2021, according to the consensual indicators and the requirements of Directive (EU) 2023/1791, is shown in

Table 1. EP assessment according to the EED (recast) indicators.

	2017	2018	2019	2020	2021	2017–2021 Average
EP1—Leaks	11.1%	12.1%	10.6%	11.2%	--	11.2%
EP2—Inability	27.0%	21.8%	18.9%	18.3%	16.7%	20.5%
EP3—Arrears	34.5%	30.5%	26.3%	22.6%	21.2%	27.0%
Poverty risk	32.2%	30.3%	29.0%	27.4%	28.3%	29.4%
Average	26.2%	23.7%	21.2%	19.9%	22.1%	22.1%

.

The share of the population living in a dwelling with a leaking roof or damp walls/floors/foundation within the study area is around 11%, without significant differences over the years of 2017–2021. The share of the population not being able to keep their home adequately warm is more than 20%, on average, but it follows a decreasing trend from 2017 onward. Similar conclusions are drawn for the share of population having arrears on utility bills, which is high (i.e., 27% on average). Yet, this indicator also follows the same decreasing trend from 2017 onward. Finally, the average EP share, including the at-risk-of-poverty rate, is 22%, remaining relatively stable over the last four years.

The relative improvement in terms of EP issues is also apparent in the results of the EP4 and EP5 indicators, which are presented in

Table 2. EP assessment according to the indicators EP4 and EP5.

	2017	2018	2019	2020	2017–2020 Average
EP4—no poverty	50.3%	54.8%	59.7%	63.3%	57.0%
EP4—25%	20.2%	21.0%	19.7%	16.4%	19.3%
EP4—50%	12.3%	10.5%	9.1%	9.2%	10.3%
EP4—75%	14.2%	10.8%	9.3%	8.6%	10.7%
EP4—100%	3.0%	2.9%	2.2%	2.5%	2.7%
EP5	49.7%	45.3%	40.3%	36.7%	43.0%

. The share of the population not experiencing EP issues increased from 50.3% in 2017 to 63.3% in 2020, whereas the share of those experiencing important or severe EP issues (i.e., the EP4 equals to 75% or 100%) dropped by more than 6%. Finally, the percentage of the population experiencing any type of EP, i.e., arrears on utility bills, inability to keep their house adequately warm, or leaks/damp walls, has constantly decreased since 2017. More precisely, a reduction of 26% was marked between 2017 and 2020.

Consensual EP indicators were investigated with respect to certain housing features and living conditions to explore the effect of the last ones on EP vulnerability in the pilot area. According to the findings of the analysis, households living in detached and semi-detached or terraced houses are more prone to almost all EU SILC EP indicators (arrears, leaks, and inability to keep home warm), with a focus on the problem of leaks, probably due to the existence of more indoor–outdoor spaces and open-to-air walls in these buildings, which makes it difficult to heat sufficiently a building. Furthermore, households living in one- or two-room houses present higher EP rates compared to the average, i.e., there are mainly problems with leaks, followed by arrears and an inability to keep the home warm, while households living in houses with four or more rooms have lower EP rates. Regarding tenure status, the most vulnerable groups to EP are tenants (mainly those at a reduced rate, followed by those at the market rate). Finally, households experiencing a great difficulty in making ends meet also face higher EP issues, with differences in EP rates of up to 21% compared to the average rates. On the other hand, households that can easily make ends meet present quite lower EP rates, of up to 29% versus average rates.

5.2. EP Level Considering Expenditure-Based Indicators

The expenditure indicators were calculated based on the HBS data and are presented in

Table 3. EP assessment according to expenditure-based indicators.

	2017	2018	2019	2020	2021	2017–2021 Average
2M	5.2%	3.7%	4.1%	3.9%	3.5%	4.1%
M/2	14.4%	15.9%	10.6%	12.7%	11.3%	13.0%
NEPI	11.5%	9.1%	9.8%	10.4%	8.8%	9.9%
Mod. NEPI	12.1%	9.5%	10.0%	10.7%	8.6%	10.2%
Mod. LIHC	24.4%	21.6%	21.2%	22.9%	18.8%	21.8%
Mod. LILEE	10.2%	7.4%	4.7%	4.3%	5.5%	6.4%

.

