We first present a temporal analysis of the ET of the EU over the period 2000-2015, and then propose a typology of the 28 member countries in relation to the three components of the ET. The additional variables described in the previous subsection will be used to enrich the description of the homogenous classes of EU countries.
4.1 Trajectory of the EU-28 energy transition
A methodological sequence of two data analysis methods [47, 48, 45] was used to group the sixteen years into homogeneous classes according to the ET components of the EU. More precisely, Hierarchical Ascendant Clustering (HAC) was used on the significant factors of the Principal Component Analysis (PCA) of annual components of the ET development.
The first two axes of dispersion of factors as revealed by the PCA form the first factorial plane, which summarizes more than 99.76% of the information regarding the evolution of the ET components of the EU over the period 2000-2015. Figure 2 illustrates representations of the components of the energy transition and years projected into the first factorial plane.
The first factorial axis opposes the early period 2000-2006, characterized by high PEI and GHGE, with the later period 2011-2015, characterized by a high SREC. Figure 2 shows the grouping of the closest years according to the first component: these groups are materialized by geometric shapes. Years with the same shape have common energy transition characteristics.
Using an HAC with the Ward criterion[6] allows us to distinguish three homogeneous sub-periods. Table 2 summarizes the main results and profiles of the EU energy transition over the three sub-periods selected from the cut in the three classes of the hierarchical tree.
The first period, comprising the six first years, 2000-2006, is characterized by high PEI, high GHGE, and low SREC.
The second class, which groups together the four succeeding years of the middle period, 2007-2010, is considered as a homogeneous class, which means that none of the three ET components on this sub-period differs significantly from the average of these components over the overall period. It can be considered as an adaptation phase, concomitant with the adoption of the climate energy package in 2008.
The ET characteristics of the last class, constituted by the last five years of the period, are opposed to those of the first class. This third class is characterised by a high share of RE in final consumption, lower energy intensity and lower GHG emissions per capita.
4.2 Energy transition of the 28 EU member countries
To better analyse and understand the evolution of the development of the ET of the 28 EU countries, we carried out an evolutionary data analysis on the three sub-periods.
The approach adopted relies on a combined use of exploratory methods of evolutionary data analysis that take into account the characteristics of the countries in terms of GHGE, PEI and SREC, as well as their evolution over each sub-period. According to the similarity of these three components, we can establish a typology of the 28 EU countries. The evolution of the countries is thus studied by a Multiple Factor Analysis (MFA), based on a weighted analysis of the principal components of all the data. The MFA [49-51] allows the simultaneous exploration of several multidimensional data tables, and it applies more particularly to time series data.
This evolutionary analysis is especially designed to study individuals (i.e. countries) characterized by a number of groups of the same variable (i.e. the components) measured at each different moment in time. The MFA highlights the common structure of a set of groups of ET components observed for the same 28 countries. The primary interest of this method is that it enables us to carry out a factor analysis in which the influence of the different groups of ET components is a priori equilibrated. An HAC was then used on the significant factors of the MFA in order to characterize homogeneous classes of countries relative to the evolution of the three ET components.
Table 3 summarizes the results of the three partitions of the EU-28 countries into four homogeneous classes as carried out over the three sub-periods, and provides the characterization of the classes[7].
Note first that even though the temporal evolution of the EU’s ET development identified three homogeneous sub-periods with distinct profiles, the three evolutionary analyses of the 28 EU countries show a certain stability in country trajectories, with all typologies having four homogeneous classes and almost identical profiles and anti-profiles. With the exception of Slovenia, Denmark and Lithuania, which had different paths in terms of ET development, the other 25 countries had an almost identical course throughout the period 2000-2015.
The first class includes six countries over the entire period, namely Austria, Croatia, Finland, Latvia, Portugal and Sweden, and is also made up of Slovenia in the first sub-period, while Denmark and Lithuania join the class in the sub-period 2011-2015. A high share of RE in final energy consumption over the three sub-periods and low GHG emissions per capita over the 2000-2006 and 2011-2015 sub-period characterize the first class. This class gathers the most successful countries in terms of the ET. This class can be called the virtuous class for ET.
The energy balance sheets of countries in this class are characterized by significantly high shares of renewable energy and waste (non-renewable) and a small share of fossil and solid fuels in domestic energy consumption. Electricity generation from hydropower and renewable sources is rather high which contributes to a low level of GHGE. The share of solar photovoltaic in gross inland RE consumption is rather low.
