Fuel Poverty
The development of fuel poverty in the
Netherlands
Claire Broeders 10365249 Bachelor’s Thesis Economics and FinanceFaculty of Economics and Business University of Amsterdam
Statement of Originality
This document is written by student Claire Broeders who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction ... 3
2. Literature Review ... 4
2.1 Definitions and measures of fuel poverty ... 4
The ten percent rule ... 4
Deviations from the ten percent rule ... 5
2.2 Causes of fuel poverty ... 6
Low household income ... 6
High energy costs ... 6
2.3 Implications of fuel poverty ... 8
Prioritizing energy spending ... 8
Reducing energy spending ... 9
2.4 Policies and measures reducing fuel poverty ... 10
Low incomes ... 10
High energy costs ... 11
3. Analysis ... 12
3.1 Methodology ... 12
Measuring the extent and development of fuel poverty in the Netherlands ... 12
Determining the influence of certain factors on fuel poverty ... 14
3.2 Results ... 15
Measuring the extent and development of fuel poverty in the Netherlands ... 15
Determining the influence of certain factors on fuel poverty ... 16
4. Conclusion and discussion ... 21
Bibliography ... 23
Appendix ... 28
Regression outcome LIHC indicator ... 28
Regression outcome 10% rule ... 29
1. Introduction
Not being able to afford adequate in-‐house warmth has become a big issue in the United Kingdom. This is due to significant energy price increases over last decades, which especially added large pressure on low income households (Hills, 2011). The term for this unaffordability of warmth is fuel poverty, which has become a concern for the UK government since 2001, when the first ‘UK fuel poverty strategy’ was issued
(Department of Energy & Climate Change, 2001). This raises the question whether fuel poverty is a problem in the Netherlands as well. The only research on this subject in the Netherlands has been done in 2012 by Wisse Veenstra, a real estate master student at the Rijksuniversity of Groningen. In his thesis, with which he won the WoOn-‐thesis prize in 2013, he researched the relationship between energy costs and incomes of
households in the Netherlands, and therefore the presence of fuel poverty in the Netherlands in 2006 and 2009, and possible influential factors.
Building upon Veenstra’s research, the purpose of this research is to get insight in the development of fuel poverty in the Netherlands over the years 2006 to 2012, and to get an impression of the most influential factors influencing the problem. This research can be used to develop appropriate policies and measures addressing the problem of fuel poverty.
This research is structured in two parts. In the literature review the possible definitions, causes, implications and solutions for fuel poverty are described. In the analysis section the presence, development and possible influence factors of fuel poverty in the
Netherlands are researched. The analysis is conducted using the WoOn datasets of 2006, 2009 and 2012. First, the presence and development of fuel poverty in the Netherlands according to two different definitions of fuel poverty is mapped. This shows that for both definitions, fuel poverty is significantly present, and it decreased substantially from 2006 to 2009, but increased, in a lesser extent, from 2009 to 2012. Next, the factors possibly influencing fuel poverty are researched using a logistic regression. It appears that for both the definitions, the biggest influence factor on fuel poverty is the ethnicity of the household members. Immigrant households, especially non-‐Western, have a higher chance of becoming fuel poor than autochthonous
households. Reasons for this could be that non-‐Western immigrants are used to higher temperatures and therefore set their heating higher, because immigrants have less knowledge of energy efficiency, and/or because immigrants have on average lower incomes.
2. Literature Review
This chapter will start with an overview of several definitions used in the last three decades. Then possible causes of fuel poverty will be outlined, followed by the
implications of the problem. The chapter ends with an overview of measures addressing the problem in the UK.
2.1 Definitions and measures of fuel poverty
Fuel poverty is still a relatively unknown term in the Netherlands, but it exists over thirty years now. The term has been redefined a number of times in the UK, which is the only country that has an official definition of fuel poverty (EPEE, 2006). Isherwood and Hancock were among the first to define ‘victims of fuel poverty’ (1979): “Households with
high fuel expenditure as those spending more than twice the median (i.e. 12%) on fuel, light and power”. In fact, twice the median was 11%, but they chose 12% in order to
correspond with other analyses that used this figure (Isherwood & Hancock, 1979). A few years later Bradshaw and Hutton (1983) mentioned that fuel poverty is a difficult
concept, and very different from poverty itself. Some people are poor, but can afford adequate warmth. Others are above the poverty line, but cannot afford to keep warm. In their article, Bradshaw and Hutton (1983) gave the following definition of fuel poverty: “the inability to afford adequate warmth at home”. This is very vague; the definition had to be specified in order to be measurable.