Considering the 2M indicator, the estimated levels of EP are very low (4% on average), which is largely attributed to the fact that the index does not count as energy-poor households that under-consume energy. Nevertheless, energy under-consumption is a situation that has been quite common in Greece during the last decade due to shrinking incomes and high energy prices. With the M/2 indicator, energy poverty levels were calculated to be between 11% and 16% (13% on average). This indicator, however, also shows significant weaknesses because it may classify as energy-poor households whose energy costs are low because they live in houses with a high energy efficiency or in warmer climate zones (e.g., the energy expenditures of households living in the AUA compared to those of households living in Northern Greece).

Following the official national indicator (NEPI) in the reference period, EP lies between 9.1% and 11.5% (about 10% on average). As expected, EP based on the modified NEPI is at the same level, given the similar structure of the two indicators. The estimated EP levels according to the modified LIHC indicator are twice as high (about 22% on average).

The discrepancy between the modified LIHC indicator and the NEPI/modified NEPI indicators primarily stems from the fact that the latter indicators necessitate that a household’s equivalized annual net income falls below 60% of the median equivalized income across all households. In essence, these indicators stipulate that a prerequisite for a household to be deemed energy-poor is to be classified below the officially recognized poverty levels. Altering these thresholds can significantly impact the estimated levels of energy poverty in a given region. To illustrate this, Figure 2 demonstrates that, by raising the levelized income threshold to 80% of the median national income, the EP rates calculated by the NEPI nearly double, reaching 18–20%, and are aligned with those of the modified LIHC. However, it is important to acknowledge the possibility that the divergence between the NEPI/modified NEPI and the modified LIHC indicators may be attributed to their distinct structures. Specifically, the NEPI and modified NEPI indicators evaluate the actual energy expenditure in relation to the required energy expenditure, while the modified LIHC indicator exclusively focuses on the expenditure required to attain satisfactory thermal comfort conditions within the home.

By applying the modified LILEE, 6.4% of households (i.e., 424 out of 6645 observations), on average, are considered as energy-poor. These households, according to the definition of the modified LILEE, are at risk of poverty (i.e., their income falls below 60% of the median equivalized national income) and live in the worst-performing buildings. To explore the relation of this indicator to the other three expenditure-based indicators, energy-poor and non-energy-poor households according to the NEPI, modified NEPI, and modified LIHC indicators were cross-tabulated against the modified LILEE (

Table 4. EP assessment according to expenditure-based indicators.

	Classification According to Mod. LILEE
	Non-EP	EP	Total
Classification according to the NEPI	329	333	662
Classification according to the mod. NEPI	336	344	680
Classification according to the mod. LIHC	1034	412	1446

).

According to the results, out of 424 EP households identified as energy-poor according to the modified LILEE, 333 (i.e., 78.5%) are also energy-poor according to the NEPI, 344 (or 81.1%) are energy-poor according to the modified NEPI, and 412 (i.e., 97.2%) are energy-poor according to the modified LIHC indicator. Hence, it can be argued that employing this simplified index allows for the identification of the vast majority of energy-poor households residing in the least energy-efficient homes. In addition to its simplicity and effectiveness, the simplified LILEE indicator offers a tool for analyzing various policy scenarios to investigate the extent of energy poverty for individuals residing in energy-inefficient homes. Yet, the main limitation in the use of this indicator, at present, is that the EU SILC survey does not include information on the age (or, if known, the energy class) of residences, and the HBS is not carried out on an annual basis in all EU countries. Also, it is not standard practice to provide information on the age of residence nor for the HBS.