In each year of the 2011-2015 period, distance to RE share target that is above the EU average, means that these countries have made greater efforts to meet or exceed their objectives and achieved their SREC target in 2012, 2014 and 2015. These eight countries were efficient throughout the 2011-2015 period in terms of reducing PEC, and achieved their PEC goal in 2012. With regard to the other two objectives, their performance is quite similar to the European average, with the exception of the year 2012 for which the PEC goal was reached.
The second class contains seven countries over the entire period 2000-2015: Romania, Slovakia, Poland, Hungary, Estonia, Czech Republic, Bulgaria. Lithuania also belongs to this class in the first two sub-periods only, while Slovenia joins in the sub-period 2011-2015. The characterization of the second class is stable over the three sub-periods. These countries have a PEI significantly higher than the average of the 28 EU countries. The other two components are no different from the EU averages. It is terms the low energy efficiency class. These are countries of the ancient eastern bloc, which still suffer from entrenched specialization in heavy industries driven by central planners. This results in a high level of PEI, i.e. a low efficacy in producing one unit of GDP.
This low energy efficiency comes also from the lower access to clean fuels and technologies for cooking. Having highly developed coal and nuclear energy sources, these countries are less dependent on energy imports. Indeed, the share of solid fuels in gross domestic consumption is significantly higher than the EU average. Coal and nuclear sources make a significant contribution to electricity production while the shares of natural gas and renewable (total and excluding hydroelectric) sources in electric power are rather low. The biomass and renewable waste sector provided an important share of RE consumption from 2011 to 2013, while the share of wind power lags behind the average of the EU countries.
Throughout the period, countries belonging to this class present a distance to their GHGE targets that is lower than the European average and made greater efforts to meet their RE consumption objectives in 2011. These countries, mainly developing countries, have made significant efforts to control their GHGs. Note that the targets assigned to them vary from + 4% for Slovenia to + 20% for Bulgaria. Countries in this class have generally achieved their GHG objectives in 2011 and 2012.
The third class contains eleven countries over all three sub-periods: Malta, Italy, Spain, France, Ireland, Greece, Germany, Cyprus, Belgium, United Kingdom and Netherlands. Denmark is also attached to this class in the first two sub-periods, while Slovenia is in it for the intermediate sub-period (2007-2010). The characterization of the class is stable over the period 2000-2015. This class gathers countries whose PEI and SREC are significantly below the respective averages of all the EU-28. We establish that the EU-28, projected a posteriori in each periodic analysis, is assigned to class 3 whatever the sub-period, meaning that the EU-28 has characteristics similar to those of class 3 with respect to the three components of the ET. We term it the energy efficient class lagging behind in RE development. These countries are diverse, gathering developed countries, which are relatively efficient in terms of production but not very aware and/or attentive to environmental concerns, and less developed countries (Malta, Greece and Cyprus) where the low PEI can be attributed to low industrialization.
These countries are the most energy efficient, but are also the most dependent on energy imports. They are highly dependent on fossil fuels, and in particular oil. Fossil fuel energy consumption and share of fossil fuel and oil (crude oil and petroleum products) in gross inland consumption are significantly higher than the average of the EU-28. Electricity production from fossil sources remains very high over the period 2011-2015, being mainly dependent on oil and gas sources. The shares of RE and hydroelectric in power generation are below the average for the EU-28. Compared to the EU average, the SREC is rather low in this class, this being explicable by the massive use of fossil fuels. However, this class is the leader in the development of new renewable sources: the shares of wind power, solar thermic and solar photovoltaic energy in the domestic consumption of RE are significantly higher than the European Union averages. On the other hand, biomass and renewable waste are less prominent in the RE energy mix.
In this class, the distances to targets are higher than the European averages for the three targets of the ET, reflecting insufficient efforts to achieve the objectives. In addition, we see that these countries never achieve their PEC objectives during the whole period and have not achieved their RE consumption objectives since 2012.
The fourth class consists solely of Luxembourg over the three sub-periods. Notably, Luxembourg is isolated in all the three sub-periods, differing from other EU countries over the three periods by having a significantly higher GHGE per capita than those of the EU over all three sub-periods, and a lower SREC from 2007. We name it the non-virtuous class as regards the energy transition. Contrary to results widely established in the literature (see supra), the positive link between economic growth and the use of renewable energy does not prevail in Luxembourg.
Since 2013, Luxembourg has been highly dependent on energy imports. Its characteristics are those of a rich country with both primary and final energy and electricity consumption per capita significantly higher than those of the EU population on average. The share of crude oil and petroleum products in gross domestic consumption has been high since 2013, and natural gas is widely used in electricity production. Note that Luxembourg is a small and densely populated country, with a high density of road freight and many “cross-border workers”, which contributes to its GHGE. In particular, fuel sales to non-residents have increased significantly, by 165% between 2000 and 2013.