The ten percent rule
In 1991 Brenda Boardman, nowadays a deep expert in this field, wrote her first book on fuel poverty, where she defined a household to be in fuel poverty “when it is unable to
obtain an adequate level of energy services, particularly warmth, for ten percent of its income”. In other words: when the fuel expenditures on all energy services exceed ten
percent of the household income. Her choice for the ten percent level was based on the Family Expenditure Survey of UK households in 1988 (Boardman, 1991, p. 207). She opted for ten percent because it represented twice the median for all UK households. Later on this definition was given a specific name by Bennett, Cooke and Waddams Price (2002): Expenditure Fuel Poverty (EFP).
In 1996 the writers of the English House Condition Survey (EHCS) Energy Report used this ten percent rule, but instead of actual energy costs, they used the costs required to achieve either a minimum heating regime to safeguard health or a standard regime to provide thermal comfort, plus adequate lighting, cooking and typical appliance use (DOE, 1996). That this is a better measure becomes clear from the results of this report, which show that low income households often spent significantly less on fuel than required and suffered cold homes as a consequence.
Still, the definitions stated above are too vague. For instance, what is adequate warmth? And which type of income is used? The first is answered in the definition used in the first version of the ‘UK Fuel Poverty Strategy’, where the issue of fuel poverty was recognized by the government for the first time (Department of Energy & Climate Change, 2001):
“A fuel poor household is one that cannot afford to keep adequately warm at
reasonable cost. The most widely accepted definition of a fuel poor household is one which needs to spend more than 10% of its income on all fuel use to heat its home to an adequate standard of warmth. This is generally defined as 21⁰C in the living room and 18⁰C in the other occupied rooms – the temperatures recommended by the World Health Organisation.”
After fuel poverty was politically recognized as a real problem, Boardman wrote another book in 2010, where she states that there is no correct definition of fuel poverty, because it depends on who you want to focus on. In her book, she outlined all the consistent parts of all possible definitions and their descriptions. It contains the definition of income in the UK, which she described as full income, including benefit and income support for mortgage interst (ISMI). However, Moore (2012) argues that net housing costs should be ommited, because housing costs cannot be spent on fuel, just like taxes, which are excluded from income. Therefore, disposable income should be used in the ten percent rule.
Deviations from the ten percent rule
Although the ten percent rule was widely accepted and used among several countries, some argued that it was not the right measure. In 2011 a report by John Hills on the problem and measurement of fuel poverty was published. From his research he
concluded that the ten percent measure has a certain weakness. It does not show what happened to the separate underlying aspects of the problem, just a result of all the aspects together. Therefore, Hills concluded that an accurate measure of fuel poverty must focus on households that both have low incomes and high costs: the ‘Low Income-‐ High Costs’ (LIHC) indicator. In this definition, a household is fuel poor when it its income is below the median – and energy expenditure pushes it below the poverty line. This new definition was adopted by the Department of Energy and Climate Change (DECC) in the UK in 2013 (DECC, 2013).
Waddams Price, Brazier and Wang (2012) wrote an article about their research on objective and subjective measures of fuel poverty. Here, the objective measure is the ten percent rule for expenditure fuel poverty (EFP), and the subjective measure is
conducting surveys to find out if people are feeling fuel poor (FFP). Waddams Price et al. (2012) conclude that the latter would be a measure that should also be considered, because if the goal of the government is to eliminate EFP, there will still be a significant amount of people feeling fuel poor.
2.2 Causes of fuel poverty
Since fuel poverty has been defined above in several ways, the next step towards stating possible measures reducing this problem is to explain its causes. As can be seen in Figure 1, there are two main causes: low household income and high energy costs. On their own, high energy costs have three different causes: high prices of energy, poor insulation of homes, and poor energy consumption behavior of inhabitants.
Figure 1 (Source: adapted from Boardman (2010))
Low household income
Using the EFP definition, low household income is one of the main causes of fuel poverty, since income is needed to pay for energy costs. If income reduces, and energy costs remain constant, the threshold of income needed to pay for the energy bill increases. In her book, Boardman (2010, p.40) proves that low income correlates strongly with fuel poverty.
High energy costs
High energy prices
In the Netherlands, the main source of heating is natural gas (CBS, 2015). Since 2002 the price for natural gas in the Netherlands has increased, but oscillates around a certain level since 2006 (Figure 2).
Fuel poverty Low household income High energy costs
High prices of energy Poor insula:on of homes Poor energy consump:on behavior
Figure 2: Development of natural gas prices in the Netherlands, compared with the consumer price index in the Netherlands, both with base year 2002 (Source: adapted from Central Bureau of Statistics Netherlands)
Another source of high energy prices are energy taxes. The Dutch government levies taxes on energy to stimulate households to consume less energy (Rijksoverheid, 2015). These taxes are being raised every year. There are no discounts or surcharges for the low income households, but taxes are levied according to a levied scale: tax rates vary in accordance with the variation of use of energy (Belastingdienst, 2015).