5.3. Models for Identifying Energy-Poor Households

As mentioned above, several multivariate regression models were tested to examine the relationship between the four expenditure-based EP indicators (i.e., the NEPI, modified NEPI, modified LIHC, and modified LILEE) and commonly referred EP drivers. The consensual EP indicators as well as the M/2 and 2M EP indicators were not presented in this analysis since, for the former, the EU SILC does not contain sufficient background information on the characteristics of the dwellings and households in order to develop meaningful statistical models, while for the latter, the statistical models developed present a low predictive capability. The explanatory variables were related to both household and house characteristics. Moreover, for comparison reasons, the explanatory variables were common to all four models. Specifically, the explanatory factors tested were the following:

• Household income (HHINC, continuous variable—HH099 in HBS);
• Total number of household members (HHMEMB; integer variable—HB05 in HBS);
• Presence of elderly people (ELDER; binary variable: 1 if there are people over 65 years old; 0 if otherwise);
• Presence of unemployed household members (UNEMP; binary variable: 1 if there are unemployed household members; 0 if otherwise);
• Area of the house (AREA; continuous variable—DS017 in HBS);
• Heating fuel cost (FUELCOST; continuous variable: expressed in EUR per kWh, estimated by the authors using the statistical data from the Liquid Fuel Price Observatory of the Ministry of Development and other relevant sources);
• Building weighted U factor (UFACTOR; continuous variable: expressed in W/m²K, estimated by the authors using the proposed coefficients by the officially adopted Regulation of Energy Efficiency in Buildings in Greece);
• Efficiency of the heating system (HEATEFF; continuous variable: expressed in %, estimated by the authors based on the typical values of the efficiency ratio of heating systems in residential buildings).

The binary logistic model results are presented in

Table 5. Binary logistic regression results.

Variables	Description	NEPI	Mod. NEPI	Mod. LIHC	Mod. LILEE
HHINC	Household income	−0.0005 ***	−0.0005 ***	−0.0007 ***	−0.0019 ***
		(0.000)	(0.000)	(0.000)	(0.000)
HHMEMB	Household members	1.355 ***	1.397 ***	5.364 ***	5.521 ***
		(0.084)	(0.085)	(0.195)	(0.432)
ELDER	Elderly people (>65 years old)	0.738 ***	0.529 ***	0.534 ***	0.809 ***
		(0.142)	(0.140)	(0.142)	(0.309)
UNEMP	Unemployed members	1.068 ***	1.040 ***	1.134 ***	0.937 ***
		(0.157)	(0.156)	(0.145)	(0.353)
AREA	Dwelling area	0.015 ***	0.018 ***	0.017 ***	0.002
		(0.002)	(0.002)	(0.002)	(0.005)
FUELCOST	Heating fuel cost	4.489 ***	4.296 ***	11.309 ***	9.783 ***
		(1.421)	(1.417)	(1.563)	(2.977)
UFACTOR	Weighted U factor	0.478 ***	0.645 ***	0.710 ***	8.780 ***
		(0.070)	(0.072)	(0.073)	(0.750)
HEATEFF	Efficiency of the heating system	−0.049	−0.118	−1.008 ***	0.248
		(0.126)	(0.127)	(0.148)	(0.257)
CONS	Constant	−1.969 ***	−2.457 ***	−5.876 ***	−20.034 ***
		(0.337)	(0.340)	(0.396)	(1.930)
Observations		6645	6645	6645	6645
Pseudo R2		55.1%	56.0%	74.6%	87.4%

Standard errors in parentheses; *** p < 0.01.

. Furthermore,

Table 6. Prediction performance of the EP models.

	NEPI	Mod. NEPI	Mod. LIHC	Mod. LILEE
Sensitivity	56.04%	58.24%	86.38%	88.68%
Specificity	97.93%	97.85%	97.19%	99.55%
Accuracy	93.75%	93.80%	94.84%	98.86%

presents the prediction performance of the models.

