The political economy of the Dutch housing market : housing tenure and voting

(1)

The Political Economy of the Dutch

Housing Market

Housing Tenure and Voting

Wessel van Proosdij 10244670

Master Thesis Political Science: Political Economy Political Economy of Financial Crises

Supervisor: dhr. dr. J.G.W. Blom Second Reader: mr. E. Harteveld MSc

(2)

(3)

Abstract

This study analyses the effects of housing tenure on the voting behaviour of Dutch house-holds. Households are divided in four groups: homeowners in positive equity, homeowners in negative equity (underwater mortgage), social renters and private renters. Using panel data and adopting a multinomial logistic regression approach, we find that housing is able to affect voting behaviour under certain circumstances. During a housing crisis, home-owners in both positive and negative equity are more likely to vote for political parties that want to restore the housing market and increase prices (pro-ownership). However, homeowners with more than 25000 euros of negative wealth are more likely to vote for parties that provide larger welfare benefits (pro-welfare). We argue that the societal and political saliency of the housing crisis is the main driver behind our results: homeowners need to be aware of their housing wealth in order to vote in line with their tenure status. Additionally, we find that social renters are more likely to vote for pro-welfare parties in both economic upswings and downturns. We find no significant effect between the tenure status of private renters and their voting behaviour due to the diverse nature of this group.

(4)

List of Figures

1 Household debt of OECD countries, 2015 . . . 12

2 Household debt as share of income in the Netherlands, 1995–2015 . . . . 13

3 Distribution of tenure status of European countries, 2015 . . . 15

4 Average nominal house prices in the Netherlands in euros, 1995–2016 . . 16

5 Political party spectrum . . . 23

6 Predictive probability of voting for a party based on housing tenure . . . 33

7 Predictive probability of voting for a party based on housing tenure . . . 38

8 Predictive probability of party choice for households (all election rounds) 40 9 Predictive probability of party choice based on extended housing tenure . 45

List of Tables

1 Party positions of Dutch political parties . . . 28

2 Baseline model (complete sample) . . . 31

3 Baseline models (all election rounds) . . . 34

4 Controlled models (complete sample) . . . 36

5 Controlled models (all election rounds) . . . 39

6 Households and flexible division parties . . . 42

7 Households and left versus right . . . 43

8 Homeowners extended equity division . . . 44

A1 Summary statistics . . . 52

A2 Definitions and expected results of variables . . . 53

Abbreviations and Acronyms

DTI . . . Debt-to-income

GFC . . . Global Financial Crisis

IPE . . . International Political Economy

LISS . . . Longitudinal Internet Studies for the Social sciences MID . . . Mortgage Interest Deductibility

NE . . . Negative Equity

NHG . . . National Mortgage Guarantee

OECD . . . Organisation for Economic Cooperation and Development OVB . . . Omitted Variable Bias

PE . . . Positive Equity

(6)