In the past six years, normal taxes on natural gas increased with over twenty percent (Belastingdienst, 2015), which is twice as much as the increase in the consumer price index (Figure 3).
Figure 3: Development of natural gas taxes in the Netherlands, compared with the consumer price index in the Netherlands, both with base year 2009 (Source: adapted from Central Bureau of Statistics Netherlands and Belastingdienst.nl)
It is noteworthy that especially the past two years these taxes have increased substantially.
Poor insulation of homes
While both low household income and high energy prices are important drivers of fuel poverty, a crucial factor is the energy efficiency of a home. A household with a low income, but a well-‐insulated dwelling, is much less likely to be in fuel poverty than a household with a poorly insulated home (Boardman, 2012).
0.00 50.00 100.00 150.00 200.00 250.00 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Price index natural gas in the Netherlands Consumer price index in the Netherlands
90 95 100 105 110 115 120 125 2009 2010 2011 2012 2013 2014
Price index taxes natural gas in the Netherlands
Consumer price index in the Netherlands
This poor insulation in itself is often a result of insufficient capital expenditure on improving the caliber of the home (Howden-‐Chapman, Viggers, Chapman, O'Sullivan, Barnard, & Lloyd, 2012). This insufficient expenditure is mostly a result of low
household incomes. Those households do not have the ability to save money to invest in the property (Boardman, 2012). An additional cause of sub-‐standard insulation can be that many poor households live in rented dwellings, where it is the property owner’s responsibility to invest, and they may choose not to do so.
The year in which a house is built can tell something about the insulation of a home. According to the ministry of housing, spatial planning and environment of the
Netherlands (2002), the degree of insulation of houses has risen substantially over the years. Especially since 1971, when regulation regarding the insulation of newly built dwellings became stricter.
Poor energy consumption behavior of inhabitants
High energy costs may also be caused by poor energy consumption behavior of the inhabitants of a home. Brounen, Kok and Quigley (2013) used a detailed survey of 1721 Dutch households to measure the extent to which consumers are aware of their energy consumption and whether they have taken measures to reduce their energy costs. Adjusting the thermostat is a clear trade-‐off between comfort temperature (higher than the required healthy temperature) and a lower energy bill, but it turns out that there is a substantial amount of households that are not aware that some actions, like turning down the temperature at night, can have such a great effect on their energy bill. On the other hand, other households may be familiar with the measures to reduce their energy costs, but with intertemporal choices like this, evidence shows that short term benefits like a higher than required temperature are worth more at the point in time, than the long term benefits, like a lower energy bill (Frederick, Loewenstein, & O'Donoghue, 2002).
2.3 Implications of fuel poverty
Households in fuel poverty face a trade-‐off: they can reduce their energy spending with the probable result of not keeping adequately warm, or they can prioritize energy spending which results in debt or reducing spending elsewhere, for example food, also causing negative effects. This trade-‐off is popularly known as ‘heat or eat?’ in the USA and England (Hills, 2011).
Prioritizing energy spending
When choosing to go into debt, there are two consequences. Firstly, there are financial implications of this unsustainable debt, such as limitation of the borrowing capacity of the household due to bad credit rating. Secondly, Brown, Taylor and Wheatley Price (2005) researched the effect of debt on mental health, and it turned out that households in unsustainable debt report significantly lower levels of psychological well-‐being than those not in debt.
The other choice is to reduce spending on another primary need, often food. This means that either the amount or the quality of food will decrease. A study by Bhattacharya, DeLeirra, Haider and Currie (2003) conducted in the United States confirms that household spending on food in the winter months decreased
substantially. Another study conducted in several urban sites in the USA found that the calorie intake of low-‐income households was ten percent lower during winter months, also for children (Frank, et al., 2006).
Reducing energy spending
On the other hand, choosing to cut out on energy spending leads to lower home temperatures. It appeared that in the UK a large percentage of fuel poor people live in homes that are persistently cold and humid (Liddell, 2008). Next to the lower comfort levels that this causes, the exposure to cold can lead to significant physical and mental health problems, but it can also have a negative social impact.
The majority of evidence of physical health effects linked to fuel poverty relates to the lower temperature. These effects vary from respiratory problems to excess winter deaths, and vary over age classifications and initial health conditions of people.