In all four models, the explanatory variables exhibit the anticipated signs, except for HEATEFF in the modified LILEE model (however, the variable is not statistically significant). More specifically, with decreased income, increased household size, increased age and unemployment of household members, increased dwelling area and fuel cost, and decreased energy efficiency of the building envelope and heating system, the probability of a household being affected by EP is gradually increased. In addition, all variables are statistically significant in the NEPI, modified NEPI, and modified LIHC models, except for HEATEFF. Although the efficiency of the heating system is an important consideration in energy cost, the fact that the NEPI and modified NEPI indicators rely not only on the required but also the real energy consumption shows that the importance of this factor may be diluted. As far as the modified LILEE model is concerned, two variables, namely HEATEFF and AREA, are statistically insignificant. However, this can be explained by the fact that the area of the house and the efficiency of the heating system do not affect the efficiency of the building envelope or the household income, i.e., the two factors that define EP according to this specific indicator.

As far as the overall performance of the models is concerned, the most promising results are obtained from the modified LILEE model, followed closely by the modified LIHC model. Specifically, the modified LILEE model presents the highest goodness of fit ratio, with a pseudo R² of 87.4%, the highest sensitivity (i.e., the percentage of energy-poor households accurately identified as such), of about 89%, and the highest overall accuracy. As mentioned above, the modified LIHC model presents a similar performance, having a pseudo R² of 74.6% and a sensitivity of 86%. The NEPI and modified NEPI models exhibit lower fit statistics and classification metrics.

Finally, the Wald test was implemented to explore whether EP indicators can be predicted by the same (i.e., pooled) model over time, i.e., to check if the differences in the values of the coefficients across the pooled and the annual sub-models are statistically significant. The results are presented in

Table 7. Results of the Wald tests.

Models	NEPI Prob > χ2	Mod. NEPI Prob > χ2	Mod. LIHC Prob > χ2	Mod. LILEE Prob > χ2
Pooled-2017	0.3025	0.0843	0.0689	0.0772
Pooled-2018	0.6613	0.4888	0.2284	0.0034
Pooled-2019	0.1259	0.2037	0.2165	0.1085
Pooled-2020	0.4676	0.2103	0.0991	0.3322
Pooled-2021	0.4474	0.4263	0.3880	0.0439

. The null hypothesis that all the coefficients in the examined models are equal cannot be rejected between the pooled model and the annual models for the NEPI, modified NEPI, and modified LIHC indicators. Nevertheless, the conclusions differ for the modified LILEE model (i.e., the null hypothesis that all coefficients in the 2018 and 2021 models are equal to those of the pooled model is rejected).

6. Discussion, Conclusions, and Policy Recommendations

Addressing energy poverty (EP) is a top-priority concern within the framework of a just and inclusive transition to green energy due to its social, economic, and environmental impacts. The primary, and perhaps most crucial, step in this endeavor is to define the problem and, above all, to identify individuals who are currently experiencing or are at risk of EP. The recent Directive 2023/1791 defines EP and outlines specific indicators for calculating the percentage of energy-poor households in EU Member States. However, these indicators rely on data collected through surveys, notably the EU SILC, making their practical application limited. It is essential not only to calculate the rate of energy poverty but also to pinpoint households facing or at risk of facing EP. In addition to these indicators, two widely recognized expenditure-based indicators, 2M and M/2, encounter specific challenges. Beyond being calculated with questionnaire data, they may yield biased conclusions. For instance, the M/2 indicator may categorize as energy-poor some households residing in highly energy-efficient houses or warmer climatic zones within the same country, where the energy expenditure may be significantly lower than the national median. In Greece, unlike many Member States, there exists an official definition of EP based on a specific indicator (NEPI). However, as previously mentioned, this indicator faces practical challenges, relying on data that can only be collected through primary surveys (the actual energy expenditure), and faces criticism regarding methodological issues and the thresholds used.

