Analyzing Energy Poverty and Its Determinants in Greece: Implications for Policy

Abstract

Energy and environmental policies in the sector of buildings aim to achieve climate targets while ensuring affordable energy services for households. This study uses the Greek residential sector as a case study and focuses on energy poverty, examining both established and novel energy poverty indicators for its measurement, analyzing the key determinants of energy poverty, and developing statistical models to identify energy-poor households. The same models are also used for assessing the effectiveness of policies and measures implemented or planned to address energy poverty with a view to develop synergies with policies aiming to reduce greenhouse gas emissions. Energy poverty levels in Greece ranged from 8.4% to 19.6% in 2021, depending on the energy poverty measure used. The evaluation of the policies aiming at tackling energy poverty showed that deep energy renovations, combined with space heating system upgrades, can reduce energy poverty by 69–99%. Shallow energy renovations and upgrades of space heating systems, implemented either individually or in combination, are less effective. Finally, while the various subsidy schemes for vulnerable households do not significantly affect energy poverty levels, they play a critical role in alleviating the depth of energy poverty and improving the quality of energy services provided to households.

1. Introduction

The sector of buildings, particularly residential buildings, contributes to climate change through the release of greenhouse gas (GHG) emissions associated with the consumption of fossil fuels and fossil-fueled electricity, as well as the use of f-gases as refrigerants in household appliances and materials (e.g., steel and cement) for the construction of dwellings [1]. On the other hand, it is itself affected by climate change, which alters the total energy requirements as well as the profile of energy consumption in the sector. In general, higher average temperatures expected under future climate scenarios are likely to reduce energy consumption for space heating while increasing energy demand for cooling [2]; however, it remains uncertain whether the net energy consumption will ultimately rise or fall.

Recognizing the critical role of the sector of buildings, the European Union and its Member States have implemented a range of policies and measures over the past two decades, aiming at upgrading the energy performance of both existing and new buildings, promoting energy efficiency, reducing total energy consumption, improving the carbon footprint of the sector, and adapting to new climatic conditions. In this context, the quantity and quality of the energy services received by the citizens is of paramount importance. The recently adopted European policy framework to address climate change (i.e., European Green Deal and REPowerEU) presents new challenges for households, particularly regarding their access to sufficient and affordable modern energy sources. For example, on the one hand, the promotion of policies and measures enhancing the energy upgrading of homes contributes to energy poverty alleviation [3,4,5], and, on the other hand, the implementation of ETS2 (The new European emissions trading system for buildings, road transport, and additional sectors), which may increase energy prices, could disproportionately affect the most vulnerable households [6,7].

Consequently, there is an urgent need for an integrated policy framework in the residential sector that simultaneously seeks to (i) reduce direct and indirect fossil fuel consumption, thus improving the sector’s carbon footprint; (ii) adapt to new climatic conditions; and (iii) improve the energy services provided to citizens with a particular emphasis on the alleviation of energy poverty, i.e., the inability of households to have affordable access to basic energy services, such as heating, cooling, lighting, etc. [8].

In Greece, the residential sector has been extensively studied as regards its GHG emissions reduction potential towards a low carbon economy (e.g., [9,10,11]), as well as its adaptation needs under future climatic conditions (e.g., [12,13,14]). However, these analyses do not adequately consider the level and quality of energy services received by citizens and the implications for energy poverty.

A growing body of the literature explores energy poverty in Greece, as the problem worsened dramatically during the economic crisis of the previous decade especially for the most vulnerable segments of the population. These studies focus on the measurement of the problem, its primary drivers, and its potential implications for public health and social welfare (see, for example, Refs. [15,16]). However, the issue of tackling energy poverty is considered individually and is not integrated in the wider framework of decarbonization, in the context of which several policies and measures target the residential buildings and attempt to reduce the energy consumption and the associated GHG emissions in the sector.

This study contributes to filling this gap, focusing on energy poverty and its determinants and identifying policies and measures that both tackle energy poverty and GHG emissions. Using Greece as a case study, it examines energy poverty through various indicators, explores its multiple dimensions and key determinants, and develops statistical models to identify energy-poor households, providing, in this way, a framework for evaluating the effectiveness of policies and measures implemented to address energy poverty, many of which are integral components of Greece’s National Energy and Climate Plan (NECP). The potential synergies among the policies aiming to tackle energy poverty and those planned to decarbonize the residential sector are also highlighted.

This paper is strongly linked to sustainability as energy poverty refers to the lack of access to affordable and modern energy services, which can hinder economic development, social equity, and environmental sustainability. Moreover, addressing energy poverty through sustainable energy solutions, such as energy efficiency interventions, can promote economic growth, improve health outcomes, and enhance the overall quality of life while simultaneously reducing environmental impacts.

2. Literature Background

Energy poverty is a multidimensional social problem, and the international literature offers a variety of indicators for its measurement. Broadly, the methods for measuring energy poverty can be classified into three categories [17,18,19,20]: (i) objective- or expenditure-based methods that asses households’ energy costs, particularly those spent or required for keeping the home adequately warm or cool, in relation to absolute or relative thresholds, which are often linked to households’ income; (ii) subjective or consensual assessments, which evaluate households’ perceptions of indoor housing conditions and their ability to achieve acceptable levels of energy services; and (iii) direct measurements, which compare the level of energy services (e.g., heating) achieved within a dwelling to a set standard.

Each energy poverty indicator has its strengths and limitations regarding its ability to describe the problem [21,22,23,24] and often covers different dimensions of energy poverty. Most researchers argue against single-indicator energy poverty metrics and advocate multiple-indicator approaches that explicitly acknowledge the shortcomings of each of the methods implemented (see, for example, Refs. [20,24]).

Objective- or expenditure-based approaches are commonly used as energy poverty measures. These approaches include absolute ratios of energy expenditure to household income, e.g., the “10%-rule” [25]; the high share of income on energy expenditure (2M) and the low absolute energy expenditure (M/2) [22]; or relative measurements, e.g., the Low Income High Cost (LIHC) indicator [26].

Subjective approaches for measuring energy poverty typically involve asking households about their ability to keep their home adequately warm, pay energy bills on time, as well as the condition of their dwelling [27]. In other words, they aim at assessing basic parameters or characteristics of a household, which are considered “socially perceived necessities” and whose absence can be taken as an indicator of energy poverty. However, they have been criticized for exclusion errors, where households may not self-identify as energy-poor despite having insufficient income to cover the basic social activities and practices of the household (for a critical review of these approaches, please see [18]).

Direct measurements of energy poverty face practical challenges, albeit with some exceptions, as datasets on energy services, such as indoor temperature surveys, are largely unavailable [20], and consequently their implementation requires extensive field research and the commitment of significant financial and human resources.

In recent years, new indicators have been developed and used by official organizations that attempt to directly relate energy poverty to the energy efficiency of housing, such as the Low Income Low Energy Efficiency (LILEE) indicator, which was proposed by the UK Department for Business, Energy and Industrial Strategy [28]. Also, researchers have developed multidimensional indicators with the aim of capturing the complex nature of the phenomenon, e.g., the Multidimensional Energy Poverty Index (MEPI) [29] and the Three-dimensional and Territorial Indicator of Energy Poverty (EPTTI) [30] or variations in objective indicators for measuring mainly the so-called ‘hidden energy poverty’ [31,32,33,34,35,36].