Evidence from multiple studies suggests that cold related morbidity is mainly present among the vulnerable and disadvantaged groups in fuel poverty, namely the elderly, very young children and people with a long-‐term sickness or disability (Hills, 2011). These people have an increased chance of respiratory problems with a home
temperature below 16°C, blood circulatory problems below 12°C and risk of hypothermia below 6°C (Marmot Review Team, 2011). A different study shows
evidence that respiratory problems at temperatures below 16°C is only significant when also the humidity level differs from the optimal level. High humidity is caused by
condensation forming on cold, poorly insulated fabric, and is detrimental in itself because it results in mold, and inhabitants may develop allergic responses to that. Besides that, it may reduce resistance to infections such as colds (WHO, 1987). These health effects mentioned above can be fatal. In the Netherlands almost ten percent more people die in the winter months compared to the rest of the year, as is shown graphically in Figure 4 (CBS, 2015). Not all of these deaths are a result of fuel poverty drivers, but these could be important factors.
However, in their article on fuel poverty and human health, Liddell and Morris (2010) conclude that while several studies on fuel poverty found evidence of a substantial negative physical health effect, this evidences is weakened by methodological limitations.
Figure 4: seasonal fluctuation in mortality in the Netherlands (Adapted from the Central Bureau of Statistics Netherlands (2015))
Next to the possible physical health problems mentioned above, recent studies provide evidence that there is a link between living at low temperatures and the mental well-‐ being of adults. According to Hills (2011) the most convincing evidence leads to two direct, and one indirect mental health issue. Firstly, living in low temperatures can cause the inhabitants physical discomfort, which in turn can cause high stress levels. Secondly, factors related to the cold can directly contribute to the development of common mental disorders. Finally, also social problems can occur due to living in a cold home, which may have potential knock-‐on effects for mental health. These social
problems seem to differ for older and younger people. Older people struggle with social isolation and exclusion, while adolescents face problems related to education and anti-‐ social behavior (Hills, 2011).
2.4 Policies and measures reducing fuel poverty
When looking at policies, the two main causes must be considered: policies addressing low household incomes, and policies addressing high-‐energy costs.
Low incomes
The Fuel Poverty Advisory Group for England (FPAG, 2012) suggests using the raised energy taxes to benefit fuel poor households in particular. The FPAG also recommends a campaign to encourage benefit take up, since nearly 25 percent of all available benefits expenditure in 2009-‐2010 were unclaimed. According to Bradshaw and Hutton (1983) there are three main options to extend existing income maintenance measures, namely increasing benefits across the board, introducing a fuel allowance and extending the existing scheme of additional requirements.
10000 10500 11000 11500 12000 12500 13000
Average number of deaths per month in
the Netherlands, 2002-‐2014
High energy costs
The FPAG (2012) suggests that all privately rented properties must be brought up to a minimum energy efficiency standard rating by 2018. With better insulation, energy costs will decrease. Bradshaw and Hutton (1983) mention three measures reducing energy expenditure. Prices could be held down by subsidies or restructured by tilting tariffs, payment methods could be changed from quarterly payments to monthly payments, so that the burden can be spread out equally, and finally the Government should take measures increasing the energy cost awareness of households.
According to Allcott and Mullainathan (2010) the relevance of information provision in changing consumer behavior has been addressed in field experiments providing
feedback on energy consumption to consumers. They suggest that policy-‐makers should encourage private-‐sector firms to generate and utilize behavioral innovations that “nudge” consumers toward reducing energy use. Nudges are small changes in context that can indirectly influence the motives, incentives and decision making of people.
3. Analysis
3.1 Methodology
As mentioned in the literature review, there are several definitions of fuel poverty. There are two definitions that are being measured for the Netherlands in this research. The first definition we use is known as expenditure fuel poverty. Actual fuel spending is not the most accurate indicator of fuel poverty, since low income households spend significantly less on energy than required (Moore, 2012). But using required energy costs requires a detailed knowledge of the energy efficiency of the housing stock, which is not available. The other definition is called the ‘Low Income High Costs’ indicator. This is measured by selecting the households that have an income less than the median, and where subtracting the energy costs pushes the income below the poverty line. For the analysis, the datasets WoOn 2006, WoOn 2009 and WoOn 2012 have been used. These datasets contain 800 to 900 variables, with 60,000 to 80,000 observations,
varying per dataset. The WoOn datasets are the results of surveys sent out every three years to research the quality of living and housing, and are primarily used in support of government policy in this area. Information on household situation, current and desired living arrangements, housing costs and incomes is merged here.
The analysis consists of two parts. First, the extent of fuel poverty in the Netherlands is measured for both definitions in 2006, 2009 and 2012, so that the development over the years, and the different outcomes from the definitions can be shown. Next, for the LIHC indicator and for the ten percent rule, both for 2012, two regressions are conducted to research the influence of certain variables on the probability of being in fuel poverty.
