Determinants of Wind Power Energy in the United Kingdom
Verena Ettlinger
Student Semester 2016 Bachelor’s Level
Degree Thesis in Economics 15 ECTS
Abstract
Today’s energy revolution proposes a pathway from conventional energy sources to more alternative and renewable energy sources in order to sustain energy supply, slow down climate change and promote political independence. As a result of this revolution most countries have to implement new technologies and ways to generate green energy and therefore ensure future energy supply. Since the late 20th century, the United Kingdom focuses on the energy source wind due to its perfect wind conditions on the Island. My study provides statistical tests to examine significant determinants that might influence the permission process for new windmill construction sites. I use the probit model to calculate the effects caused by changes in the installed capacities, employment rates, energy consumption levels, income levels and population density on the probability of attaining construction permission for new wind farms.
My analysis shows that only capacity levels might play a significant role regarding the permission for new wind farm projects. The other variables play minor roles, meaning that employment rates, energy consumption levels, income levels and population density do affect the permission process of windmills but their impacts are not great enough to cause substantial changes.
Acknowledgement
This paper and my studies at Umea Universitet would have never been possible without the fantastic help and support from Mikeal Lindbäck. I want to thank him for giving me the opportunity to study in Umeå in order to fulfil all requirements for my dream Master studies. I would also like to thank Linda Lindgren for her help and patience regarding all the paper work for my credit transfer from Germany.
Special thanks goes to my supervisor Jurate Juraite-Kazukauske for her help, support and inspiration throughout the entire time. During the thesis work, she has given me help with the structure, data and especially with the Statistic program STATA and I am grateful for her fast responds on emails and her daily reachability.
Big thanks also goes to my extraordinary friend and classmate Christoffer Johansson. He has always been there for me during my studies, gave fantastic advices, cheered me up and pushed me to reach my goals and meet my deadlines.
Finally, I would like to thank my family and friends for proofreading and valuable support and comments.
Umeå, May 2016 Verena Ettlinger
Table of Contents
1. Introduction ...1
1.1 Background ...1
1.2 Research question ...3
1.3 Purpose ...4
2. Theory ...4
2.1 Physical, social and economic determinants ...4
2.2 Rural vs. urban locations ...5
2.3 Employment ...6
2.4 Gross Domestic Product (GDP) ...6
2.5 Investment ...7
3. Methodology ...8
3.1 Empirical Strategy ...8
3.2 Diagnostic methods ...9
3.2.1 Mis-specification ...9
3.2.2 Multicollinearity ...9
3.2.3 Hetero- and Homoscedasticity ...10
3.2.4 Goodness of fit ...10
3.2.5 Marginal effects ...11
3.3 Variables of the probit model ...11
3.4 Data collection ...12
3.5 Probit regression model ...13
3.5.1 First stage ...13
3.5.2 Second stage ...15
4. Results ...18
4.1 Descriptive analysis ...18
4.2 Probit regression ...20
4.2.1 Probit model with robust standard errors ...20
4.2.2 Marginal effects ...22
4.2.3 Probit regression excluding Scotland ...23
5. Discussion ...24
5.1 Independent continuous variables ...25
5.1.1 Capacity ...25
5.1.2 Employment ...26
5.1.3 Income levels and GDP ...27
5.1.4 Population...27
5.1.5 Energy consumption ...28
5.2 Significant regions and their variables ...29
5.3 Theoretical approaches ...31
6. Conclusion ...32
7. References ...33
8. Appendix ...37
8.1 Table 10 – Multicollinearity ...37
8.2 Table 11 – Mis-specification - Linktest ...37
8.3 Table 12 – Energy consumption NUTS2 ...38
8.4 Table 13 – NUTS2 ...39
8.5 Table 14 – Employment NUTS2 ...40
8.6 Table 15 – Income of households NUTS2 ...41
8.7 Table 16 – GDP NUTS2 ...42
8.8 Table 17 – Population NUTS2 ...43
8.9 Table 18 – Goodness of fit ...44
8.10 Table 19 – Probit regression excluding Scotland ...45
1
1. Introduction
1.1 Background
Wind power – one of the most rapidly developing energy technologies both in Europe and globally, with great potential to promote today’s energy revolution by leading the way from conventional energy sources to alternative energy sources. It is common knowledge that we have to reduce our CO₂emissions in order to protect our environment and stop global warming.
Therefore, it is indispensable that our energy production has to change from conventional sources of energy including oil, gas and coal to alternative sources like wind, water, sun or biomass. In addition, energy is one of the most important component of economic infrastructure and plays a crucial factor in terms of policy as well as power struggle in and between countries.
In 1981, the US Congressional budget Office published that the first oil crisis in 1973 cost the US economy $350 billion (US Congressional Budget Office, 1981). In 2000, Greene and Tishchishyna (2000) stated that the oil price movements between 1970 and 2000 induced up to
$7 trillion in costs on the US. These and several other studies substantiate that oil prices affect economic activity negatively. A rise in oil prices may result in deteriorating terms of trade for oil-importing countries, an increased money demand or generating inflation as well as a change in production structure and an impact on unemployment (Lardic and Mignon, 2006). This effect, known as the “oil-GDP effect”, becomes even more important in situations where industrialised countries depend on the import of fossil fuels from politically unstable areas at unpredictable and higher prices. Furthermore, conventional energy sources are limited in supply and will eventually be completely used up in the future. The Statistical Review of World Energy predicts that our coal reserves will be exhausted by 2128 (112 years), natural gas by 2069 (53 years) and crude oil reserves already in 2067 (51 years) (Knoema, 2015).