1 Introduction

Housing markets are unique and important markets. As Reinhart and Rogoff (2009) show, many episodes of financial instability are associated with housing market booms. All six major episodes of banking crises in advanced economies since the 1970s were preceded by a housing boom and corresponding bust. The global financial crisis (GFC) was no different, with a huge housing bust in the United States and several European countries (Schwartz, 2012). One of these countries was the Netherlands, where a large housing boom during the 1980s accelerated increases in the price of real estate, and where the bubble busted during the GFC. Besides widespread economic issues such as unem-ployment, bankrupts and reduced income, Dutch homeowners saw a housing crisis that evaporated the value of their homes. This crisis increased their (already high) mortgage debts, led to difficulties in selling their house and financing a new home. These problems forced the government to change their housing policies in order to smother the crisis and stabilize the market. Deemed untouchable subjects such as homeowners’ fiscal benefits were made debatable by several political parties. Since the state of the housing market affects households directly, it is interesting to see how households respond to it. When housing prices decline, do homeowners vote for political parties that want to restore the market? And do renters respond to housing market developments? These are questions this study will try to answer. We will analyse how the Dutch housing market is able to affect voting behaviour in both economic downturns and upswings. Our research ques-tion is as follows: How does the housing tenure of Dutch households affect their voting behaviour? We will look at the Netherlands over the period 2006–2017, in which a hous-ing crisis erupted and four elections rounds were held. In order to answer our research question we make use of data of the LISS (Longitudinal Internet Studies for the Social sciences) panel administered by CentERdata (Tilburg University, The Netherlands). Households are divided in four groups: homeowners in positive equity, homeowners in negative equity (underwater mortgage), social renters and private renters. We find that housing can affect voting behaviour, but only under certain circumstances. Our results suggest that during a housing crisis, homeowners in both positive and negative equity are more likely to vote for political parties that want to restore the housing market and increase prices (pro-ownership). However, homeowners with more than 25000 euros of negative wealth are more likely to vote for parties that provide larger welfare benefits (pro-welfare). We argue that the saliency of the housing crisis is the main driver behind our results: homeowners need to be aware of their housing wealth in order to vote in line with their tenure status. In addition, we find that social renters are more likely to vote for pro-welfare parties, in both economic upswings and downturns. We find no significant effect between the tenure status of private renters and their voting behaviour.

(7)

This research extends the literature in several important aspects. First of all, this study is the first that analyses how the housing tenure of all households (both homeowners and renters) is able to affect voting behaviour. By using the LISS panel a dataset is constructed containing data of party choice, housing status and socio-economic indica-tors of Dutch households over four election rounds (2006, 2010, 2012 and 2017). While similar studies have been conducted in both the Netherlands (Andr´e, Dewilde, Luijkx, & Spierings, 2016) and other countries (Ansell, 2014), the scope of this study is unique. Additionally, this study researches housing and voting during both economic downturns and upswings, enabling us to compare how housing affects voting behaviour in differ-ent economic environmdiffer-ents. Second, in order to include all sorts of households in our research, we construct a new theoretical framework that connects housing with the per-manent income hypothesis. Third, in order to research the relationship between housing tenure and voting behaviour, we adopt a multinomial logistic regression method that provides us with predictive probabilities that show the likelihood that a household group is predicted to vote for a political party group.

This study is constructed as follows. The first chapter is devoted to existing literature that is relevant for our own research. Second, we provide a short descriptive overview of the Dutch housing market and its unique institutions and policies. Third, we construct our theoretical framework to aid us in testing our hypotheses and answering our research question. In the fourth chapter we explain our empirical set-up: the use of our data, the operationalization of our variables and our choice of methodology. The remaining two chapters are devoted to analysing the obtained results and concluding our research.

(8)

2 Literature review

This chapter describes recent and relevant empiric academic literature. The selected studies are relevant as they give us an insight in the economics and politics of housing markets, how the Dutch housing market compares to markets in other countries and which methodologies are used to analyse housing markets. This literature review is divided in three parts. First we summarize cross-country studies that analyse housing markets and how institutions and housing policies affect them. We then discuss studies that analyse the Dutch housing market, its unique housing policies and corresponding political issues. In the last paragraph we discuss studies that analyse how housing and politics connect.

2.1 Housing markets in a cross-sectional setting

Dobbs, Lund, Woetzel, and Mutafchieva (2015) show that mortgages account for most of the growth in household debt and that increases in house prices are largely determined by land prices and the availability of credit. They further show that countries that have one megacity or a few large urban agglomerations rather than multiple large cities have higher real estate prices in the central city and therefore higher levels of household debt. They also argue that the use of macroprudential tools and changes in taxation can im-prove the sustainability of housing markets in the longer run.

Johnston and Regan (2015) claim that analyses in international political economy (IPE) often argue that convergence of interest rates in light of European monetary integration and financial market liberalization are causal factors behind the rise of house prices. De-spite common credit supply shocks, European countries experienced diverging trends in housing inflation in the period 1990–2008. The authors argue that wage-setting institu-tions blunt the impact of financial liberalization since they restrain inflationary pressures on income. By using a panel regression and by comparing the housing markets in the Netherlands and Ireland, they find that income growth is a strong predictor of a housing bubble.