Measuring the extent and development of fuel poverty in the Netherlands
In this part, the analysis using the ten percent rule is done in a similar way to Veenstra (2012). However, he did not measure fuel poverty with the LIHC indicator.
The ten percent rule
Only two variables are used to research the development of fuel poverty according to the ten percent rule in the Netherlands: disposable household income (as defined by the CBS) per year, denoted by cbssch (or BESTINKH in 2012), and average total energy costs per month, denoted by totener. Disposable income per household defined by the CBS is gross income minus paid income transfers, income insurance costs, health insurance costs and taxes on income and capital.
For this analysis, the energy costs/disposable income ratios are tested against the ten percent rule. Firstly, all the cases for which the data on one of the two variables is
missing, due to incomplete survey results, are deleted from the datasets. Then the ratios are obtained by multiplying the total energy costs per month with twelve, which gives the total energy costs per year, and dividing that by the total disposable income per year:
𝑅𝑎𝑡𝑖𝑜 = 𝑡𝑜𝑡𝑒𝑛𝑒𝑟 ∗ 12
𝑐𝑏𝑠𝑐ℎℎ 𝑜𝑟 𝐵𝐸𝑆𝑇𝐼𝑁𝐾𝐻
A binary variable is then created to indicate the presence of fuel poverty. It gets a value of 1 if the ratio is below 0.10, and therefore fuel poverty is present in that household, and a value of 0 if the ratio is 0.10 or higher, so the household is not in fuel poverty. After doing this for all three years, the development over the years can be shown in a graph.
The ‘Low Income High Costs’ indicator
For this measure, some more information is needed. Besides the two variables from the dataset, namely disposable household income and total energy costs per month, the poverty line for the Netherlands is needed.
Low income threshold per month for the Netherlands
No kids 2 parents 1 parent
Single (1 pers) Couple (2 pers) 1 kid (3 pers) 2 kids (4 pers) 3 kids (5 pers) 1 kid (2 pers) 2 kids (3 pers) 3 kids (4 pers) 2006 €880 €1210 €1470 €1660 €1820 €1170 €1330 €1550 2009 €930 €1270 €1550 €1750 €1910 €1240 €1400 €1640 2012 €990 €1350 €1650 €1850 €2030 €1310 €1490 €1740
Table 1: Per month poverty line for the Netherlands according to the Low Income Threshold Measure (Source: adapted from SCP/CBS 2010 and 2014)
Because the information about persons in each household in the WoOn datasets is not this detailed, the thresholds have to be revised. The WoOn dataset only gives the number of people in each household. Therefore, the average of the poverty lines with the same amounts of people are taken. The assumption has to be made that there is no difference in required expenditures for a kid and an adult, which there may be in practice. In addition, there is no poverty line for households with more than five persons, so the line for five persons is also used for more persons.
Low income threshold adapted for number of persons in a household
1 person 2 persons 3 persons 4 persons 5 or more persons 2006 €880 €1190 €1400 €1605 €1820
2009 €930 €1255 €1475 €1695 €1910 2012 €990 €1330 €1570 €1795 €2030
Table 2: Per month poverty line for the Netherlands, adapted for number of persons in a household, according to the Low Income Threshold Measure (Source: adapted from SCP/CBS 2010 and 2014).
For this analysis, first the median of all incomes in the dataset is determined. Then, the incomes above the median are assigned a value of zero, the incomes below median are assigned a value of one. From data with a value of one, for each observation the energy costs are subtracted from the income. The income that is left is then tested against the poverty lines for each number of household members. A binary variable is then created, indicating the presence of fuel poverty. If this leftover income is below the poverty line, it is assigned a value of one, and if not, the value becomes zero. After doing this for 2006, 2009 and 2012, the development of fuel poverty can be
Determining the influence of certain factors on fuel poverty
To determine the influence of certain factors on fuel poverty, two regression analyses are constructed: one for the LIHC indicator, and one for the EFP measure. Both were done for the year 2012, as it is the most recent year, leading to the most accurate recommendation. The dependent variables for the regressions are the binary variables created for indicating if a household is in fuel poverty or not. Because the dependent variable is binary, a binary logistic regression is constructed.
The independent variables are chosen based on the literature review, and the thesis of Veenstra (2012), and are outlined in Table 3. The number of inhabitants may influence the household income, but also the size of a house, and therefore the usage of energy. The age of a house can influence the degree of insulation, and therefore fuel poverty. If a house has more rooms, there are usually also more rooms to heat. The same counts for the surface area of a home, if it is larger, more energy is needed to heat the house to adequate warmth. Finally, it will be studied if the province in which the household lives, the urbanity of the town in which the household lives, and the ethnicity of the
inhabitants have an influence on fuel poverty.