As far as Greece is concerned, previous research efforts have used a variety of energy poverty indicators such as the original or modified versions of the “10% rule” [37,38,39,40,41,42]; subjective indicators [16,21,43,44]; a combination of subjective, objective, and composite indicators [15,45,46,47]; as well as new indicators like the “Degree of Coverage of Energy Needs” (DCEN) [48] or variations in existing indicators [45,46]. The main characteristics of these research efforts are summarized in the following

Table 1. Main characteristics of previous research efforts on energy poverty (EP) in Greece.

Authors	Type of Data	Year of Data	Geographical Extent	EP Indicators	EP Results
[40]	Primary data (1110 households)	2004	Local (Athens)	“10% rule”	1.63%
[41]	Primary data (598 households)	2012	Nation-wide	“10% rule”	Winter 2009-10: 11.1% Winter 2010-11: 11.7%
[37]	Secondary data (EU-SILC and HBS)	2004 and 2008–2013 (HBS) 2003–2014 (EU-SILC)	Nation-wide	(i) “10% rule” using the full income (ii) “10% rule” using the equivalent income (iii) subjective indicators from EU-SILC	Objective indicators: from 2.5–4.8% (2004) to 20–25% (2013) Subjective indicators: from 17.4% to 30.5% (2003) to 13.7–37.3% (2014)
[44]	Primary data (400 households)	March to May 2015	Nation-wide	(i) subjective: EU-SILC’s indicators, health problems related to poor indoor conditions, and restriction of other essential needs to have adequate heating (ii) “10% rule”	Health issues: 22% Arrears: 18.8% Damp/mold problems: 37.5% Inadequately heated home: 42.2% Cutbacks in essentials to cover energy needs: 75% “10% rule”: 58%
[43]	Primary data (762 households)	November 2015 to July 2016	North Greece	(i) inability to maintain the temperature inside the house (ii) arrears in energy bills	(i) Inability to maintain the temperature inside the house: 22.6% (“rarely so”) (ii) Arrears in energy bills: 10%
[38]	Primary data (400 households living in mountainous areas)	March to May 2015	Nation-wide (mountainous areas)	(i) subjective: EU-SILC’s indicators, health problems related to poor indoor conditions, and restriction of other essential needs to have adequate heating (ii) “10% rule”	Health issues: 25% Arrears: 20% Damp/mold problems: 46% Inadequately heated home: 38.5% Cutbacks in essentials to cover energy needs: 73.5% “10% rule”: 73.5%
[39]	Combination of primary and secondary data	2015	Nation-wide	(i) “10% rule” with modeled energy consumption	“10% rule”: 18%
[47]	Primary data (451 households)	September and November 2017	Regional (Attica)	(i) objective: “10% rule” with real (O1) and estimated energy costs (O2) (ii) subjective: EU-SILC’s indicators (S1–S3), inability to keep home adequately cool during summer (S4), health problems related to poor indoor conditions (S5), and restriction of other essential needs to have adequate heating (S6) (iii) composite indices combining objective and subjective indicators (C1 and C2)	(i) Objective: O1: 24.6% and O2: 42.9% (ii) Subjective: S1: 35.7%; S2: 46.8%; S3: 33.5%; S4: 25.1%; S5: 16.6%; and S6: 58.3% (iii) Composite: C1: 37% and C2: 43.5%
[48]	Primary data (800 households—400 living in mountainous areas) combined with a stochastic model	March to May 2015	Nation-wide	Degree of coverage of energy needs	Country average: 0.98 Mountainous areas: 0.89
[42]	Primary end-user energy data provided by a natural gas company (4027 households)	2018	Regional (Attica)	“10% rule” adjusted by a set of weather-driven, income-oriented, and energy-oriented variables	A total of 15–27% spend more than 10% of their income for heating purposes, and more than 12% consume less energy than the desired amount
[45]	Primary data (300 households)	December 2018–January 2019	Local (Metsovo town)	(i) subjective indicators from EU-SILC (ii) subjective indicators by the authors (e.g., health problems due to inadequate heating) (iii) “10% rule	Health issues: 2% Arrears: 8% Damp/mold problems: 34% Inadequately heated home: 38% Cutbacks in essentials to cover energy needs: 65% “10% rule”: over 90%
[46]	Secondary data (EU-SILC and HBS)	HBS: 2008–2019 EU-SILC: 2003–2020	Nation-wide	(i) objective indicators: 10%, 2M, 2M EXP, M/2, and M/2 EXP (ii) subjective indicators from EU-SILC	10%: 10.6–25.1% 2M: 13.5–17.9% M/2: 12.5–18% Inadequately heated home: 12–32.9% Arrears: 15.7–42.2% Damp/mold problems: 12.5–20.9%
[16]	Secondary data (EU-SILC)	2010–2019	Nation-wide	(i) subjective indicators from EU-SILC (ii) a composite indicator	Inadequately heated home: 15.4–32.9% Arrears: 18.8–42.2% Damp/mold problems: 12.5–17.1% Composite indicator (2017 only): 41%
[21]	Secondary data (EU-SILC—4466 identical households over time)	2017–2020	Nation-wide	Subjective indicators from EU-SILC	Inadequately heated home: 22.5% Arrears: 28.2% Damp/mold problems: 14.9%
[15]	Secondary data (EU-SILC and HBS)	2017–2021	Local (Athens urban area)	(i) subjective indicators: EP 1-leaking roof, damp, and rot; EP 2-inability to keep home adequately warm; EP 3-arrears on utility bills; EP 4-weighted composite index; EP 5-any form of EP (ii) objective indicators: 2M, M2, official national energy poverty index (NEPI), modified NEPI, modified LIHC, and modified LILEE	Average values EP 1: 11.2% EP 2: 20.5% EP 3: 27% EP 4: 43% EP 5: 43% 2M: 4.1% M/2: 13% NEPI: 9.9% Modified NEPI: 10.2% Modified LIHC: 21.8% Modified LILEE: 6.4%

.

3. Materials and Methods

As already mentioned, this paper explores energy poverty and its determinants in Greece and attempts to provide policy recommendations in order to both tackle energy poverty and reduce the GHG emissions from the residential sector. This requires a deep understanding of the various dimensions of energy poverty, its determinants, and an evaluation of the effectiveness of the various interventions planned to address energy poverty with a view of developing synergies with carbon mitigation plans. The methodological framework developed and implemented in this paper comprises the following:

• A comprehensive analysis of energy poverty in Greece through existing and new indicators.
• An understanding of the various dimensions of energy poverty in Greece through an analysis of the energy poverty indicators examined and the segment of population that is affected in each case.
• The identification of the main determinants of energy poverty in Greece through the development of several logistic regression models, based on selected energy poverty indicators examined.
• The utilization of these models for evaluating the effectiveness of a number of policies and measures used or planned to tackle energy poverty in Greece and the exploitation of these results in order to maximize synergies with decarbonization strategy.

A more analytical discussion of these issues is presented below.