Wind, Lersch, and Dewilde (2016) analyse the distribution of housing wealth and how institutions have shaped this distribution. They argue that subsidies for homeownership, privatisation of social housing and mortgage finance liberalisation have influenced the distribution of housing wealth in recent decades. Using surveys, the authors analyse two different birth cohorts in 16 European countries with different welfare states and housing systems. Their results show that the expansion of homeownership in a market-based housing system is associated with a less equal distribution of housing wealth as more marginal homeowners are drawn into mortgage debt. Such a pattern is not found in regimes with a more state- or family-based provision of housing.

(9)

Fernandez and Aalbers (2016) argue that there has been a growing imbalance between the growth rate of the capital stock and GDP which resulted in a ’wall of money’ looking for profitable investment. A large part of this money has been invested in the financing of real estate, leading to an explosion of mortgage debt and housing prices. This led to problems during the GFC, when the flow of money dried up. The authors show that there are four kinds of housing markets which all have different forms of housing-based financialization. The first kind of market combines high rates of homeownership with low cross-border capital flows and a modest financial sector which has not been finan-cialized yet (e.g. Greece, Portugal, Brazil, Turkey). The second kind has high levels of homeownership with high mortgage-to-GDP levels and a deep and sophisticated financial sector (i.e. Ireland, Canada, the UK and the US). The third kind combines moderate homeownership with very high mortgage-to-GDP levels (i.e. the Netherlands and Den-mark). These countries have extremely high-cross border flows making them vulnerable to economic downswings. These countries seem to have reached the limits of housing-based financialization. The fourth kind includes countries with low ownership rates, low mortgage-to-GDP levels (e.g. Germany, Switzerland, France). These countries have strict institutional barriers against the inflow of foreign capital into the national housing sector, not allowing financialization to overflow the market.

2.2 The Dutch housing market

Boelhouwer (2005) analyses the coordination problems in the privatization of the housing market in the Netherlands in the period of 1990–2004. He finds that there is a mismatch between the explosion in house prices and the stagnation in house-building activities at the end of the 1990s. It is argued that the Dutch government stimulates housing demand with a variety of subsidies, such as the mortgage interest deduction, which caused the price explosion. In addition, the Dutch government is not focusing on the weakest groups in the housing market: it is only further increasing income and wealth gaps. Boelhouwer argues that the Dutch government has liberalized the supply side of the housing market by no longer planning and subsidizing housing production, but by still subsidizing the demand side, an environment is created in which housing supply and demand cannot react in a proper way.

In a follow-up study, Boelhouwer and Hoekstra (2009) analyse Dutch housing policies and their effects on the housing market. It is argued that although these policies seem developed in terms of money and instruments, they are inconsistent and ineffective in stabilizing the market. Housing shortages are prevalent in areas of economic growth, property prices are high, and substantial segments of the Dutch population face accessi-bility and affordaaccessi-bility problems. The government provides support of the demand side of

(10)

the housing market via mortgage interest deduction for homeowners and rent allowances for tenants but simultaneously enforce regulations and restrictions in the production of housing. The authors argue that many Dutch political parties - both at the left and the right side of the political spectrum - choose to sustain the status quo. However, outside of the government, there are many calls for reform. The Dutch central bank and the min-istry of Housing, Spatial Planning and the Environment (ministerie van Volkshuisvesting, Ruimtelijke Ordening en Milieu: VROM), project developers, building contractors and local authorities all argued that the Dutch housing market is in urgent need of reform. De Wit, Englund, and Francke (2013) research the positive relationship between housing price increases and the number of houses sold in the Dutch housing market for the period 1985–2007. They estimate a model that shows the mechanism that is the source of this correlation. According to their model, a shock in the interest rate has a continuously increasing effect on house prices, an immediate but temporary effect on sales and an insignificant effect on the rate of entry of new houses for sale. This evidence shows that the correlation between housing prices and the amount of sales is almost fully caused by the effect of interest rates changes on housing prices, which implies that the fiscal benefits the Dutch homeowners receive is highly important for the price level of homes and the amount of sales and thus the efficiency of the market.