Due to these and many other reasons, European countries started to implement and promote the development of renewable energy sources in the public and private sector. Within the last 26 years, the European Union increased its primary production of renewable energies by more than 174%, and is targeting 20% final energy consumption by 2020 (Eurostat, 2016). The EU member countries try to reach their own targets by producing energy from a wide variety of sources including wind, solar, hydro, tidal, geothermal and biomass. Although, each energy source has its own advantages and disadvantages, wind power is rumoured to be one of the favoured energy forms for the future. Its development has been rapidly increasing over the last 20 years resulting in the second largest contributor of electricity in the EU from renewable
2
sources by 2014 (European Commission, 2016). A distinct benefit of wind energy is the independence from fluctuating fuel prices and consequential uncertainties that may imply considerable risk for future generation costs. Instead, wind turbines are capital-intensive compared to conventional fossil fuel fired technologies meaning that approximately three- fourths of the total cost are related to upfront costs. However, these costs have continuously reduced over the past years due to technological improvements for instance larger turbines and enhanced cost-effectiveness. The International Energy Agency (IEA) even expects new wind power capacity to be cheaper than coal and gas in 2030 (Krohn et al., 2009). Besides the economic and political benefits of this energy source, wind is also widely available throughout the entire world. The production process is environmentally friendly and is conducted without the direct emission of greenhouse gases, other pollutants or the consumption of water (International Energy Agency, 2016). Therefore, wind power distinguishes clearly from other conventional and alternative energy forms and plays a central role in the energy mix of several countries.
One of the countries that requires wind power in order to achieve its renewable energy target and move towards a low-carbon economy is the United Kingdom. According to the European Environmental Agency, England, Wales, Scotland and the Northern Ireland have the potential to produce 11% of the total wind energy in the European Union (Bassi et al., 2012). In 2014, 19.1% of UK’s electricity was generated from renewable energies, showing a steady increase.
Wind energy accounted for 20.3% and scientists predict even higher numbers for the upcoming years due to the abundant wind resource availability on the island (MacLeay et al., 2015). These future prospects are also reflected in the past and current investment trends with a stable market situation at around 10€ billion/year investments until 2015, followed by a gradually increasing share of investments (Krohn et al., 2009).
In total, 927 onshore wind energy projects containing 5226 turbines providing 8577MW onshore capacity were located and operating in the four UK’s countries in the end of 2014. The generated capacity and electricity produced was around 17 terawatt-hours (TWh) equals the electricity consumption of more than 4 million households (RenewableUK, 2014). Therefore, it is obvious that wind energy makes a significant contribution to the UK’s energy supplies, reduces CO₂ emissions and affects the UK’s national and local economies respectively.
3 However, the actual procedure of
developing wind power in the United Kingdom is much more intricate than illustrated in the previous paragraphs.
Figure 1 shows the geographical distribution of wind farm projects in the United Kingdom and their status:
consented, planning, under construction and operational. At first glance, it seems like a map full of random located coloured dots, each symbolizing a wind energy project. But the truth is, that every single point required a long, intense and partial complicated planning process including multiphase voting
procedures. Although, renewable energy projects are highly recommended, they are still not without impacts and its benefits usually accrue globally rather than locally (Cowell, Bristow and Munday, 2011). Furthermore, the success of wind energy projects depends predominately on geographical, environmental as well as economic circumstances within certain areas. It is important to be aware of the main determinants in order to choose the most profitable, environmentally friendly and sustainable location for the construction of new windmills. Hence, this multitude of influencing factors brings us to the actual purpose of my study.
1.2 Research question
There exist different papers and research theories about possible relationships between environmental, geographical and economic factors and the development of wind energy in general. However, no paper yet examined these relationships focusing on all wind farm projects in the United Kingdom, which envelops several questions. The most relevant question concerns the different factors determining the development of wind energy in England, Wales, Scotland and Northern Ireland. Which factors are more significant and how strong is their effect on the permission process for new windmills in the UK? Furthermore, do my results conform with previous studies and outcomes regarding the development of wind energy?
The Telegraph, 2012
Figure 1: Wind farm projects in the United Kingdom
4 1.3 Purpose
The aim of this study is to investigate factors that determine the successful approval of wind energy projects and thereby the development of wind energy in the United Kingdom. Within this paper, I will focus on the location of the power farms, subdivided by the Nomenclature of Territorial Units for Statistics (NUTS1 and NUTS2), and their specific geographical, environmental and economic data. Examining the relationships between the different variables will give me information about the significance of factors determining the permission process for new construction sites of windmills. With this study, I want to find out whether it is possible to predict the probability of operating windmills in certain areas depending on different variables.
2. Theory
The following paragraphs describe the theory and past research which envelop wind energy development and power generation. In general, people tend to think that wind conditions, strength and direction, are the most important factors to consider when choosing the right area for wind farms - but finding a suitable location to attain construction permission involves much more than just wind speeds. We can group the most relevant determinants into three main categories: physical factors, social factors and economic factors (Clint, 2011).
2.1 Physical, social and economic determinants
Physical constraints comprise the wind farm’s overall productivity and the surrounding environment. Landscape characteristics including hills, forests or lakes have to be considered due to negative and limiting effects of flora and fauna that could impact the power generation.