3.1. Measures of Energy Poverty

In the context of this study, we have utilized the microdata of the Household Budget Survey (HBS) for the year 2021, publicly available through the Hellenic Statistical Authority (ELSTAT), along with additional confidential survey data provided by ELSTAT in order to estimate energy poverty in Greece and analyze its characteristics through various indicators. The HBS in Greece is undertaken on an annual basis since 2008, collecting information on households’ composition, members’ employment status, living conditions, and mainly focusing on their members’ expenditure on goods and services as well as on their income [49]. The survey is conducted on a representative random sample of all private households of the country, and it is carried out by applying the two-stage stratified sampling method with the primary sampling unit for the area (one or more building blocks) and the ultimate unit for the household and its members [49]. The year 2021 was chosen because it is recent, includes detailed HBS data, and avoids the anomalies of 2020 (when household incomes and energy consumption were impacted by COVID-19 restrictions) and 2022 (when energy prices rose dramatically due to the war in Ukraine). In other words, 2021 is considered more representative of the typical living conditions of Greek households. The final sample size was 6053 households (the sampling fraction is about 1.5‰), which were equally distributed within the year [49].

A preliminary assessment of energy poverty in Greece is made using the 2M and M/2 indicators, which have been widely used across Europe [19]:

• The 2M indicator identifies households as energy-poor if their energy expenditure, as a share of income, exceeds twice the national median. For this calculation, both energy expenditure and income were equivalized to consider the differences in households’ size and composition (more details are presented in Section 5.1). Although this indicator is based on two key determinants of energy poverty, namely income and energy costs, it overlooks factors such as energy efficiency and housing characteristics. Also, a major drawback of this indicator is that it excludes households that under-consume energy, which has been a common occurrence in Greece over the last decade due to declining incomes and rising energy prices.
• The M/2 indicator classifies households as energy-poor if their equivalized absolute energy expenditure is less than half the national median, or, in other words, abnormally low. Despite its simplicity, this indicator presents significant weaknesses, as it may classify households with low energy costs due to high energy efficiency as energy-poor while overlooking households with high energy expenditures necessary to maintain adequate thermal conditions.

Aiming to provide a more comprehensive view of energy poverty, the Greek State, through the Action Plan to Combat Energy Poverty (APCEP), adopted the national energy poverty index (NEPI), which is also examined in this study [50]. According to the NEPI, a household is considered 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; (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 indicator incorporates key dimensions of energy poverty, namely the discrepancy between actual and required energy consumption for adequate thermal comfort and household income. However, the NEPI has been criticized for ambiguities in defining minimum required energy consumption (condition (i)) as well as for adopting a 60% income threshold (condition (ii)), which assumes that a household must first be classified as income-poor to be considered energy-poor. While energy poverty is more prevalent among households below the poverty line, it also affects higher income households. 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.

To address these limitations, additional energy poverty indicators were developed in this study based on the work of [15], with appropriate adjustments. Specifically, we formulated the HBS-4 energy poverty measure, which follows the NEPI structure but modifies condition (i), classifying a household as energy-poor if its annual energy consumption costs are less than 60% of the expenditures associated with the theoretical required energy consumption of its dwelling (as defined by the national regulation on energy performance of buildings, KENAK). Condition (ii) was also adjusted, with the equivalized annual net income threshold raised to 70% of the median equivalized income of all households in Greece.

Additionally, two energy poverty indicators commonly used in the UK were examined, namely

• The Low Income High Cost (LIHC) indicator [51] classifies a household as energy-poor if the following conditions apply simultaneously: (i) its required equivalized energy expenditure exceeds the national median; (ii) if the household were to spend that amount, its residual income would fall below the official poverty line. In the calculation process, the equivalent income is used (based on the equalized factors of the OECD), while the housing costs (i.e., rent and mortgage payments) is subtracted.
• The Low Income Low Energy Efficiency (LILEE) indicator, which was recently adopted in England [52], considers a household energy-poor if (i) it resides in a building with an energy efficiency rating of band D or lower and (ii) its disposable income (after housing and energy costs) is below the poverty line, mirroring the second criterion of the LIHC indicator. With respect to condition (i), in England this is determined by the most up-to-date Fuel Poverty Energy Efficiency Rating (FPEER) Methodology [52], while in this study, the fulfillment of this criterion is based on the energy performance certificates (EPCs) that consider all households living in a dwelling with EPCs D, E, F, and G as potentially energy-poor.

3.2. Detecting the Characteristics of Energy-Poor Households in Greece

It is widely recognized that in developed economies, energy poverty is associated with a low income, poor housing, inefficient heating systems, and high energy prices [18,20,53,54]. Additionally, these factors are related to the geographical location of the dwelling, the size of the household, the presence of elderly or unemployed members, the characteristics of the dwelling such as the surface area, the building type (detached house or apartment), its tenure status, and the education level of household members.

In this study, the relationship between energy poverty indicators and these factors was explored using microdata from the HBS, which is conducted annually in Greece. Logistic regression models were employed to establish a mathematical relationship between the dependent variable (whether a household is energy-poor or not) and a set of nominal-, ordinal-, interval-, or ratio-level independent variables that serve as predictors of energy poverty. The dependent variable Y is binary, taking the value of 1 if the household experiences energy poverty based on a specific indicator and 0 otherwise. The probability p that Y = 1 is modeled using the logistic regression equation:

$$logit(p)=ln\left({\displaystyle \frac{p}{1-p}}\right)={\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+\cdots +{\beta }_k{x}_k$$

(1)

where β₀, β₁, …, and β_k are the coefficients to be estimated from the regression analysis, and x₁, x₂, …, x_k are the independent variables of the model.

The logistic regression models developed incorporate as independent variables, and consequently potential factors influencing energy poverty levels, some of the parameters previously discussed. Specifically, the explanatory variables used as a starting point in developing the models in question, include:

• Household size (HB05): A categorical ordinal variable based on the number of household members.
• Presence of young children (KidsL4): A dummy variable taking the value 1 if the household has at least one member under four years old and 0 otherwise.
• Presence of elderly members (Elderly): A dummy variable taking the value 1 if the household has at least one member aged over 64 and 0 otherwise.
• Age of the dwelling (DS018): A categorical ordinal variable with categories as follows: 1: dwellings built before 1946, 2: dwellings built between 1946 and 1960, 3: dwellings built between 1961 and 1980, 4: dwellings built between 1981 and 1995, 5: dwellings built between 1996 and 2005, 6: dwellings built between 2006 and 2011, 7: dwellings built between 2012 and 2016, and 8: dwellings built after 2016.
• Dwelling area (DS017): A continuous variable representing the floor area of the residence in square meters.
• Housing tenure status (DS012): Three dummy variables to capture owner-occupied housing (reference category), housing with a mortgage (DS012-IM), accommodation provided for free by the employer or the family (DS012-F), and rented housing (DS012-R). Depending on the tenure status of the residence, the corresponding dummy variable takes the value 1 and 0 otherwise.
• Building type (DS011): Three dummy variables representing semi-detached houses (DS011-SD), apartment buildings with fewer than 10 units (DS011-FSB), and apartment buildings with 10 or more units (DS011-FBB), with detached houses as the reference category. Depending on the building type of the residence, the corresponding dummy variable takes the value 1 and 0 otherwise.
• Heating system (DA028): Seven dummy variables that simulate eight different residential heating systems, i.e., central heating systems with diesel oil (reference category), heating systems with natural gas (DA028-NG), oil or LPG stoves (DA028-S), heat storage appliances (DA028-HS), firewood stoves (DA028-W), electric heater appliances (DA028-EH), air conditioning or heat pumps (DA028-HP), other systems (DA028-OTHER). Depending on the heating system of the residence the corresponding dummy variable takes the value 1 and 0 otherwise.
• Geographical location of the dwelling (ELXX): Twelve dummy variables that simulate the 13 regions of the country based on the NUTS2 coding as defined by EUROSTAT, with Attica as the reference region. Specifically, we have included the following dummy variables: EL41 for North Aegean, EL42 for South Aegean, EL43 for Crete, EL51 for Eastern Macedonia and Thrace, EL52 for Central Macedonia, EL54 for Epirus, EL61 for Thessaly, EL62 for the Ionian Islands, EL63 for Western Greece, EL64 for Central Greece, and EL65 for Peloponnese. Again, depending on the region of the residence, the corresponding dummy variable takes the value 1 and 0 otherwise.
• Household income (HH095th): A continuous variable representing the total household income in thousands of euros.
• Unemployment status (UNEM-C): A dummy variable taking the value 1 if there is at least one unemployed household member and 0 otherwise.