2.3 Housing and politics

An extensive list of research exists that looks at the relationship between housing and pol-itics, especially in Anglo-Saxon political science. Kelley, McAllister, and Mughan (1985) analyse England between 1964 and 1979 and find that changes in homewownership affect voting behaviour stronger than changes in education or income. In England (1960–1980) there were two dominant housing types; owner-occupiers, who were usually Conservative in partisanship, and council housing renters, who were subsidized by the government and usually voted Labour. Conservatives have strongly encouraged homeownership and were the main architects of the huge growth in homeownership in the 1960s and the early 1970s. Furthermore, for vote-maximizing reasons the Conservatives have advocated lower interest rates in home loans and introduced policies to allow council tenants to purchase their homes at advantageous prices. Labour has traditionally supported coun-cil housing, originally as a social welfare measure, and have been identified as strongly opposing the Conservative initiatives to reduce the size of the council rental market. Con-sequently, Labour was unable to reverse its stance on housing without alienating many of its supporters. As a result, Conservatives could hold the initiative on housing issues and increase the electoral salience of housing to their own advantage (Dunleavy, 1979; McAllister, 1984).

(11)

Lee and Yu (2012) find that widespread homeownership not only makes democracies more stable, it also supports the legitimacy of authoritarian regimes. By studying the housing markets in Singapore and Hong Kong they find that more homeownership increase the political support for authoritarian governance. More widespread and equal homeowner-ship can lead to a sense of equality among citizens (Barr, 1998) and it provides households with a sense of economic security that in turn relates to (increased) support for the ruling government.

Ansell (2014) uses both micro- and macrodata to analyse how housing prices affect the support for redistribution and the welfare state in the US and the UK. The results show that house price appreciation reduces the support for social security spending. This effect is consistent regardless of increases or decreases in house prices: decreases in house prices lead to increased support for welfare spending. This effect is more strongly represented among right-wing voters, since left-wing voters are primed to resist cuts to social spending due to ideological reasons. In a similar manner, Ansell finds that right-wing governments cut social spending more vigorously than left-wing governments during housing booms. This research suggests that social insurance policies are likely to increase in popularity among the public during a housing crisis, but that the political ideology from both voters as the parties in power matters.

Weiss (2014) studies the housing market in Israel after the 2011 housing protests. These protests were caused by a doubling in housing prices and rents in less than a decade. Weiss finds that the reliance of homeowners on the availability of credit compels them to function as an investor despite themselves wanting to make homeownership synony-mous with achieving economic security. Consequently, the pursuit of homeowners for this security contributes to a contradictory collective insecurity of the middle class through mortgage-enabled homeownership. Credit-leveraged accumulation thereby widens the gap between housing market growth and public welfare, even as they are widely seen as interlinked.

Andr´e et al. (2016) analyse three Dutch national elections and show that housing wealth is an important predictor of party choice. By using survey data, they quantitatively show that individuals in households with negative equity are more likely to vote for parties that want to abolish the fiscal benefits of homeowners and want to expand the social rental sector, while individuals with housing wealth are more likely to vote for parties that want to protect these benefits for homeowners. Respondents who see their housing wealth decrease are likely to change their voting preference from the first group to the latter, indicating that changes in household wealth result in changes in political behaviour.

(12)

3 The Dutch housing market: a background

In this chapter we give a short descriptive overview of the Dutch housing market. This overview gives us an insight in how the market works and how it affects the housing tenure of households. First, we show how it compares with housing markets in other countries and what makes the Dutch one unique. Secondly, we show that this uniqueness is the result of the institutional setting of the Dutch housing market which consists of several policies that are intended to affect both supply- and demand-side of the market. We discuss these policies since they affect the housing tenure of households directly.