There exist, for instance, a wide variety of papers and protocols that address the interaction of wind energy and birds with different results and outcomes (Anderson, Morrison, Sinclair and Strickland, 1999). Another issue are obstacles or valleys that might change the speed and direction of wind flows. According to Chowdhury et al. (2010), so-called wake effects will also attribute to the loss in availability of wind energy. Their study examines the optimal placement of turbines in a wind farm in order to maximize energy generation. Generally, the category of physical constraints determines the quality and quantity of wind that can be found in chosen areas (Clint, 2011).
5
Social factors, in contrast to physical factors, represent the politics of generating wind energy.
Important determinants are policy frameworks, public engagement and participation, noise levels and visual impacts, just to mention a few (European Wind Energy Association, 2009).
These factors affect the permission process for project approvals and became a subject of contested debates in several countries, especially in the UK. McLaren Loring (2007), Wolsink (2000) as well as Ladenburg und Dubgaard (2007) argue that landscape intrusions and local ecosystem impacts caused by wind farms are the main reasons for public concerns and oppositions. Furthermore, the most popular phenomena amongst local disputants is called NIMBY (not in my backyard) behaviour, emphasizing the relevance of proximity on public attitudes on proposed wind energy projects (Van der Horst, 2007).
The final group categorizes the economics of wind power. These determinants comprise all economic factors that affect the location and size of a wind farm. For instance, site accessibility, proximity to the grid or the local infrastructure that in turn depends on population, employment rates and wages, Gross Domestic Products or consumer behaviour (Clint, 2011).
Furthermore, Bergmann et al. (2007) investigate the different preferences of renewable energy development in rural and urban areas on a choice experiment in Scotland. Previous studies from Narayan and Smyth (2008), Apergis and Payne (2009) as well as Blanco and Rodrigues (2008) describe the relationship of renewable energy development on the economic variables growth and employment. However, the major parameter in terms of wind energy development is most likely investment. In order to get a deeper understanding of related determinants and its impacts on wind power, the following paragraphs will give a short review of the most important theoretical outcomes from the past.
2.2 Rural vs. urban locations
Renewable energy projects, especially wind farms, are usually built in rural areas most directly affecting population in rural communities. Certainly, this is not always the case. Due to this and the multifaceted environmental and economic side effects of energy projects, the relative valuation of these impacts is of special interest and has to be taken into account. The choice experiment in Scotland revealed the different preferences of urban and rural inhabitants, focusing on landscape change, wildlife, job opportunities, income and other variables. In general, people from rural areas are more likely to support wind projects that create new job opportunities, have low impacts on the environment and a small effect on the price attribute of
6
electricity. Preferences of urban residents, on the other hand, are not determined by creating new jobs but rather driven by environmental friendly projects that do not cause any harm to wildlife, air or water (Bergmann et al., 2007). Similar results regarding environmental impacts and preferences towards energy projects and their location can be found in the study of Ek and Persson (2014) that performed a choice experiment in Sweden with slightly different attributes.
2.3 Employment
The evolution of wind energy came hand in hand with the development of a cross-sectoral industry that centres around installation, maintenance and operation of wind farms. Blanco and Rodrigues (2008) used four different categories to group the employment sector for wind energy: first, manufactures and major sub-manufacturers, second, companies that generate and distribute electricity from wind energy, third, promoters of wind energy and fourth, centres that are specialized in wind energy activities (Blanco and Rodrigues, 2008). The reason why it is difficult to quantify the number of job directly coming from wind energy development is that each country has a different company profile which makes up the sector. For instance, the United Kingdom is one of Europe’s biggest wind farm operators but the construction of turbines takes place in other European countries like Germany, Denmark or Spain (Gamesa, 2016).
According to RenewableUK and their final balance sheet in October 2015, the wind energy sector contributed 8932 direct and 8677 indirect jobs to the total employment in the United Kingdom. Compared to the number of direct employment in 2010 of approximately 6000 full time employee, the United Kingdom experienced a year-on-year increase in installed capacity for wind farms and a similar developing trend in direct and indirect employment rates (RenewableUK, 2011). In conclusion, wind energy represents an attractive source of employment but the number and location of jobs created are not directly related to the site of the wind farm.
2.4 Gross Domestic Product (GDP)
The direct and indirect effects of wind energy development on GDP is similar to the deployment trend of employment described in the previous paragraph. Several recent studies and reports are consistent that the expansion of wind power has a crucial effect on growth rates. In 2012, the European Wind Energy Association (EWAE) published growth rates and facts about the overall impact of the wind energy sector on the EU economy. The observed growth rates in
7
2010 caused by the wind energy sector directly contributing to the EU GDP had been higher than the overall growth of EU’s GDP itself (European Wind Energy Association, 2012).
According to Apergis and Payne (2009) who examined the relationship between renewable energy consumption and economic growth by using a heterogeneous panel co-integration test, real GDP will increase by 0.76%, if renewable energy consumption increases by 1%. A similar study has already been conducted in 2006 resulting in a weaker outcome of 0.12-0.39% increase in GDP (Narayan and Smyth, 2008). Thus, the development of wind power is an important requirement for economic growth and likewise economic growth encourages the further expansion in the wind energy sector (Apergis and Payne, 2009).