All data used to develop these models were derived from the 2021 HBS, utilizing both public-use datasets and additional data from the Hellenic Statistical Authority (ELSTAT).

Having determined the logistic regression models, the probability that a household j is classified as energy-poor, p_j, is calculated using the following equation:

$$p_j={\displaystyle \frac{e^{z_j}}{1+e^{z_j}}}$$

(2)

where

$$z_j={\beta }_0+{\beta }_1{x}_{j,1}+{\beta }_2{x}_{j,2}+\cdots +{\beta }_k{x}_{j,k}$$

(3)

The logistic regression models were evaluated using model fit statistics, including goodness-of-fit measures and pseudo R-squared values, as well as classification metrics such as accuracy, precision, sensitivity, and specificity. These metrics are derived from the confusion matrix, which summarizes the model’s classification results. The confusion matrix includes the following elements:

• True Positive (TP): The number of households correctly classified as energy-poor by the model.
• True Negative (TN): The number of households correctly classified as not energy-poor by the model.
• False Positive (FP): The number of households incorrectly classified as energy-poor by the model.
• False Negative (FN): The number of households incorrectly classified as not energy-poor by the model.

From these values, the following performance metrics were calculated:

The accuracy of the model is the most frequently used metric in order to evaluate the overall performance of the model’s predictions, where

$$Accuracy={\displaystyle \frac{TP+TN}{TP+TN+FP+FN}}$$

(4)

The precision of the model focuses on the accuracy of positive (energy-poor) predictions in relation to cases wrongly predicted as positive, where

$$Precision={\displaystyle \frac{TP}{TP+FP}}$$

(5)

The sensitivity of the model evaluates the ability of the model to correctly identify energy-poor households, where

$$Sensitivity={\displaystyle \frac{TP}{TP+FN}}$$

(6)

The specificity of the model assesses the model’s ability to correctly classify households that are not energy-poor, where

$$Specificity={\displaystyle \frac{TN}{TN+FP}}$$

(7)

Higher values for all these metrics indicate better classification performance, though there is often a trade-off—improving sensitivity can reduce specificity, and vice versa.

The final structure of the models developed is presented in Section 5.2. They are also used to evaluate the effectiveness of a number of policies and measures that can be implemented to tackle energy poverty. Specifically, each measure in question is associated with one or more explanatory variables of the models developed. The implementation of the measure means that the associated explanatory variables change values, and consequently the energy poverty levels are re-estimated through the corresponding logistic regression model adopted. The achieved reduction in the energy poverty levels shows the effectiveness of the implemented measure. More details on this approach are presented in Section 5.3.

4. The Residential Sector in Greece

The existing building stock in Greece comprises approximately 4.6 million dwellings. A significant portion of this stock (56%, or about 2.6 million dwellings), was constructed before 1980, prior to the introduction of the first regulation on the thermal insulation of buildings and therefore has very poor energy performance. Only 1.7% of dwellings were built after 2010, which was when the new regulation on the energy performance of buildings (KENAK) was adopted.

As shown in Figure 1, the total energy consumption in the Greek residential sector decreased from 4.6 Mtoe in 2000 to 4.3 Mtoe in 2021 (−9.4%). The trend, however, was not uniform. For about 12 years (2000–2011), energy consumption is seen to increase albeit with fluctuations. Its de-escalation begun in 2012. It coincided with the financial crisis that hit the Greek economy in the 2010s and seems to have played an important role in changing the energy behaviors of households. During the first 2–3 years after the onset of the financial crisis, the use of central heating systems with diesel oil was gradually abandoned by roughly one in two households, and the share of petroleum products in the total energy consumption of the residential sector decreased from 53% in 2000 to ca 25% by 2013 and to 27% in 2021, mainly because of increased oil prices and reductions in households’ income. In the same period, the role of natural gas and electricity increased, with the former covering 12% of total energy needs in the residential sector in 2021 from almost zero in 2000 and the latter covering 36% of the energy needs in 2021 from 27% in 2000. The share of RES increased from 19% in 2000 to 25% in 2021, with biomass used for space heating and solar energy for hot water being the renewables (henceforth RES) with the highest contributions in the sectoral energy mix. Figure 1 also shows the evolution of direct (i.e., from the use of fuels) and indirect (i.e., from the use of electricity) CO₂ emissions of the sector. The reductions achieved in the period 2012–2021 were impressively large (more than 54%) compared to the period 2000–2012, during which a 4% increase was recorded.

To a large extent, the significant improvement in the carbon footprint of the Greek residential sector during the period 2000–2021 and particularly after 2012 came from forced behavioral changes due to the increase in fuel costs (e.g., the gradual phasing out of heating oil), as well as from the decarbonization of the power generation sector. The former is strongly related to the dramatic deterioration in energy poverty indicators observed during this period, as energy prices increased and the available income of households decreased. Figure 2 presents the evolution of energy poverty in Greece during the last 20 years (2003–2022) based on the three subjective indicators from the EU-SILC survey, namely the (i) inability to keep homes adequately warm (S1), (ii) arrears on utility bills (S2), and (iii) dwellings with leaking roof; damp walls, floors, or foundation; or rot in window frames or floor (S3). The analysis of these indicators shows that energy poverty is significantly more pronounced among lower-income households. The fluctuations in indicators S1 and S2 during the period 2010–2021 indicate that the very high energy poverty rates (above 25%) observed during this period were largely driven by the economic crisis, and the associated reductions in household income (the average income of households declined by more than 35% due to the austerity policies implemented [47]) and rising energy prices were due to new taxes imposed [37] during the period 2010–2017. Moreover, 15% of households live in houses with significant structural issues, indicating that energy poverty in these cases is likely to be more deeply rooted in structural conditions. Energy-saving interventions promoted during the same period were limited and yielded much lower energy savings, leaving a significant percentage of the Greek households without adequate heating for their homes and with delays in utility bill payments.