3.1 Characteristics

The Netherlands is one of the countries with the highest level of household debt in the world. In Figure 1 the amount of household debt for OECD countries in 2015 is illustrated. This figure shows that Dutch households (after the Danish) have the highest level of household debt: 277 per cent as share of household’s disposable income. In comparison with the United States, that has known a severe housing crisis since the onset of the GFC, and neighbouring countries Germany, France and Belgium, household debt is two-and-a-half to three times as high. Comparing the Netherlands with some of the European countries (i.e. Portugal, Ireland, Spain and Greece) that were affected heavily by the European sovereign debt crisis shows that Dutch household debt is significantly higher.

Figure 1: Household debt of OECD countries, 2015

Hungary Latvia Slo v enia P o la nd Slo v ak Republi c Czec h Republic Estonia Italy German y Austria France United States Belgium Greece Spain Finland _Portugal United Kingdom Korea Canada Sw eden Ireland Switzerland Australia Norw a y The Netherlands Denmark 0 50 100 150 200 250 300 47 52 57 64 68 69 82 89 93 94 108 112114 118 122130 136 150 170 175178 178 211 212222 277 292 Household debt (% of income)

(13)

In Figure 2 we have plotted the level of household debt for the Netherlands over the period of 1995–2015. This graph shows that Dutch debt-to-income (DTI) levels have always been relatively high, and that they have been rising steadily since 1995. This growth continued until the GFC, when debt levels reached a high of almost 300 per cent of households’ income in 2010. Debt levels decreased slowly after that due to stricter rules for mortgage availability, but are still extremely high in comparison with other countries. Multiple actors such as the IMF and the Dutch central bank have warned that these high debt levels can be hazardous for the Dutch economy.

Figure 2: Household debt as share of income in the Netherlands, 1995–2015

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 100 150 200 250 300 350 Household debt (% of income)

Source: Author’s graph: data from OECD (2017)

Household debt is generally problematic for several reasons. First of all, a high debt means that a large part of income has to be devoted to this debt, which implies lower consumption levels and in turn depresses economic activity. Consumption can further be lowered by shocks in the presence of housing uncertainty, leading to more precautionary savings (Carroll, Slacalek, & Sommer, 2012; Guerrieri & Lorenzoni, 2011). Such a shock in consumption would provide a motive for pursuing expansionary macroeconomic policies such as monetary or fiscal policies aimed to restructure household debt (Eggertsson & Krugman, 2012; Guerrieri & Lorenzoni, 2011), but which also affect other sectors of the economy. Second, a high household debt in combination with an economic downturn (and thus higher unemployment) leads to a reduction in households ability to service their debt. This can lead to household defaults, foreclosures and fire-sales, which decrease the prices of real estate. These negative price effects in turn reduce economic activity through a number of self-reinforcing contractionary spirals, such as lower collateral values, negative wealth effects, negative impacts on bank balance sheets, and a credit crunch (IMF, 2012). A third problem is that household debt overhang can lead to various problems. First of all,

(14)

households can forgo investments that improve the net present value of their homes, which can slow down the recovery of the housing market. Second, foreclosures and bankruptcy as result of too large debt overhang can lead to vacant houses that negatively affect neighbourhoods social cohesion and can lead to more crime (Immergluck & Smith, 2006a, 2006b). In order to reduce the problems of debt overhang and foreclosures government involvement is necessary to lower the costs of restructuring debt, facilitating the writing down of household debt and preventing foreclosures (IMF, 2012).