2.5 Investment
The United Kingdom is one of Europe’s most attractive locations for domestic and foreign direct investments due to its low tax burden, transparent, flexible company law and cooperate governance, stock market strength and dynamic attraction as well as its proximity and concentration of high skilled employees (UK Trade&Investment, 2016). Being attractive for investors is beneficial in terms of technology, knowledge, innovations or local capacities and employment rates and might therefore be an important factor for developing wind energy in the UK (Antonescu, 2014). Most investments are usually concentrated in the richest parts of the country, areas with high income, good infrastructure and easy accessibility. According to Inward Investment Report 2013/14 from the UK, most inward investment projects (1496) were located in England with high concentration in London, the economic centre of the UK.
Followed by 122 projects in Scotland, 79 in Wales and 50 in Northern Ireland (UK Trade&Investment, 2014). Regarding this numbers, England and Scotland might be favoured over Wales and the Northern Ireland as new wind farm locations due to investment advantages.
In 2014/2015, the United Kingdom invested £855m in the industrial onshore wind energy sector and an even higher amount of capital was invested in 2013/2014 of approximately £1,1bn.
Nevertheless, current investments are not high enough to realize the transition to a low carbon economy and reach the 2050 target – an 80% cut in emissions (Department of Energy&Climate Change, 2011). Therefore, further actions have to be taken in the field of technological innovations as well as in the social and institutional context, in order to increase the share of wind energy and develop a sustainable long-run energy generation (Krewitt et al., 2007).
Wüstenhagen and Menichetti (2011) introduced the issue of strategic choices on renewable energy investments and represent investment as a function of risk, return and policy. Previous
8
papers state that appropriate policy packages are required to attract investors to integrate climate-change considerations into decision-making processes. This means, that countries have to improve their policy frameworks to effectively reduce the investor’s risk and correct externalities caused by wind energy projects. In general, the promotion of wind energy will result in direct and indirect investment impulses on the entire economy.
3. Methodology
3.1 Empirical strategy
The aim of my study is to investigate variables that might determine the probability of successful wind farm project applications focusing on the windmill’s characteristics, location and economic variables. Therefore, I need a model that is designed for binary dependent variables that forces the values to be either one or zero. In general, two variants of models are used to deal with binary outcomes, the probit and the logit model. According to Cameron and Trivedi (2009), both models have very similar fitted probabilities but the probit model predicts a higher log likelihood compared to the logit model. Furthermore, the probit model is easier to defend in terms of nonsense predictions and linearity (Söderbom, 2009). Hence, the most fitting option is the probit-model within Stata, a data analysis and statistical software. It is a nonlinear regression model that estimates the probability of a specific observation with multiple characteristics. The model consists of a dependent binary variable Y that can only take two values, one and zero, conditional on the regressors. The probit model with multiple regressors is given as
Pr (Y = 1 | X1,X2,X3,X4,X5,X6,X7) = Φ (ß0 + ß1X1 + ß2X2 + ß3 X3 + ß4 X4 + ß5X5 + ß6X6 + ß7X7)
where Xi represent the different regressors (independent variables), 𝛷 symbolizes a normal distribution function and ßi are the individual coefficients for each independent variable. The estimation of the coefficients is based on the method of maximum likelihood, which produces estimators with minimum variance. In combination with the cumulative functional form, the model predicts the effect of an estimated change in X on the probability of Y=1 (Stock and Watson, 2015).
9
In addition, some other measurements are desirable for the binary outcome model, including mis-specification, multicollinearity and hetero/homoscedasticity as well as the overall goodness of fit.
3.2 Diagnostic methods
This section describes the elimination of possible issues occurring during the calculation process. Furthermore, I will evaluate the fit of the model and introduce the importance of the marginal effects in a non-linear regression model.
3.2.1 Mis-specification
The estimation of the probit model is based on the cumulative distribution function (CDF), which might cause mis-specification in two ways. First, the argument of the CDF, that is a specific input in the function, could have the wrong functional form. Second, the CDF, usually normal distributed in the probit model, could be entirely wrong (Söderbom, 2009). In order to prohibit mis-specifications in binary choice models, several specification tests are explained and discussed by Cameron and Trivedi (2009) in their book about “Microeconometrics Using Stata”. Most of these tests are conducted under a null hypothesis and proceed either in the semi- parametric and informal way or in the parametric and formal way. In case of mis-specification, there will be two alternatives to choose from: either change the argument of the CDF or change the functional form of the CDF (Söderbom, 2009).
3.2.2 Multicollinearity
Multicollinearity is a disturbance in the data, usually caused by high inter-correlations or inter- associations among the independent variables. Other reasons can be an inaccurate use of dummy variables, inclusion of a variable computed by another variable or a repetition of similar variables (StatisticsSolutions). Thus, multicollinearity can result in several problems for instance larger standard errors, over- and underestimation of parameters or even changes in signs. Therefore, it is important to find a suitable solution if multicollinearity appears in the data. Williams (2015) analysed the problem of correlated variables and suggests increasing the sample size, creating a scale from the X’s, using joint hypothesis tests or simply dropping the offending variable (Williams, 2015). In my case, I will simply use the command “corr” within
10
Stata to provide pair-wise correlations between the independent variables. This procedure does not include any formal test on the model but its outcome is sufficient to come to reliable conclusions whether multicollinearity is present or not. If so, it is important to find the independent variable that causes the disturbance and eliminate or rather solve the problem in order to avoid possible changes within the results.