The poor energy performance of the Greek building stock seems to be the common ground for contributing on the one hand to energy losses and high GHG emissions, and on the other hand to the worsening of energy poverty indicators, as shown by recent studies [15]. Hence, there could be potential for synergies between policies to tackle energy poverty and the transition to a carbon-neutral economy, especially when the focus is on energy efficiency promotion [55]. Nevertheless, these two related agendas could also easily create undesirable trade-offs if both goals, i.e., energy poverty alleviation and decarbonization, are not tackled together. The latter is even more significant for low-income, vulnerable households who enter the energy transition at a disadvantage. The poorer households, who are amongst the lowest carbon emitters and struggle to pay their energy bills, may not be able to invest in energy efficiency or suffer from higher energy prices due to carbon taxes [56].

The Greek NECP, which was submitted to the European Commission in December 2024 [9], highlights the building sector as particularly critical for achieving the goals of decarbonization and improving energy efficiency. In this context, a target to reduce the average primary energy use in residential buildings by 16% in 2030 and by 20–22% in 2035 has been adopted. The main measures adopted to this end include (i) the promotion of efficient heating and cooling systems, with an emphasis on heat pumps; (ii) energy renovations, with a priority to shallow renovation (approximately 63–83 thousand dwellings per year up to 2050); (iii) the promotion of efficient electric appliances; and (iv) the installation of RES systems, particularly solar systems for water heating and photovoltaics (with or without storage) for electricity generation. Overall, emphasis is given primarily on electrification and to a lesser extent on reducing energy demand.

The NECP also acknowledges energy poverty as a significant issue with broad economic, social, political, health, and environmental implications and notes the potential synergies that the policies and measures planned for reducing the GHG emissions in the residential sectors and tackling energy poverty may have. The NECP sets ambitious targets to reduce energy poverty in Greece by 50% in 2025 and by 75% in 2030 compared to 2016 levels. A key tool to achieve these goals is the Action Plan to Combat Energy Poverty (APCEP) that was adopted in September 2021, which outlines the definition of energy-poor households and includes various existing and new policies and measures to address the problem by utilizing a number of financial programs as well as available market mechanisms [50].

The policy measures included in the APCEP are structured around three axes [50]: (i) policy measures aimed at protecting vulnerable consumers, particularly during periods of high energy prices and extreme energy poverty; (ii) policy measures aimed at long-term structural reductions in energy poverty, through improvements in energy efficiency and the promotion of renewables in residential buildings; and (iii) information and education actions targeting vulnerable households that are supported by energy efficiency obligation schemes and centrally planned outreach efforts. In other words, the APCEP aims to address energy poverty in the short run by providing financial aid and ensuring access to minimum levels of energy services and in the long run by upgrading home energy efficiency, modernizing heating systems, and promoting renewable energy solutions.

5. Results and Discussion

5.1. Rate of Energy Poverty in Greece Using Different Measures

To estimate the energy poverty indicators described in Section 3.1, some additional data and assumptions need to be clarified. As already mentioned above, in the calculation framework equivalization factors are often used to appropriately adjust the household income and energy costs, allowing for the effective integration of differences in household size, composition, and characteristics. Table 2 summarizes the equivalization factors used in this analysis. In addition,

• For the 2M and M/2 indicators, equivalent energy costs were calculated based on the scale shown in Table 2, while for the 2M indicator, the total household income from all sources was used and equivalized on the basis of the modified OECD scale.
• For the NEPI indicator, net household income was equivalized based on the original OECD scale, while energy costs were used in their primary form.
• As regards the HBS-4 indicator, two variants were calculated: The HBS-4a uses the old OECD scale to equivalize net incomes, while the HBS-4b uses the modified OECD scale. In both cases the energy costs were used in their primary form.
• The LIHC indicator considers the total income after housing costs. This is calculated by deducting from the total income of households the annual rent or mortgage payments for the permanent residence, which are both provided by the HBS. The resulting after housing cost income along with households’ energy expenditures are then equivalized using the corresponding scales provided in Table 2.
• As regards the LILEE indicator, since the HBS does not provide the energy class of the dwellings, it was assumed that in order to meet condition (i), a household should reside in a dwelling built before 1980. Regarding condition (ii), the same assumptions with those adopted for calculating the LIHC indicator have been used.

Table 2. Equivalization factors for the income and energy costs used in calculating the energy poverty indicators. Source: OECD ¹ and BRE ².

Income				Energy Costs
Composition of the Households	Old OECD Scale	Modified OECD Scale	Scale for After Housing Costs Income Used in the UK	Number of People in the Household	Scale
First adult	1	1	0.58	One	0.82
Subsequent adults (includes partners and children aged 14 or over)	0.7	0.5	0.42	Two	1.00
Children under 14	0.5	0.3	0.20	Three	1.07
				Four	1.21
				Five or more	1.32

¹. What are equivalence scales? OECD Project on Income Distribution and Poverty via www.oecd.org/social/inequality.htm (accessed on 13 November 2023). ². Department of Energy and Climate Change. Annual Fuel Poverty Statistics Report, 2015, England.

Table 2. Equivalization factors for the income and energy costs used in calculating the energy poverty indicators. Source: OECD ¹ and BRE ².

Income				Energy Costs
Composition of the Households	Old OECD Scale	Modified OECD Scale	Scale for After Housing Costs Income Used in the UK	Number of People in the Household	Scale
First adult	1	1	0.58	One	0.82
Subsequent adults (includes partners and children aged 14 or over)	0.7	0.5	0.42	Two	1.00
Children under 14	0.5	0.3	0.20	Three	1.07
				Four	1.21
				Five or more	1.32

¹. What are equivalence scales? OECD Project on Income Distribution and Poverty via www.oecd.org/social/inequality.htm (accessed on 13 November 2023). ². Department of Energy and Climate Change. Annual Fuel Poverty Statistics Report, 2015, England.

Figure 3 shows the energy poverty levels in Greece for 2021, which were calculated using the seven objective indicators considered. For comparison, subjective indicators S1 and S2 from the EU-SILC survey are also presented. Based on the objective indicators examined, the energy poverty rate in Greece in 2021 ranged from 8.4% to 19.6%. The official energy poverty indicator NEPI estimates energy poverty levels at 12.9%. It seems that condition (ii) of the NEPI, which determines that the equivalent income of the households should be below the poverty line, limits significantly the number of households classified as energy-poor. On the other hand, the two HBS-4 indicators (which assess energy needs using KENAK and a higher household income threshold), as well as the LIHC and LILEE indicators, result in higher energy poverty rates, ranging from 18.2% to 19.6%.