To address these challenges, this research introduced three new EP indicators. The first, the modified NEPI, follows the same calculation methodology as the official Greek indicator (NEPI), introducing methodological adjustments while preserving its basic philosophy. It was anticipated to yield results similar to those of the NEPI. The second indicator, the modified LIHC, builds on a modification of the UK LIHC indicator. Calculated by deducting 60% of equivalized estimated energy costs from the equivalized total income of the household, it classifies a household as energy-poor when the equivalized residual income is lower than 60% of the equivalized median national income. This indicator estimates an EP rate approximately twice as high as the official and modified NEPI. The third indicator, the modified LILEE, is a simplified version of the UK LILEE indicator. Calculated through a relatively straightforward procedure, it identifies low-income households (below the poverty line) living in very-low-energy-efficient buildings. According to this indicator, the proportion of energy-poor households is about 40% lower than that calculated by the NEPI and modified NEPI and 70% lower than that of the modified LIHC.

The modified NEPI presents both the strengths and weaknesses of the official Greek indicator. Its primary drawback lies in the necessity to gather data through questionnaires to ascertain the actual energy costs. Conversely, the principal advantages of the modified LIHC and modified LILEE indicators lie in their ability to be computed using data readily available from governmental agencies and state authorities (e.g., tax authorities). Among all indicators, the modified LILEE stands out for its simplicity of calculation, while the modified LIHC offers the ability to gauge the depth of EP. Specifically, it enables the calculation of the EP gap, representing the difference between residual income and the threshold used (i.e., 60% of the equivalized median national income). However, it is important to note that these indicators are not without challenges. For instance, the modified LIHC, based on the theoretically required energy for thermal comfort, may overestimate the actual energy needs. While the modified LIHC adopts a highly conservative approach, emphasizing the worst-case first principle, EP issues may still affect households in moderately energy-efficient homes.

In terms of multivariate analysis, considering fit statistics and classification metrics, the modified LIHC and modified LILEE indicators outperform the NEPI and modified NEPI indicators. Nonetheless, all four indicators exhibit a satisfactory predictive performance, with the NEPI, modified NEPI, and modified LIHC models demonstrating also stability over time. Therefore, a key conclusion is that relevant government agencies and competent authorities can employ these models to estimate the likelihood of a household experiencing EP without requiring a primary questionnaire survey.

In conclusion, particularly from a policy perspective, this research introduces new tools for tackling EP. The novel indicators, particularly the modified LIHC and modified LILEE, which do not necessitate primary surveys or complex multivariate models, can help in identifying energy-vulnerable households. This, in turn, could facilitate the fulfillment of obligations arising from the European green energy transition policy. However, these findings should not be considered definitive. The new indicators, along with the multivariate models, need further exploration using national data, both in Greece and other European countries. An important obstacle in this regard, especially concerning their application to other EU countries, is the lack of data, as the HBS survey is not conducted annually in most Member States. Additionally, in future surveys, calculating consensual and expenditure-based indicators for the same counties would be beneficial. However, the EU SILC survey, used for consensual indicators, and the HBS survey, used for expenditure-based indicators, currently employ different samples. In this way and, due to the different households examined in the two surveys, the two datasets cannot be connected, and, thus, the outcomes derived from the analysis of consensual indicators (EU SILC survey) cannot be combined with those derived from the analysis of expenditure-based indicators (HBS survey). To address these challenges, certain measures could be considered. For example, a mandatory annual implementation of the HBS could be adopted, aligning it with the EU SILC schedule or, alternatively, the EU SILC dataset could be augmented with targeted variables for measuring energy poverty, like the year of dwelling construction, energy class certification (if any), expenditure on electricity, gas, and other fuels, as well as demographic information, such as the number of unemployed or economically inactive individuals in specific age groups (e.g., the number of persons aged lower than or equal to 4 years and number of persons aged more than or equal to 65 years). Furthermore, a specialized survey along with the HBS could be conducted every three years to obtain data about the energy performance and use of the utilized energy systems and equipment for all the end-uses (space heating, space cooling, and domestic hot water, cooking, lighting, and electric appliances), the potential implementation of energy efficiency interventions, and the energy behavior of the households. Therefore, to enhance the data used for policymaking, adjustments to the methodological framework of the two surveys should be considered, either aligning the samples or incorporating specific additional questions to enable the calculation of consensual and expenditure-based indicators.