3.2.3 Hetero- and Homoscedasticity
Another important test that should be done before interpreting any results given by the probit model, applies to homo- and heteroscedasticity. Homoscedasticity can be translated as “same variance” whereas heteroscedasticity is the violation of homoscedasticity (StatisticsSolutions, 2013). In general, a probit model consists of one dependent variable and a single or multiple independent variables, each individually connected to the dependent variable. This relationship can be described by an error term and its variance determines homo-and heteroscedasticity.
Ideally, the variance of errors is around the same level for all regressors, a relatively even distribution, indicating homoscedasticity. Heteroscedasticity is present if the variances of the error term differ significantly and are not evenly scattered (Osborne and Waters, 2002). If the relationship between the variables is described as heteroscedastic, the functional form of the probit model can change and lead to serious misinterpretations of the results (Söderbom, 2009).
One way to avoid this issue is using the robust standard errors to supplant the default variance matrix estimate within the probit model. Hence, the new outcome of the regression model is based on heteroscedasticity-robust estimates and can be used for further calculations without difficulties (Cameron and Trivedi, 2009).
3.2.4 Goodness of fit
A goodness of fit test for the probit model examines the adequacy of the observed results given by the regression. A good fitting model is usually described by only small differences between the observed and the fitted values as well as no systematic contributions of the differences to the error structure (Archer and Lemeshow, 2006). Therefore, I will run a post-estimation test within Stata, using the command “estat classification”, which will compare the fitted and actual values of the outcome. This test will show me the percent of my calculations classified as correctly and incorrectly (Cameron and Trivedi, 2009).
11 3.2.5 Marginal effects
In general, marginal effects are more informative than coefficients in nonlinear models. This is due to the fact, that the coefficients in a linear model already represent the marginal effects. The problem in nonlinear models is that the marginal effects are not fully given by the coefficients and have to be calculated separately. This can be done within Stata by using the mfx command, which computes the marginal effects of all independent variables (Norton et al., 2004).
3.3 Variables of the probit model
The probit model regression requires one dependent binary variable (Y) and several independent variables (Xi) related to the dependent variable Y. As dependent variable Y, I use the “operation status” of each wind farm that states whether a wind farm is successful and gained construction permission (Y=1) or not (Y=0). Based on my research question, I try to include variables (regressors, Xi) from three different sectors, environment, geography and economy, influencing the probability of Y=1 in order to cover the most probable determinants.
These variables comprise environmental, geographical and economic effects regarding the construction of new wind farm sites. Environmental impacts are thereby described by the variables “installed capacity” and “number of turbines”. Both determine the vertical and horizontal size of a wind park and hence, the effect on wildlife and landscape. Furthermore, I use the independent variables “population” and “location” to categorize geographical factors.
The population density in certain areas describes rural and urban preferences towards wind energy development and “location” divides my research area, the United Kingdom, in clear subareas. In terms of location, the European NUTS system (Nomenclature of Territorial Units for Statistics) provides a suitable division of England, Wales, Scotland and Northern Ireland into 12 NUTS1 regions and 36 NUTS2 regions. NUTS1 areas are equivalent to regions whereas NUTS2 areas are equivalent to single counties or grouped counties. Lastly, the variables employment rates, energy consumption, income rates and growth rates describe the economic status in each NUTS2. I chose these economic values due to earlier empirical studies, examining the relationship between employment/growth rates and the development of renewable energies.
In summary, my probit model is constructed by 7 different independent variables, covering environmental, geographical and economic issues, plus the location variables, adding 12 NUTS1 variables. These NUTS1 variables are, just as the dependent variable, dummy variables meaning that each of them produce two outcomes, one and zero. One symbolizes that the
12
implemented wind project in a certain area gained permission and is successfully operating whereby zero states the opposite, no permission and therefore no current operation or generation of energy.
3.4 Data collection
The data for my study is collected from two different sources, the Department of Energy &
Climate Change (DECC) in the United Kingdom and Eurostat, a leading provider of high quality statistics on Europe.
The results of DECC’s renewable project monitoring work is monthly recorded and published on their webpage. Their dataset enables me to extract the most relevant information I need for my project, a list of all onshore wind energy projects in the UK, their installed capacities and total number of turbines, the development status and the exact location of each wind farm. The installed capacity refers to an entire wind farm and is calculated by the number of wind turbines times the turbine capacity. It measures the electrical capacity in megawatts (MW), reaching from a minimum generation of 1 MW up to 115 MW. The number of turbines used to calculate the installed capacity differs between 1 and 36 turbines per wind farm. In addition, each wind farm is individually categorized regarding its development status. The 7 different statuses are called abandoned, application refused, application submitted, application withdrawn, awaiting construction, under construction and operational. Furthermore, the dataset divides the United Kingdom into 12 different regions (NUTS1) as locations for wind farms. England consists of 9 regions, the East Midlands, Eastern, London, North East, North West, South East, South West, West Midlands as well as Yorkshire and Humber. Wales, Scotland and Northern Ireland do only consist of one region each. Within these regions, the dataset gives also information about the counties where each wind farm is exactly located. Thus, it is possible to categorize the data into the NUTS2 system, by selecting the right county for the right region.
In order to come to reliable conclusions I need additional data containing economic numbers that might affect and determine the location of wind farms. Therefore, I will focus on employment rates, growth rates, total energy consumption, income of household and population density. This information is disposable on Eurostat’s webpage, a leading provider of high quality statistics on Europe. All data is based on the NUTS2 system, meaning that Eurostat provides these economic data for every NUTS2 area within the United Kingdom. Employment rates are given in percentage, measured from 1999 until 2014. The dataset for growth rates
13
covers the period 2000 to 2011 and is measured as gross domestic product (GDP) at current market prices by NUTS2 regions. The total energy consumption is given in thousand TOE (tonnes of oil equivalent), covering the years between 2005 and 2009. Finally, the income per household is tabulated in euro per inhabitant between 2000 and 2011 and the population has been counted since 1990.