Finally, Luxembourg, which seemed to be “the least virtuous” country, is still at the level of the European average in terms of distances to the objectives that have been set for it. This clearly indicates that, despite an unfavourable situation, Luxembourg has made considerable efforts to reach its objectives, albeit insufficient to achieve its three targets for 2020. It should be emphasized that over the whole period 2011-2015, Luxembourg has met its PEC objective.
The results we have revealed are not in line with the ambitions and commitments of the EU. Although progress has been made, we note that the EU Member States’ performances fluctuate greatly from one year to the next, and that no major trend toward achieving the SREC and PEC goals is emerging. In particular, the large western countries in Class 3 are significantly behind in achieving their objectives, notably Belgium, France, Germany, the Netherlands and the United Kingdom, as well as Ireland with regard to the development of renewables and the reduction of primary energy consumption. Germany, Ireland and the Netherlands are also lagging behind in reducing GHG emissions.
4.3 Discriminating effects of themes on the trajectories of the energy transition of the 28 EU countries
As predictive model, we choose here Discriminant Analysis (DA) which is a modeling method of decision-making. DA is a multidimensional method; it allows one to highlight the links existing between a target qualitative variable which can help explain, in this case, the variable synthesis of energy transition into several modalities corresponding to the previously discussed classes (three classes for the temporal analysis and four for the spatial analysis) and a set of continuous explanatory variables relating to a homogeneous theme. Four explanatory themes were considered: economic performance, trade performance, policy mix design and innovation system.
The DA method is a special PCA; it produces discriminant factors which are linear combinations of the explanatory variables and establishes graphical representations on discriminant factorial planes making it possible to distinguish the classes, and then explain their respective positions[8].
It has two main objectives: the first is descriptive and consists in determining which of the explanatory variables are discriminating. The second objective is predictive or decision-making and is concerned with classifying new anonymous explanatory data in these known classes using the discriminant linear functions established previously. Our goal is a search to identify themes -homogeneous sets of explanatory variables -, which discriminate between the classes presented in sections 4.1 and 4.2.
4.3.1 Discriminant effect on the temporal typology
Table 4 below summarizes the main results of the four models of DA[9] according to each theme. For each theme, the explanatory variables that discriminate between and separate each of the energy transition sub-periods characterized by the HAC are mentioned.
All these models are globally significant. Indeed, for each model, the critical probability or p-value Pr> F of the Wilks’Lambda statistic is less than the significance level of 1%[10]. We can therefore conclude that economic and trade performance, policy mix design and innovation system themes have a significant effect on the three sub-periods of the EU energy transition.
Thus, among the three explanatory variables of the economic performance theme, only the GDP growth rate is not discriminating. The other two variables perfectly differentiate the three sub-periods of the energy transition of the 28 EU countries.
The first significant discriminating factor restates 81.38% of the discriminating power of the model. It separates and opposes the third period 2011-2015 characterized by a high unemployment rate and GDP per capita to the first period 2000-2006. As for the second discriminating factor (18.62%), it distinguishes the third period 2011-2015 characterized by a high unemployment rate, which opposes the second period 2007-2010[11].
Our results suggest that during the second period coinciding with the economic and financial crisis, the economic performance of the 28 EU countries did not have a significant impact on environmental performance. More generally, we find:
1) The economic performance’s model shows that the GDP growth rate of the EU28 over the period 2000-2015 does not induce significant effects on the trajectory for energy transition in the 28 EU countries. There seems to be a decoupling between economic growth and environmental performance measured from the three targets. The two other variables perfectly differentiate the three sub-period of the ET of the 28 EY countries.
2) Concerning the trade performance’s model, only the energy dependence is not discriminating. EU energy dependence has slightly increased over the period from 54.8 in 2000 to 56.1 in 2015. However, its evolution has not had any significant effects on the trajectories for energy transition in the 28 EU countries. On the other hand, the improvement in energy terms of trade over the 2011-2015 period, probably related to the development of RE, led to better environmental performance.
3) Regarding the model of discrimination according to the policy mix design, all the variables of this theme are discriminating. The first discriminating factor (90.74%) opposes the first period 2000-2006 characterized by a high rate of environmental taxes, to the third period 2011-2015, which is distinguished by a high rate of public research and development expenditure. The second discriminating factor (9.26%) separates the third period 2011-2015 characterized by high-energy taxes and environmental taxes and opposes it to the second period 2007-2010. These results show the effectiveness of the public policies adopted at European level to meet the 2020 climate targets; the energy taxes and public expenditure in R&D have strongly contributed to improve environmental performance over the period 2011-2015.