Factor Code in data
Number of inhabitants Aantalpp Year of construction Bjaar Number of rooms Kamers Surface area in m2 Opptbin
Province Prov
Urbanity of town Stedgem Ethnicity of inhabitants Etniop3
Table 3: the independent variables used in the regression.
The nominal variables are broken down into dummy variables. For ‘province’ the reference dummy is Limburg. ‘Urbanity’ is broken down into ‘very strong’, ‘strong’, ‘moderate’, ‘weak’ and ‘not’. ‘Not’ is the reference dummy. Finally, the ethnicity of inhabitants is broken down into ‘autochthonous’, ‘immigrant Western’ and ‘immigrant non-‐Western’. Here ‘autochthonous’ is the reference dummy.
The regression model
A logistic binary regression model has an outcome variable, denoted as Y, which is categorical, and mostly also dichotomous, having either a success or failure as outcome. Here the logistic binary regression model describes the chance that a respondent is in fuel poverty relative to the chance that he or she is not. When these probability values are divided by each other, the ‘odds-‐ratio’ is created, which has a range from zero to infinity. In a logistic regression, the natural logarithm is taken from this odds ratio, which makes the regression look like this:
𝐿𝑛 𝐹𝑢𝑒𝑙 𝑝𝑜𝑣𝑒𝑟𝑡𝑦 (𝑌 = 1)
𝑁𝑜 𝑓𝑢𝑒𝑙 𝑝𝑜𝑣𝑒𝑟𝑡𝑦 (𝑌 = 0) = 𝛽!+ 𝛽!!"#$$𝑋!!"#$$+ 𝛽!"##$𝑋!"##$+ ⋯ + 𝛽!"#$%&#𝑋!"#$%&# To make the model easier to interpret, the odds ratios of the independent variables are determined by taking the exponent ‘e’ of the betas. This gives the following regression:
𝑃 𝑌 = 1 = 𝑂𝑅! +𝑂𝑅𝑎𝑎𝑛𝑡𝑝𝑝𝑂𝑅𝑎𝑎𝑛𝑡𝑝𝑝 + 𝑂𝑅𝑏𝑗𝑎𝑎𝑟𝑂𝑅𝑏𝑗𝑎𝑎𝑟+ ⋯ + 𝑂𝑅𝑛𝑜𝑡𝑤𝑒𝑠𝑡𝑋𝑛𝑜𝑡𝑤𝑒𝑠𝑡
The logistic regression differs from a multiple regression in the fact that it does not assume a linear relationship between the dependent and independent variables. Therefore, homoscedasticity and normal distribution are not required.
The Nagelkerke R2 test will show to what extent the data is explained by the model. If
the Chi2 test is significant, it means that the model fits the data better than a model
without the independent variables. 3.2 Results
The results are divided in two sections: the results of measuring the extent and development of fuel poverty, and the results of the factors influencing fuel poverty.
Measuring the extent and development of fuel poverty in the Netherlands
The outcomes of the analysis are shown in Table 4, and graphically in Figure 5. The results show that the percentage of households in fuel poverty both for the ten percent rule and for the LIHC indicator decreased substantially from 2006 to 2009. During the crisis, from 2009 to 2012, the percentage of households in fuel poverty in the
Netherlands increased again, to a lesser extent. A reason for this increase could be that because of the crisis, household incomes did not grow during those years, and in 2010 even decreased relative to the year before (CBS, 2014).
With the ten percent measure, substantially more people are in fuel poverty than with the LIHC indicator. The reason for this is probably that in the ten percent measure people with a high income, and still high energy costs, are also included.
2006 2009 2012
Freq. Percentage Freq. Percentage Freq. Percentage In fuel poverty (ratio ≤ 0.10) 11727 21.61% 9504 14.13% 9429 15.59% Not in fuel poverty (ratio > 0.10) 42545 78.39% 57747 85.87% 51058 84.41%
Total 54272 100.00% 67251 100.00% 60487 100.00%
Average ratio 7.79% 7.46% 7.62%
In fuel poverty (LIHC) 6509 11.99% 5327 7.92% 6087 10.10% Not in fuel poverty (no LIHC) 47762 88.01% 61924 92.08% 54200 89.90%
Total 54271 100.00% 67251 100.00% 60287 100.00%
Average energy costs per year (€) 1977.10 1988.27 2056.69 Average disp. inc. per year (€) 31295.07 35185.23 35476.92
Table 4: results from analysis.
Figure 5: percentage of households in fuel poverty in the Netherlands using the 10% rule and the LIHC indicator.