3.5 Probit regression model 3.5.1 First stage
Based on these data, I can start constructing my probit regression model, characterized by the independent variables capacity, employment rates, turbine number, energy consumption, income per household, growth rates, population as well as the location described by the NUTS1 system (dummy variables).
First, the different datasets have to get sorted and organized in order to create one file with all relevant information. The data from Eurostat has already been tabulated in the NUTS2 system and I only have to decide about the time frame I want to use for calculating the average values.
The mean value for population is based on the period from 2002 till 2014 (appendix, Table 17), employment rates reach from 2005 to 2014 (appendix, Table 14), Energy consumption covers the period 2005 till 2009 (appendix, Table 12) and the mean value for GDP growth and income is based on the period 2001 – 2011 (appendix, Table 16 and Table 15). Although, the variables are calculated throughout different periods, it will not influence the results due to no significant economic changes over the missing periods in each category. The more difficult part is organizing the main data from the Department of Energy & Climate Change (DECC). I have to select the relevant information, all onshore wind energy projects, the installed capacity and total number of turbines, the development status and the exact location and sort it with respect to the NUTS2 system. After matching all the data within one file, I can start creating my probit model.
The dependent binary variable “STATUS” shows one for all wind farms that obtained construction permission and are currently awaiting construction, under construction or operating. It shows zero for all wind farm projects that did not gain construction permission.
These unsuccessful projects are either abandoned wind farms or projects with refused, submitted or withdrawn applications. Twelve binary independent variables are created from the 12 NUTS1 regions to get information about wind farm locations. Table 1 shows these created REGION_DUMMIES with their associated regions.
14 Table 1 – NUTS1 dummy description
Dummy Region
REGION_DUMMY1 East Midlands
REGION_DUMMY2 Eastern
REGION_DUMMY3 London
REGION_DUMMY4 North East
REGION_DUMMY5 North West
REGION_DUMMY6 Northern Ireland
REGION_DUMMY7 Scotland
REGION_DUMMY8 South East
REGION_DUMMY9 South West
REGION_DUMMY10 Wales
REGION_DUMMY11 West Midlands
REGION_DUMMY12 Yorkshire and Humber
Subsequently, I have all my data in correct terms and order and can start running the probit model. As mentioned before, the operational status of a wind project is used as dependent variable and the independent variables comprise the 12 regions, the capacity of each wind farm, the number of turbines, the mean value of employment, the energy consumption, the GDP, income per household and the number of average population. Furthermore, I have to choose a reference dummy out of the 12 regions, meaning I include only 11 regions in my regression and omit one region. My reference dummy is Scotland (DUMMY_Region7) due to its high number of wind projects compared to the total number of wind farms in the UK. This is a necessary step in my analysis because the calculated coefficients from the regression will be interpreted in reference to the excluded region, Scotland. Table 2 shows the results of the first regression, including all independent variables.
15
Table 2 – Probit regression including all independent variables REGION_DUMMY3 != 0 PREDICTS SUCCESS PERFECTLY REGION_DUMMY3 DROPPED AND 4 OBS NOT USED NUMBER OF OBS = 1578
PROB > CHI2 = 0.0000 PSEUDO R2 = 0.0699
STATUS Coef. Std. Err. z P>z [95% Conf. Interval]
REGION_DUMMY1 .077319 .1640752 0.47 0.637 -.2442625 .3989004 REGION_DUMMY2 .3768663 .2235594 1.69 0.092 -.0613021 .8150347
REGION_DUMMY3 0 (omitted)
REGION_DUMMY4 .276977 .2433572 1.14 0.255 -.1999943 .7539484 REGION_DUMMY5 .0015988 .1534948 0.01 0.992 -.2992455 .3024431 REGION_DUMMY6 .5883869 .2179926 2.70 0.007 .1611293 1.015644 REGION_DUMMY8 -.3095126 .313284 -0.99 0.323 -.9235379 .3045127 REGION_DUMMY9 -.1434923 .1654321 -0.87 0.386 -.4677333 .1807487 REGION_DUMMY10 -.0078415 .1692096 -0.05 0.963 -.3394862 .3238031 REGION_DUMMY11 -1.066419 .3906605 -2.73 0.006 -1.8321 -.3007388 REGION_DUMMY12 .1901314 .1924862 0.99 0.323 -.1871347 .5673975
CAPACITY -.0315894 .0035024 -9.02 0.000 -.0384539 -.0247249
TURBINES .0909643 .0102106 8.91 0.000 .070952 .1109767
EMP_MEAN .0158722 .0309029 0.51 0.608 -.0446965 .0764409
ENERGY_MEAN 5.29e-06 .0000517 0.10 0.918 -.000096 .0001066
GDP_MEAN -7.55e-06 .0000291 -0.26 0.796 -.0000647 .0000496
INCOME_MEAN -.0000185 .0000565 -0.33 0.744 -.0001292 .0000923
PP_MEAN -1.17e-07 1.86e-07 -0.63 0.528 -4.82e-07 2.47e-07
_CONS -.6892277 2.169126 -0.32 0.751 -4.940637 3.562182
3.5.2 Second stage
The next step is analysing the first stage regression outcome to avoid possible mistakes that might occur, including multicollinearity, mis-specifications and hetero-/homoscedasticity.