It is worth noting that households identified as energy-poor by one indicator may not be classified as such by another, reflecting the multidimensional nature of energy poverty, and that each indicator, by design, captures different aspects of the problem. Figure 4 illustrates the degree of overlap between selected energy poverty indicators. Specifically, Figure 4a clearly shows that the 2M and M/2 indicators capture different dimensions of energy poverty focusing on different population segments (the former on households that have high energy costs in relation to their income and the latter on households that under-consume energy). The NEPI indicator by its structure focuses on households under the line of poverty that under-consume energy. The strict income criterion adopted significantly reduces the number of households with low energy expenditures that are classified as energy-poor. On the other hand, it includes a significant portion of households under the line of poverty that, despite under-consuming energy, have resulting expenditures that constitute a significant part of their income. However, almost two thirds of households with high energy costs in relation to their income are not classified as energy-poor with the NEPI. Figure 4b compares the NEPI, HBS-4b, and LIHC indicators. Energy poverty rates calculated with these indicators were 12.9%, 19.3%, and 18.7%, respectively. Notably, 6.9% of households were classified as energy-poor by all three indicators. As the HBS-4b indicator has a similar structure with the NEPI and uses broader criteria for identifying energy-poor households, not surprisingly, it captures nearly all households classified as energy-poor by the NEPI. In contrast, the LIHC indicator shows a moderate correlation with the other two, with approximately 44% of households identified as energy-poor by LIHC but not classified as such by either of the other two measures. This is attributable to the structure of LIHC, which focuses on the theoretical energy expenditures of the households and does not take into account the difference between the theoretical and real energy expenditures that is considered by the NEPI and HBS-4b. Finally, Figure 4c examines the overlap between the NEPI, HBS-4b, and LILEE indicators, demonstrating a similar pattern of relevance and coverage across the different dimensions of energy poverty. The difference between the LILEE and the other two indicators is again attributed to the fact that the former focuses on the energy performance of the dwelling without taking into account the energy behavior of the users. From the above, it is clear that the problem of under-consumption of energy also exists in households that live in residences with relatively higher energy performance. On the other hand, the NEPI and HBS-4b indicators fail to identify energy-poor households that cover their energy needs by spending a significant portion of their income.

Based on the above it should be noted that the indicators 2M and M/2 cover specific dimensions of energy poverty; NEPI and HBS-4 also focus on the under-consumption of energy of lower-income households, while the indicators LIHC and LILEE adopt a broader approach to specify households under energy poverty based on the available income and the energy performance of the dwellings. Also, the indicators NEPI, HBS-4, and LIHC implement a quite complex calculation framework to estimate the required energy costs, which are incorporated in the evaluation criteria adopted.

5.2. Models for Detecting the Characteristics of Energy-Poor Households

Table 3. Logistic regression models for the identification of energy-poor households based on three different energy poverty indicators. B: the coefficients of the model used for predicting the dependent variable; S.E.: the standard errors associated with the coefficients; Sig.: the p-value of the independent variables showing their significance; and Exp(B): the exponentiation of the B coefficients, which are the odds ratios for the predictors.

	NEPI				HBS-4b				LIHC
	B	S.E.	Sig.	Exp(B)	B	S.E.	Sig.	Exp(B)	B	S.E.	Sig.	Exp(B)
HB05	1.909	0.089	0.000	6.743	1.475	0.072	0.000	4.373	1.137	0.071	0.000	3.116
DS018	−0.454	0.056	0.000	0.635	−0.703	0.053	0.000	0.495	−1.494	0.067	0.000	0.224
DS017									0.017	0.002	0.000	1.017
EL42									−0.754	0.313	0.016	0.471
EL51									2.046	0.206	0.000	7.734
EL52									2.150	0.154	0.000	8.584
EL53	0.930	0.283	0.001	2.534					2.540	0.252	0.000	12.674
EL54									2.113	0.226	0.000	8.270
EL61					0.474	0.194	0.015	1.606	1.992	0.209	0.000	7.327
EL62									−0.978	0.351	0.005	0.376
EL63	0.712	0.208	0.001	2.039	0.625	0.196	0.001	1.869	0.470	0.204	0.021	1.600
EL64	−0.608	0.273	0.026	0.545
KidsL4									−1.539	0.348	0.000	0.215
Elderly	0.573	0.140	0.000	1.773	0.631	0.120	0.000	1.879	1.045	0.139	0.000	2.844
UNEM-C									0.399	0.143	0.005	1.490
HH095th	−0.667	0.025	0.000	0.513	−0.615	0.021	0.000	0.540	−0.320	0.013	0.000	0.726
DS012-IM									0.866	0.274	0.002	2.376
DS012-R									2.121	0.173	0.000	8.337
DS011-FSB					−0.282	0.113	0.013	0.754	−1.891	0.144	0.000	0.151
DS011-FBB									−2.119	0.180	0.000	0.120
DA028-NG									−1.797	0.247	0.000	0.166
DA028-S	0.926	0.253	0.000	2.524	0.964	0.238	0.000	2.622	1.826	0.237	0.000	6.207
DA028-HS									−4.218	1.075	0.000	0.015
DA028-W									−0.966	0.167	0.000	0.381
DA028-EH	1.041	0.162	0.000	2.833	1.261	0.151	0.000	3.529	1.405	0.173	0.000	4.074
DA028-HP	0.975	0.179	0.000	2.651	0.920	0.164	0.000	2.510	−1.366	0.244	0.000	0.255
DA028-OTHER					0.445	0.162	0.006	1.560	0.987	0.157	0.000	2.684
Constant	1.310	0.262	0.000	3.705	3.571	0.254	0.000	35.538	2.090	0.290	0.000	8.082
−2 Log likelihood	2025.488				2586.692				2568.087
Cox & Snell R-Squared	35.1%				42.6%				41.8%
Nagelkerke R-Squared	65.6%				68.1%				67.5%

presents the logistic regression models developed in this study, which aim to identify energy-poor households based on key factors that are easily detectable, measurable, and statistically significant. In total three logistic regression models were developed, with each using a different energy poverty indicator as a dependent variable to characterize the energy poverty status of the households. Specifically, we present here the models based on the NEPI, HBS-4b, and LIHC indicators, while the analysis was carried out with the SPSS computer package (version 25). The analysis was not performed for the 2M and M/2 indicators as they cover specific dimensions of energy poverty without providing a more holistic view of the problem, and the analysis is not provided for the HBS-4a and LILEE indicators as their characteristics present significant similarities with the HBS-4b and LIHC indicators, respectively (see Figure 4).

Based on the signs of the estimated coefficients in each model, the following conclusions can be drawn:

• Household income (HH095th): In all three models, an increase in net household income decreases the probability of being characterized as energy-poor.
• Building age (DS018): The age of the dwelling was found to be a statistically significant factor in all models, affecting the probability of a household being characterized as energy-poor. Older homes generally have higher energy losses, requiring more energy to achieve adequate thermal comfort conditions, which increases the likelihood of energy poverty.
• Household size (HB05): Larger households are associated with a higher likelihood of energy poverty in all three models. This is likely due to increased energy needs and costs that these households have, as well as the necessity of maintaining adequate thermal comfort conditions for longer periods due to more household members being present.
• Household composition: The presence of elderly individuals in the household correlates positively with energy poverty across all models. In the model using the LIHC indicator, the presence of unemployed members is also associated with increased energy poverty, while the presence of young children (aged up to 4 years) is associated with a lower likelihood of energy poverty.
• Floor area (DS017): Larger floor areas are positively correlated with energy poverty only in the LIHC model. Larger dwellings usually require higher energy expenditure to achieve adequate thermal comfort conditions, and therefore the households living there have increased odds of energy poverty.
• Geographical location (ELXX): To some extent, regional differences in energy poverty are evident in all models, with Attica serving as the reference region. If a region is not included as an independent variable in the model, it indicates no significant difference compared to the reference region (i.e., Attica). More specifically, the NEPI and HBS-4b models present small geographical variations, with only 2–3 regions per model showing increased or reduced probabilities of energy poverty from the reference region. As these indicators focus on the under-consumption of energy, people have the flexibility to adjust their behavior and the use of heating systems, considering the climatic conditions, the energy prices, their income, etc. On the other hand, the LIHC model presents significant geographical differences, with six regions in Northern and Central Greece presenting increased rates of energy poverty, and two island regions reduced the chances of energy poverty. As the energy poverty indicator used in this model take into account the required energy expenditures for ensuring adequate indoor conditions, the climatic conditions, which differ significantly across the country, emerge as a key determinant of energy poverty.
• Heating systems: In all three models, heating with oil or LPG stoves (DA028-S) and electric appliances (DA028-EH) is linked to a higher likelihood of energy poverty. In the LIHC model, natural gas (DA028-NG), heat storage (DA028-HS), and biomass (DA028-W) systems are linked to lower energy poverty odds. Air conditioning and heat pumps (DA028-HP) are associated with increased energy poverty in the two models (NEPI and HBS-4b) but decreased odds in the other (LIHC). This could be attributed to the fact that these three indicators cover different populations affected by energy poverty. The NEPI and HBS-4b focus on households that under-consume energy, which, in many cases, use air conditioners to heat only some rooms of their dwelling. On the contrary, the LIHC incorporates the required energy needs of the dwellings, and as the air conditioners/heat pumps are characterized by relatively high efficiency, the model identified this technology as an option for reducing energy expenditures. Heating technologies not included as independent variables do not show a statistically significant impact compared to the reference technology (diesel oil).
• Housing type: In two models, households living in apartment buildings have a lower probability of being energy-poor compared to those in detached houses. Generally, the energy demand in these buildings can be satisfied more easily due to the proportionally smaller openings in their shell, thus resulting in less energy losses. Moreover, due to economies of scale, the fixed heating costs, such as maintenance of the boiler, cleaning costs, etc., for these dwellings are relatively lower.
• Tenure status: In the LIHC model, households paying a mortgage (DS012-IM) or rent (DS012-R) have a higher likelihood of energy poverty.

The three logistic regression models developed in the context of the present analysis include only statistically significant independent variables, and their initial evaluation was performed using the −2 Log likelihood and pseudo-R2 statistics (see Table 3). A further assessment of models’ performance is presented in

Table 4. Prediction performance of the logistic models developed.

	NEPI	HBS-4b	LIHC
TP	478	845	775
TN	5126	4645	4678
FP	149	238	241
FN	300	325	359
Accuracy	92.6%	90.7%	90.1%
Precision	76.2%	78.0%	76.3%
Sensitivity	61.4%	72.2%	68.3%
Specificity	97.2%	95.1%	95.1%

, using accuracy, precision, sensitivity, and specificity as evaluation metrics, as discussed in Section 3.2. All three models demonstrate high accuracy, exceeding 90%. Precision is between 76% and 78%, while sensitivity (the ability to correctly identify energy-poor households) ranges from 61% to 72%, which is satisfactory for the purposes of this analysis (i.e., evaluation of energy poverty policies). The performance of the models can be further improved by incorporating additional parameters as independent variables (e.g., energy expenditure). However, this information is not readily available in government databases and requires additional modeling in cases where the models will be used for identifying energy-poor households with a view to implement targeted policies to alleviate them.

5.3. Evaluation of Main Policies Planned in Greece for Tackling Energy Poverty

The logistic regression models developed in Section 5.2 provide a framework for evaluating the effectiveness of the planned interventions for tackling energy poverty in Greece (presented in Section 4). The policy measures examined and the assumptions made for incorporating them into the logistic regression models developed are summarized below:

• Μ1—Deep energy renovations: Interventions aimed at significantly upgrading the energy efficiency of the building envelope on a scale where the energy performance of the buildings constructed before 2010 will be comparable to the houses built in the country after 2010 based on the specifications of KENAK. The evaluation of the measure is performed through the variable DS018, considering that all renovated dwellings have a performance like those constructed in the period 2012–2016 (category 7).
• Μ2—Shallow energy renovations: Interventions aimed at achieving moderate upgrades in the energy efficiency of the building envelope at a scale where the buildings constructed before 1980 acquire an energy performance similar to that of buildings constructed in the period 1981–1995 and residential buildings constructed in the periods 1981–1995 and 1996–2005 obtain the energy performance of the buildings of the immediately following time periods, namely 1996–2005 and 2006–2011, respectively. The evaluation of the measure is performed again through the variable DS018, considering that the category of the renovated dwellings is improved according to the above-mentioned assumptions.
• Μ3—Modernization of heating systems: From the econometric analysis carried out in Section 5.2, it was found that houses that use oil and LPG stoves, electric appliances, and in some cases air conditioners (which usually do not cover the total area of the house) as their main heating system are positively correlated with increased odds of energy poverty. In the context of this analysis, the heating systems in question are upgraded or replaced so that the households living in these buildings can obtain energy services similar to those received by households living in residences having a central heating system. More specifically, the evaluation of the measure is performed through the parameter DA08, considering that the households with heating systems associated with high energy poverty levels start to use a central heating system with diesel oil.
• Μ4—Combined application of the measures M1 and M3. The evaluation of the measure is performed through the parameters DS018 and DA08.
• Μ5—Combined application of the measures M2 and M3. Again, the evaluation of the measure is performed through the parameters DS018 and DA08.
• Μ6—Provision of a subsidy of 750 EUR/year to households with an annual income below EUR 20,000. The financial assistance provided corresponds to the maximum amount of the space heating subsidy provided by the Greek State in 2021 for low- and middle-income households. It has been incorporated as additional income in the models used. So, the evaluation of the measure has been performed through the parameter HH095th.
• Μ7—Provision of a subsidy of 750 EUR/year to energy-poor households. The measure considers the targeted provision of the subsidy considered in M6 exclusively to energy-poor households based on the energy poverty indicators considered. It has been incorporated as additional income through the parameter HH095th in the models used.
• Μ8—Provision of a subsidy of 400 EUR/year to energy-poor households. The subsidy of M7 is adjusted to the mean value of the space heating subsidy provided by the Greek State in 2021. Again, it has been incorporated as additional income through the parameter HH095th in the models used.

Table 5. Reduction in energy poverty levels in Greece measured with three alternative indicators due to the implementation of various policy measures.

Policy Measures	NEPI	HBS-4b	LIHC
Μ1—Deep energy renovations	59%	77%	99%
Μ2—Shallow energy renovations	22%	37%	72%
Μ3—Use of modern heating systems	17%	15%	13%
Μ4—Combined application of measures M1 and M3	69%	84%	99%
Μ5—Combined application of measures M2 and M3	36%	49%	79%
Μ6—Provision of a subsidy of 750 EUR/year to households with an annual income of less than EUR 20,000	22%	22%	9%
Μ7—Provision of a subsidy of 750 EUR/year to energy-poor households	15%	15%	6%
Μ8—Provision of a subsidy of 400 EUR/year to energy-poor households	9%	7%	3%

shows the estimated reductions in energy poverty for each of the measures considered based on the three main energy poverty indicators used in this analysis. The results confirm that deep energy renovations are of paramount importance for structurally addressing the problem of energy poverty. Combined with upgraded heating systems, these interventions could reduce energy poverty by 69–99%. Shallow renovations paired with heating system upgrades also show significant potential, though they may not fully meet the APCEP’s 2030 targets. Upgrading heating systems alone, without improving building insulation, has a limited effect. While the subsidy schemes examined do not drastically reduce energy poverty levels, they mitigate the depth of energy poverty and improve the quality of energy services received by households. Also, if the subsidy is high enough and provided to vulnerable households, the effectiveness of the measure can be improved considerably. However, it is important to note that the models may underestimate the subsidies’ impact, as they were modeled as additional income rather than a targeted intervention specifically for improving heating conditions.