Determining the influence of certain factors on fuel poverty
The outcome of the regressions are shown in detail in the Appendix, and summarized in Table 5. The Chi2 test is significant for both the regressions, which means that both the
models fit the data better than the models without the independent variables. The Nagelkerke R2 shows that for the ten percent rule, the model explains about six percent
of the total variation of the outcomes. For the LIHC indicator this is about four percent. 0.00% 5.00% 10.00% 15.00% 20.00% 25.00% 2006 2009 2012
Fuel Poverty in the Netherlands 2012
Fuel poverty -‐ 10% rule Fuel poverty -‐ LIHC
10% rule LIHC indicator
Var. Odds ratio P>|z| [95% conf. Interval] Odds Ratio P>|z| [95% conf. Interval]
aantalpp 0.6077735 0.000*** 0.59281 0.62311 0.9668172 0.010** 0.94239 0.99188 bjaar 0.9944567 0.000*** 0.99393 0.99499 0.995087 0.000*** 0.9945 0.99568 kamers 1.010261 0.245 0.99301 1.02781 0.8325766 0.000*** 0.8116 0.8541 opptbin 0.9999535 0.805 0.99958 1.00032 0.998143 0.000*** 0.99759 0.9987 dum_gr 1.038488 0.650 0.88226 1.22239 1.438565 0.000*** 1.1829 1.74948 dum_fr 0.8603064 0.081* 0.72667 1.01852 1.25253 0.037** 1.01405 1.54709 dum_dr 0.9117647 0.324 0.75894 1.09536 0.9032113 0.445 0.69552 1.17292 dum_ov 0.7553719 0.000*** 0.66643 0.85618 0.9528775 0.551 0.81317 1.11659 dum_fl 0.6717675 0.000*** 0.54675 0.82538 0.8011934 0.088* 0.62132 1.03314 dum_ge 0.6793147 0.000*** 0.61076 0.75557 0.8802117 0.073* 0.76557 1.01202 dum_ut 0.576961 0.000*** 0.49943 0.66653 0.6517757 0.000*** 0.54557 0.77866 dum_nh 0.5184045 0.000*** 0.45896 0.58555 0.5950345 0.000*** 0.51019 0.694 dum_zh 0.4867602 0.000*** 0.43508 0.54458 0.5204562 0.000*** 0.45018 0.6017 dum_ze 0.5267439 0.000*** 0.4544 0.6106 0.7306926 0.001*** 0.60219 0.88662 du_nb 0.761636 0.000*** 0.68001 0.85306 0.8329675 0.016** 0.71739 0.96716 du_li 1 1 du_stzs 0.8092828 0.000*** 0.72329 0.9055 1.725055 0.000*** 1.49497 1.99056 du_sts 0.7948052 0.000*** 0.72002 0.87735 1.212185 0.004*** 1.06414 1.38083 du_stm 0.7129658 0.000*** 0.64294 0.79062 0.9414662 0.391 0.82035 1.08046 du_stw 0.9191783 0.097* 0.83203 1.01546 0.8805871 0.068* 0.7683 1.00928 du_stn 1 1 du_etnnietwes 2.334219 0.000*** 2.15017 2.53402 3.047472 0.000*** 2.80932 3.30582 du_etnwes 1.326441 0.000*** 1.2285 1.43219 1.370314 0.000*** 1.25023 1.50193 du_etnaut 1 1 _cons 46429.04 0.000*** 16117 133750 4804.327 0.000*** 1478.04 15616.3 Table 5: Outcomes regressions 10% rule and LIHC indicator, 2012.
* P<0.1 **P<0.05 ***P<0.01 Number of inhabitants
Using the ten percent rule, the number of people in a household has a negative
correlation with fuel poverty. The chance of becoming fuel poor decreases with 39.2% when one extra person is in a household, with a P-‐value below the significance level of 0.01. This means that there is a less than 1% chance that the null-‐hypothesis (number of inhabitants does not have an influence on fuel poverty) is true. When using LIHC
indicator there is still a negative effect, but it is smaller. The chance of becoming fuel poor decreases with only 3.3%, and the P-‐value is below the significance level of 0.05.
A probable explanation for this negative correlation could be that when there are more people in a household, there are also more people with an income. The hypothesis that more people in a household also indicates a bigger home, and therefore more energy costs, clearly does not outweigh the income effect.