The remarkable observation regarding the high Dutch household debts is that they hardly lead to the problems mentioned above. During the height of the housing crisis, the amount of households that were in arrears reached only 0.8 per cent in 2013 (Fitch, 2014). Banks play an important role in solving these payment problems: they can provide loan modifications and budget counselling aimed at balancing income and costs. The Dutch legal framework further contributes to solving early stage problems: the high level of welfare support strongly reduces the chance that a mortgage cannot be paid back. Besides housing, the large pension savings of Dutch households also provides them with a buffer for their old-age income. Besides households being in arrears, foreclosures are even rarer. If payment problems occur and these problems cannot be solved via the above manners, households prefer to voluntary sell their homes since it usually leads to better sales values than foreclosures. The number of foreclosures in the Netherlands is therefore low: foreclosures spiked to 0.08 per cent in 2011. As a result, banks losses have been modest over the past years (Dutch Banking Association, 2014).

Another remarkable observation is that the high debt levels of Dutch households are not accompanied by very high homeownership levels. In Figure 3, we have depicted the distribution of population by tenure status in European countries. As can be seen, around 67 per cent of all Dutch dwellings are owner-occupied, from which 7 per cent without a mortgage. This homeownership level is well below the average of around 75 per cent in the rest of Europe. The figure further shows that 26 per cent of all Dutch homes are social rental ones, and that the remaining 6 per cent belong to the private rental sector. In comparison with other countries we can see that the social rental sector is the largest, while the private rental sector is relatively small. The Dutch housing market thus has a rather uncommon division of tenure types, accompanied by high household debts. Besides the relatively stability of household debt levels, the tenure division has also been relatively stable over time. In Paragraph 2.2 we will explain how and why this division is so stable, how it is established and why Dutch household debt is so high when only 60 per cent of all dwellings is owner-occupied with a mortgage.

(15)

Figure 3: Distribution of tenure status of European countries, 2015 Latvia Slo v enia P oland Slo v ak Republi c Greece Italy Czec h

Republic Hungary Estonia Austria _German

y F rance Spain Ireland United Kingdom Portugal Switzerland Belgium Finland Luxem b ourg Denmark The Netherlands Norw a y Iceland Sw e d e n 0 10 20 30 40 50 60 70 80 90 100 P ercen tage(%)

Owner mortgage Owner no mortgage Private renter Social renter

Source: Author’s graph: data from Eurostat (2017)

In Figure 4 the housing prices of Dutch real estate over the last two decades are plotted. This figure leads to two important observations. First, prices have increased sharply since 1995. Where a average house costed around 85,000 euro in 1995, the price increased to around 250,000 in 2016. This increase lies well above the increase in the national inflation rate for this period. The GFC and the concurrent housing crisis have reduced prices in 2011–2013, but in 2016 prices have almost returned to pre-crisis levels. The increase in prices have been most prevalent in the municipality of Amsterdam, where prices have more than tripled in twenty years. In the rural province of Groningen prices have also almost tripled, but the price differential relative to Amsterdam grew from 1.5 to 2.0: where average prices in Amsterdam in 1995 were only 1.5 times as high as in Groningen, they are now twice as high. The main reason for the increase in this difference is the higher demand for housing in Amsterdam relative to Groningen. Economic growth in Amsterdam is high, which attracts citizens who demand housing. A reverse development is seen in Groningen, which is slowly becoming a shrinking region. Furthermore, Amsterdam is a popular destination for tourists, increasing the demand for hotels or other forms of renting which increases the prices for land. Lastly, there is less room to build real estate in Amsterdam than in Groningen, making the supply of housing more inelastic, which increases the price further.

(16)

Figure 4: Average nominal house prices in the Netherlands in euros, 1995–2016 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 50,000 100,000 150,000 200,000 250,000 300,000 350,000 400,000 Av erage house prices in euros

Source: Author’s graph: data from CBS (2017) Amsterdam

National average Groningen

Why are these changing and diverging prices of real estate relevant? Strong fluctuations in housing prices affect the wealth of households directly. When housing prices decline, the possibility exist that a mortgage becomes ’underwater’: the value of the mortgage is then higher than the market value of the property. This affects the ability of households to refinance or sell their homes. However, too high house prices are also be hazardous: housing becomes less available for lower incomes and the chance of a speculative bubble increases. Fluctuations in housing prices can make housing salient, leading to citizens thinking more about their position in the housing market which can affect their economic behaviour. Furthermore, diverging house prices can lead to country-wide inequalities, which can lead to political saliency of housing as well.