The results, though dependent on the predictive capabilities of the models used, provide a clear picture of the effectiveness of the planned measures. The measures aimed at structurally addressing energy poverty (M1–M5) improve to a larger extent the HBS-4b and LIHC energy poverty indicators compared to the NEPI. This is mainly attributed to the fact that these two indicators apply broader criteria to define energy-poor households and thus capture more households close to the energy poverty line. Economic measures (M6–M8) have a relatively smaller impact on the LIHC indicator, since the subsidies are modeled as additional income rather than direct reductions in energy costs. This limitation, which to a lesser extent may also affect the results derived by the other two models, potentially underestimates the effectiveness of the financial policies examined. Future research should consider including energy expenditures as an independent variable to improve the evaluation of targeted subsidies for energy-poor or vulnerable households.

In general, the results of the analysis indicate that decision-makers should place greater emphasis on policies aiming at energy upgrading of the buildings’ envelope in conjunction with the modernization of heating systems. This will substantially contribute to both decarbonizing the buildings’ sector and tackling energy poverty. As subsidies improve energy poverty indicators to a much lesser extent, the economic resources available for this purpose should be directed more specifically to the most vulnerable households, thus allowing for an increase in the amount of subsidy provided per household. In the long run, subsidies should be gradually limited, and a larger portion of available resources should be directed to the renovation of the building stock and the upgrading of heating systems.

6. Conclusions

As in several other developed countries, energy and environmental policies in the Greek residential sector face a triple challenge: (i) decarbonizing the sector to meet climate targets; (ii) developing appropriate climate adaptation strategies, including strengthening the resilience of the building stock; and (iii) improving energy services for citizens and alleviating energy poverty. This research contributes to bridging the gap between climate mitigation policies and the need to improve the energy services received by households, with a focus on energy poverty. It examines different indicators for measuring energy poverty, analyzes its determinants, and develops statistical models for identifying energy-poor households. These models also provide a framework for evaluating the effectiveness of various policies and measures aimed at tackling energy poverty, many of which are integral components of the national plans for climate mitigation.

For Greece and other countries in Southern and Eastern Europe, these issues are particularly important, as much of the building stock exhibits low energy performance, and the lingering effects of the economic crisis during the 2010s continue to compound the challenges posed by global health and energy crises. Specifically, the problem of energy poverty was explored through ten different subjective and objective indicators, with some proposed in the international literature and policy documents, including the official national energy poverty indicator, and others developed as part of this research. Using the 2021 microdata of the Household Budget Survey, energy poverty levels in Greece were estimated to range from 8.4% to 19.6% based on the seven objective indicators considered. The official energy poverty indicator was estimated at 12.9%, considerably lower compared to energy poverty levels estimated with the indicators used in other countries (e.g., in the United Kingdom). This clearly shows that Greece uses quite strict criteria for classifying households as energy-poor, and the problem of energy poverty might have a wider social basis. Moreover, decision-makers need to be aware that energy poverty indicators, by design, capture different aspects of the problem, and the selection of the appropriate indicator (or set of indicators) is of paramount importance for monitoring energy poverty and developing effective mitigation strategies. The use of more simplified indicators, which are based on the available income and energy performance of the dwellings, as well as appropriately adjusted to incorporate additional parameters such as housing costs and/or reasonable living costs, may be more suitable for capturing an extremely complex phenomenon, which should however be monitored at the national, regional, and local levels based on parameters that are easily available to public authorities. With such an adjustment, it is possible to monitor the problem more effectively, thus covering also the rising issue of summer energy poverty.

As energy poverty is a multidimensional problem, understanding its determinants is crucial for developing effective mitigation policies. This study used the above-mentioned dataset to develop logistic regression models that link the probability of a household being energy-poor to specific household and dwelling characteristics. These models serve two purposes: to identify energy-poor households based on the independent variables included and to provide a mathematical basis for evaluating the effectiveness of various policies and measures implemented or planned to address energy poverty. Three alternative logistic regression models were developed based on three different objective energy poverty indicators. All models demonstrated high accuracy and predictive power.

The models highlighted key determinants of energy poverty, namely decreased income of households, the poor energy performance of the building envelope (strongly related to building age), a larger household size, the presence of elderly and/or unemployment persons in the household, a larger dwelling area, the use of oil or LPG stoves and electric appliances for space heating instead of central heating or other modern systems, the climate with northern and mountainous areas to experience higher levels of energy poverty, increased housing costs (rents or mortgages), and living on detached or semi-detached houses. All these form the basis for developing targeted policies with a view to mitigate energy poverty levels nationally or for specific population segments.

The evaluation of the various policies implemented or examined to reduce greenhouse gas emissions in the sector and/or tackle energy poverty levels clearly shows that deep energy renovations, combined with heating system upgrades, can substantially reduce energy poverty levels (by 69–99% in Greece). Shallow renovations, when combined with heating system upgrades, also help alleviate the problem but to a lesser extent and may not be sufficient to meet the ambitious targets adopted by countries for reducing energy poverty levels by 2030. Upgrading heating systems without improving the building envelope only lifts a small portion of affected households out of energy poverty. Finally, while the subsidy schemes examined do not significantly reduce energy poverty levels, they do help mitigate the severity of energy poverty and improve the quality of energy services provided to households. In addition, the targeted provision of higher subsidies to vulnerable households can significantly improve their effectiveness. Consequently, the energy upgrade of the existing building stock is of paramount importance for decarbonizing the residential sector and tackling energy poverty. As these interventions present also a significant greenhouse gas emission reduction potential, a larger portion of available resources should be directed to the renovation of the residential buildings in conjunction with the upgrading of heating systems. At the same time appropriate policies should be implemented to ensure the participation of the most vulnerable households (e.g., those identified as energy-poor by various indicators) in such energy upgrading programs. This type of analysis could also be used to develop more stratified and classified strategies for specific groups of energy-poor households, but this requires the development of new logistic regression models based on those specific sub-populations.

Indisputably, the presented analysis has several limitations. It is based on the microdata of the HBS for the year 2021, so it will be particularly useful to check whether the findings of the study are confirmed or differentiated over a longer period. Also, changes in the structure of the logistic regression models developed by incorporating additional parameters may further improve their predictive capacity and effectiveness. As already mentioned, including energy expenditures as an independent variable to the models developed may improve the evaluation of targeted subsidies for energy-poor or vulnerable households. In this context, enriching the HBS with a small number of questions on the energy performance certificate of the dwellings, the existence of renewable systems for electricity generation, etc., would be particularly useful. Furthermore, limitations in capturing informal or undocumented energy practices by the HBS may also have influenced the results. To filling this gap, the combination of quantitative models like those presented in this paper with qualitative approaches that explore in more detail subjective coping strategies (see, for example, the analysis presented in [57]) may improve the level of information provided to policy-makers. Lastly, the whole analysis is focused on space heating even though the various indicators considered take into account the total (real or required) energy expenditures of the households. Summer energy poverty is an important dimension of the problem, and future analyses should incorporate it more systematically.