Year of construction
As expected, for both the measures, it appears that the more recent the house is built, the lower the chance of fuel poverty. For each year, the chance of becoming fuel poor decreases with less than 1%, but with a significance level smaller than 1%. This
correlation is likely caused by the increasing knowledge of dwelling insulation over the years. Also, in 1983 a law was set with minimum insulation requirements
(Rijksoverheid, 2015). Over the years, these requirements have become even stricter. Boumen, Kok and Quigley (2012) confirm that the relation between the age of a house and energy consumption exists: “We document a strong relationship between the
vintage of dwellings and their resource consumption. Dwellings constructed after World War II use about 65 percent less gas for heating than those constructed prior to World War II.”
Number of rooms
For the LIHC measure, the number of rooms in a dwelling has a negative effect on fuel poverty. With each extra room, there is a 16.7% smaller chance to become fuel poor, with a significance level below 0.01. This effect could be explained by the correlation between low income and small houses. Even though it could be expected that more rooms means more heating, it seems that the income effect is bigger.
Using the ten percent rule, the number of rooms in a dwelling has no significant effect on fuel poverty, because the p-‐value is above 0.1. A reason could be that the income effect and the more heating effect balance each other out.
Surface area in m2
For the LIHC measure, the surface area of a dwelling has a negative effect on fuel poverty. One square meter extra surface area, correlates with a decrease in the chance of fuel poverty of 0.19%, with a significance level below 0.01. This corresponds to the explanation for the negative effect of the number of rooms: the lower the income, the smaller the house, because the household cannot afford a bigger house. This income effect seems to be bigger than the effect of a bigger surface area meaning a larger area to heat.
The ten percent rule indicates no effect of surface area on fuel poverty. This can be seen by the value of 1 lying in the confidence interval. Also this outcome is not below any significance level.
Province
For the regression with the ten percent rule, Overijsel, Flevoland, Gelderland, Utrecht, Noord-‐Holland, Zuid-‐Holland, Zeeland and Noord-‐Brabant have a significant negative effect on the chance of fuel poverty, compared to Limburg. The chance of being in fuel poverty compared to living in Limburg increases with respectively 24,4%, 32,8%, 32,0%, 42.2%, 48.2%, 51.3%, 47.3% and 23.8%. These effects are all below the
significance level of 10%. This decrease is probably because these provinces, except for Overijsel and Flevoland, have a higher average income compared to Limburg. A reason that Flevoland has a negative correlation with fuel poverty could be that Flevoland exists for 29 years now (Flevoland, 2015), meaning that the houses in Flevoland have all been built after 1983, which indicates a good insulation.
For the regression with the LIHC indicator, living in Groningen or Friesland has a positive effect on the chance of being in fuel poverty, compared to Limburg. The chance of being in fuel poverty increases with respectively 43,9% and 25.3%. This is probably because inhabitants of Groningen and Friesland have the lowest average income of the Netherlands (CBS, 2014). Inhabitants of Flevoland, Gelderland, Utrecht, Noord-‐Holland, Zuid-‐Holland, Zeeland and Noord-‐Brabant have a significant negative effect on the chance of fuel poverty, compared to Limburg, respectively 19.9%, 12.0%, 34.8%, 40.5%, 48.0%, 25.2% and 16.7%. These effects are all below the significance level of 10%. Urbanity of town
Using the ten percent rule, it appears that there is a negative correlation between a highly urbanized town and the chance of becoming fuel poor, compared to a town that is not urbanized at all. A very strongly urbanized town decreases the chance of being in fuel poverty with 19.9%. A reason for this could be that in urbanized towns there are usually more apartment buildings, where energy efficiency is higher as apartments have less outside surface where heat can escape, compared to houses.
The LIHC indicator, however, shows a positive correlation between a high urbanization of a town and the chance of becoming fuel poor, compared to a non-‐urbanized town. Inhabitants of a town that is very strongly urbanized have a 72.5% higher chance of becoming fuel poor than a non-‐urbanized town. An explanation for this could be explained by the fact that the most low-‐income neighborhoods are located in urban towns (CBS, 2008). Why the LIHC indicator and the ten percent rule give such different outcomes should be further researched.
Ethnicity of inhabitants
Finally, it appears from both the regressions that non-‐western immigrant households have a higher chance to become fuel poor than immigrant western people, and a much higher chance than autochthonous people. Comparing to an autochthonous household, a Western immigrant household has a 32.6% higher chance to become fuel poor, using the ten percent rule. Using the LIHC indicator this is 37.0%.
Using the ten percent rule and comparing to an autochthonous household, a non-‐ Western household has a 133% higher chance to become fuel poor, while using the LIHC indicator this is 205%.
All are below a significance level of 0.01. A possible explanation can be that non-‐ Westerners heat their houses to a higher temperature, because they might be used to higher temperatures in their home country. Another explanation can be that immigrants have insufficient knowledge about energy efficiency. Finally, it appears that immigrants have significantly lower incomes than autochthonous people (CBS, 2014).