Subsequently, I will calculate the marginal effects for the probit model to interpret the impact on Y caused by the different regressors.
To track down multicollinearity, Stata uses the command “corr” to compute the percentage of correlation between the independent variables. Its outcome is shown in the appendix in Table 10, where the numbers are given in percentage and the signs symbolize a positive or negative relationship. The paired variables Capacity and Turbines as well as GDP_MEAN and Income_MEAN accord by about 90% and are therefore highly correlated. The result is a logical consequence of the fact that the calculation of the installed capacity is based on the number of turbines and growth rates usually affect income levels and the other way round. Consequently,
16
one variable of each correlated pair can be dropped and ignored within the regression. I delete the number of turbines as well as the growth rates and keep the variables installed capacity and income per household. If the kept variables were changed with the dropped ones, the results would not differ due to the high correlation causing multicollnearity between the paired variables.
In terms of mis-specification, there exist a couple of methods to detect this error. One possible way to perform a model specification test is the so-called “linktest” within Stata. The test creates two new variables, _hat and _hatsq, the variable of prediction and the variable of squared prediction. The idea behind the test is that no additional independent variables will be found and specified as significant when searching for a link error. A model without a link error will have a nonsignificant z test for _hatsq versus _hat (idre-UCLA: Institute for digital Research and Education). The linktest for my probit model is shown in the appendix in Table 11, where _hat is significant since it is the predicted variable and _hatsq is 0.311 for P>|z| and therefore not significant. Hence, my regression was specified correctly and I could exclude mis- specification from my regression.
The last provision to conduct concerns homo- and heteroscedasticity. Therefore, I use heteroscedasticity-robust standard errors within my probit regression. To deal with the problem of heteroscedasticity, a binary outcome model requires the addition of the heteroscedasticity- robust standard errors. The robust option within Stata is the most common and easiest way to avoid and eliminate the problem of heteroscedasticity. When using this method, it is important to compare the outcome of the probit regression with the new outcome of the probit regression in addition of the robust standard errors, with special focus on the signs of the coefficients.
Table 3 shows the estimated data with the heteroscedasticity- robust standard errors resulting in small changes of significance. Compared to Table 2, the regression with robust standard errors has a new significant dummy, DUMMY_REGION2 (Eastern). Furthermore, the variable Capacity lost its significant status but due to its P value close to 0.05, I will still treat it similar to significance.
17
Table 3 – Probit regression with robust standard errors REGION_DUMMY3 != 0 PREDICTS SUCCESS PERFECTLY REGION_DUMMY3 DROPPED AND 4 OBS NOT USED NUMBER OF OBS = 1578
PROB > CHI2 = 0.0000 PSEUDO R2 = 0.0229
STATUS Coef. Std. Err.
Robust
z P>z [95% Conf. Interval]
REGION_DUMMY1 .048451 .1502398 0.32 0.747 -.2460136 .3429157
REGION_DUMMY2 .366328 .1811577 2.02 0.043 .0112655 .7213905
REGION_DUMMY3 0 (omitted)
REGION_DUMMY4 .0640502 .228226 0.28 0.779 -.3832645 .511365
REGION_DUMMY5 -.0533811 .149751 -0.36 0.721 -.3468877 .2401256
REGION_DUMMY6 .4800599 .2027364 2.37 0.018 .0827038 .8774159
REGION_DUMMY8 -.342131 .2790369 -1.23 0.220 -.8890333 .2047713
REGION_DUMMY9 -.212417 .1604643 -1.32 0.186 -.5269212 .1020871
REGION_DUMMY10 -.0335836 .1588962 -0.21 0.833 -.3450144 .2778473
REGION_DUMMY11 -.874812 .3623503 -2.41 0.016 -1.585006 -.1646184
REGION_DUMMY12 .0772015 .1833263 0.42 0.674 -.2821115 .4365145
CAPACITY -.0017433 .0010741 -1.62 0.105 -.0038486 .000362
EMP_MEAN -.0016227 .029271 -0.06 0.956 -.0589929 .0557474
ENERGY_MEAN .0000124 .0000499 0.25 0.803 -.0000853 .0001102
INCOME_MEAN -.0000219 .0000323 -0.68 0.497 -.0000852 .0000414
PP_MEAN -2.10e-07 1.68e-07 -1.25 0.212 -5.39e-07 1.19e-07
_CONS .755983 1.985852 0.38 0.703 -3.136216 4.648182
Thus, my probit model meets the required assumptions regarding multicollinearity, mis- specification as well as hetero-/homoscedasticity and can be used for further calculations. The first stage regression model differs in its number of regressors and the used standard errors compared to the second stage regression model. Table 3 shows the second stage regression, excluding the two variables, number of turbines and GDP, and adding heteroscedasticity-robust standard errors.