4) As for the three variables introduced into the innovation system’s model, all are discriminating and well separating the three periods. The discriminating factor (84.12%) opposes the third period 2011-2015 characterized by a high number of environmental, a high share of environmental technology patents and high R&D expenditure, to the first period 2000-2006. The preoccupation of global warning has oriented research and development towards environmental issues. The evolution of the innovation system is a significant driver of the European energy transition.
4.3.2 Discriminant effect on the spatial typology over the period 2011-2015
As we showed in section 4.1, the energy transition is on track for the 2011-2015 period that is why we focus our research on the drivers of the EU energy transition on this sub-period. We use DA[12] models to see how each of the four thematic makes it possible to distinguish different classes of EU countries grouped according to their energy transition performance. Table 5 presents the overall results of the discriminant analysis models for each theme. Note that only two models, economic and trade performance are significant and therefore discriminating. Indeed, the Wilks Lambda of these two models are less than the significance level of 5%. In contrary the innovation system and policy mix design have no effect on the energy transition development of the EU countries over the 2011-2015 sub-period.
Thus, for the 2011-2015 sub-period, among the three explanatory variables of the economic performance theme, GDP growth and unemployment rates are not discriminating. The first discriminating factor (72.61%) is statistically significant and opposes Luxembourg (class 4, non-virtuous class as regards the energy transition), characterized by a high GDP per capita, to the low energy efficiency class (class 2 grouping Eastern and Central European countries). So, national disparities in terms of standard of living induce contrasted environmental performances. This finding contradicts the results of [53, 16, 17], since we show that a high level of development does not necessarily lead to virtuous growth. Nevertheless, it should be noted that Luxembourg is a major transport node and has the highest road freight density in Europe. The share of CO2 emissions from transport in total fuel consumption is the highest in Europe; it reached 66.8 % in 2015 compared to 31.1% for the European average. Furthermore, Luxembourg is also a small country with low relief, which considerably limits the development of RE.
In the trade performance model, only the energy dependence variable is discriminant with a significance level less than 5%. The first factor (75.24%) is significant and therefore discriminating. It opposes Luxembourg characterized by a high rate of energy dependence to class 2. Luxembourg has no domestic energy resource and is highly dependent on its energy imports, mainly oil and gas. On the other hand, the Central and Eastern European countries have developed national energy production based on coal and nuclear energy sources, so they are less dependent on energy imports.
DA models highlight that temporal and spatial determinants of the energy transition with respect to the three targets defined by the 2020 European Climate Energy Package differ. Indeed, while the four themes selected make it possible to discriminate the trajectory of the energy transition of the European Union over the period 2000-2015, only the themes relating to the economic and commercial performances explain the contrasted environmental performances of the countries over the period 2011-2015. More specifically, only two variables, namely GDP per capita and energy dependence are discriminant. Policy mix design and innovation system do not contribute to discriminate the four classes of the typology. Our findings provide strong evidence that policy mix design and innovation system have been particularly effective in promoting sustainable development in the last sub-period, but national differences regarding these two themes do not explain the contrasting results observed at the country level.
Footnotes
[6] Generalised Ward’s Criteria, i.e. aggregation based on the criterion of the loss of minimal inertia.
[7] In order not to overload the article, we do not present the results related to illustrative variables but they are available on request.
[8] The DA is based on the normality of populations. The discriminant functions are linear if the matrices of variances and co-variances of these populations are equal; otherwise they are quadratic. All these conditions of application have been checked.
[9] The DA is based on the normality of populations. The discriminant functions are linear if the matrices of variances and co-variances of these populations are equal; otherwise they are quadratic. All these conditions of application have been checked.
[10] The value of Wilks' lambda varies between 0 and 1, the more it tends to 0, the best is the discrimination model (the class centers are well separated). A probability distribution of Fisher approximates the Wilks test statistic.
[11] Nevertheless, unemployment rates evolution is very different among European countries during the 2007-2017 period [52]. If Slovakia was at the top of the list of the highest unemployed countries before the crisis, it ranks eleventh in ten years. On the other hand, it is mainly the countries of Mediterranean Europe such as Greece, Spain, Cyprus or Italy, which are at the top of the ranking of the countries most affected by unemployment ten years later –France, is at the sixth positon-. Conversely, Germany (-56%), which has had its lowest unemployment rate since reunification at 5.7%, Hungary (-49%) and Poland are the three states to have known the biggest decline.
[12]DA is based on normality of the variables in the populations. The discriminant functions are linear if the matrices of variances and covariances of the variables are equal, otherwise they are quadratic. All these application conditions have been verified except for class 4.