18
The final step before interpreting the results is the calculation of the marginal effects. This is due to the fact that I try to determine the partial effect of each independent variable on the probability that the observed dependent variable Y=1, meaning that a wind project gained permission and is currently awaiting construction, under construction or operating. The results of the calculated marginal effects are represented in the following Table 4:
Table 4 – Marginal effects probit regression with robust standard errors
Variable dy/dx Std. Err. z P>z [ 95% C.I. ] x
REGIO~Y1*
.0192974 .05976 0.32 0.747 -.097821 .136416 .057034 REGIO~Y2* .1425976 .06757 2.11 0.035 .010169 .275026 .058302 REGION~4* .0254976 .09066 0.28 0.779 -.15219 .203185 .051965 REGION~5* -.0212922 .05973 -0.36 0.721 -.138359 .095774 .063371 REGION~6* .1847681 .07311 2.53 0.011 .041471 .328065 .08872 REGION~8* -.1347 .10649 -1.26 0.206 -.343418 .074018 .020913 REGION~9* -.0844229 .06316 -1.34 0.181 -.208214 .039368 .048796 REGIO~10* -.0133945 .06338 -0.21 0.833 -.13762 .110831 .091255 REGIO~11* -.3144581 .10274 -3.06 0.002 -.515819 -.113097 .010773 REGIO~12* .0307195 .07274 0.42 0.673 -.11185 .173289 .057034 CAPACITY -.0006951 .00043 -1.62 0.105 -.001534 .000144 19.7071 EMP_MEAN -.000647 .01167 -0.06 0.956 -.023521 .022227 71.9891 ENERGY~N 4.96e-06 .00002 0.25 0.803 -.000034 .000044 3008.22 INCOME~N -8.73e-06 .00001 -0.68 0.497 -.000034 .000016 17482.4 PP_MEAN -8.36e-08 .00000 -1.25 0.212 -2.1e-07 4.8e-08 1.3e+06
4. Results
4.1 Descriptive analysis
The descriptive analysis will present a short overview over the most significant numbers and patterns that might emerge from the data. This concise summary of the following tables gives the base for further calculations and interpretations.
To begin with, Table 5 gives a numerical and percental abridgment of all wind farms located in the different NUTS1 regions across the United Kingdom.
19
Table 5 – Operating and non-operating wind farms in NUTS1
Region Status Total Percentage
0 1
East Midlands 45 45 90 5.69
Eastern 41 51 92 5.82
London 0 4 4 0.25
North East 37 45 82 5.18
North West 49 51 100 6.32
Northern Ireland 47 93 140 8.85
Scotland 352 361 713 45.07
South East 24 9 33 2.09
South West 44 33 77 4.87
Wales 74 70 144 9.10
West Midlands 14 3 17 1.07
Yorkshire And Humber
40 50 90 5.69
Total 767 815 1582 100
In total, the dataset comprises 1582 wind farms but merely every second farm is currently operating and generating energy. When looking at the numbers of wind farms per region, it is obvious that most of these wind energy projects are located in Scotland, accounting for 45% of all windfarms in the UK. Followed by England with 37% and then Wales and the Northern Ireland that provide together about 18% of UK’s wind farms. The same ranking within the regions can be used in terms of average growth rates (GDP), employment rates and income per household, based on their last measured updates in Table 6. It also shows that Scotland’s total population is ten times smaller than England’s population, but it has the highest income per household and the highest GDP growth rates (Table 6). Comparing Wales and the Northern Ireland to the rest of the UK, both countries are disadvantaged regarding average income, GDP growth rates and employment rates.
20
Table 6 – Average population, income levels, GDP, employment rates and energy consumption in England, Wales, Scotland and Northern Ireland
This trend changes when analysing the total energy consumption in the four countries. The lowest consumption is measured in England, about 1000 thousand TOE less than in Wales, the country with the highest energy uptake rate of almost 4000 thousand TOE. Another interesting aspect is that every NUTS2 that shares more than 3% of UK’s total wind energy generation (appendix, Table 13) has a growth in employment rates since 1999. The most relevant counties are Highlands and Islands as well as East Yorkshire and Northern Lincolnshire with an increase in employment of 5%, Cumbria with an increasing rate of 6% and in South Western Scotland, the employment rate went from 62.3 in 1999 up to 70.1 in 2014 (appendix, Table 14). However, there is no significant growth in income per household during that period. All counties that share more than 3% of UK’s wind energy measure small fluctuations between +/- 100-800 € per inhabitant except for Highlands and Islands. This county provides 28.55% of UK’s wind farms and increased its average income per household by 2700€ per inhabitant within 11 years (appendix, Table 15). Compared to the other economic variables, there is no observable trend or relationship between changes in GDP and the development of wind energy in England, Wales, Scotland or Northern Ireland between 2000 and 2011 (appendix, Table 16).
Furthermore, the population in the contemplated counties is comparable with other parts of the UK except of Cumbria and the Highlands and Islands that stand out with a relatively low population of less than 500,000 people (appendix, Table 17).
The descriptive analysis shows that the favoured location for wind farms in the United Kingdom is definitely Scotland. I also observe a possible relationship between employment rates and the development of the energy sector whereas the other attributes have to get tested by a statistical model in order to come to reliable conclusions. Special focus will be put on the reference variable Scotland due to its high percentage of wind power farms, its unusual income, its growth rates and its low population compared to the other countries.
Average values / Region England Wales Scotland Northern Ireland
Population (2011) 53,372.83 3,110.14 5,356.48 1,847.09
Income per hh in euro per
inhabitant(2011) 18,351.72 15,150.00 19,325.00 15,300.00
GDP in euro per inhabitant (2011) 24,872.41 20,850.00 28,550.00 21,000.00
Employment in % (2014) 72.56 68.80 74.63 66.90
Energy consumption in thousand TOE
(2009) 2947.03 3994.56 3010.75 3238.82