3.2 Market interventions and policies

As can be seen, the Dutch housing market is quite unique: high debt levels, low fore-closures, low levels of homeownership and a tripling in the prices of real estate in just twenty years. This uniqueness is largely due to government policies and the institutional setting of the housing market. As the word market indicates there are two forces at work: a demand- and a supply-side. The Dutch government has introduced policies and regulations to control both sides.

3.2.1 Supply-side influence

The Dutch government influences housing supply in two ways: by spatial planning policy and by rules and support for housing corporations which include subsidies and regulations regarding a maximum of rental prices.

(17)

Spatial planning

Dutch spatial planning is mainly influenced by a legal framework in which the national government provides rough guidelines for real estate which trickle down to the provincial and the municipal level. Municipal zoning plans then designate a detailed function (i.e. housing, industry, offices, shops) to each lot of land. This Spatial Planning Act is legally binding and the government largely determines the supply of residential land. Market forces can only affect this to the extent that government institutions are sensitive to them. Even if these institutions are responsive to price signals, then legal procedures significantly delay such responses. The main reasons for government intervention are the preservation of landscape heritage and open space and the provision of public goods. Due to the small size and the population density of the Netherlands, the government sees it fit to not leave this up to the market (Rouwendal & Vermeulen, 2007: 10–14).

Housing corporations and rental prices

Housing corporations in the Netherlands are fairly independent, with the government as role of financier, regulator and supervisor. They are expected to provide affordable and liveable housing for Dutch citizens with a low income. Housing corporations can fulfil this task thanks to their accrued wealth that comes partly from the government and the implicit and explicit subsidies and guarantees from which they profit. Due to the government support, the housing corporations are able to lower the rental price of their houses below the market price and even below the maximum the government has bound the rental prices to (Donders, Dijk, & Romijn, 2010: 7). Due to these lower-than-market prices, social renters are implicitly subsidized, which largely explains why this form of housing is so large in the Netherlands.

3.2.2 Demand-side influence

The demand-side is largely influenced by the government in the form of several policies, which are set to steer the demand for housing in a certain way. We will shortly describe the three most important policies for our research, the rationale behind them and their effects on the housing market.

Mortgage Interest Deductibility

The most prominent policy in the Dutch housing market is the mortgage interest de-ductibility (MID). The MID is a fiscal policy that gives homeowners that have mortgaged their house a tax reduction. Homeowners can reduce their taxable income by the amount of interest paid on the mortgage which is secured by the collateral of their houses. This means that the larger the mortgage, the higher the interest costs, and the higher the income, the larger the benefits of the MID. This is exactly the reason why Dutch house-holds hold so much debt: they are incentivized to hold debt since it is fiscally beneficial.

(18)

The MID is a subsidy that is constructed in such a way that households with all levels of income can benefit. This means that almost all households profit from the subsidy but simultaneously contribute to it through taxes. The MID is therefore some sort of money circulation scheme which is accompanied with a loss of welfare. The subsidy can be typed as a form of bounded income transfer: households can only benefit if they buy a house by using a mortgage. This means that the money circulation of the MID has the disadvantage of restricting the choice of citizens in consuming their income and that they may spend more resources on housing than socially optimal (Donders et al., 2010). National Mortgage Guarantee

The National Mortgage Guarantee (Nationale Hypotheek Garantie: NHG) is a form of upfront premium-based insurance established in 1993. The NHG covers losses from non-voluntary sales of the primary residence. This means that homeowners who bought their residence with NHG-coverage and are in arrears because of unemployment, disability, widowhood or relationship dissolution are eligible for cancellation of their debt in case of a non-voluntary sale (Andr´e & Dewilde, 2016: 5–6). However, this only works when the purchasing price of the house lies below the NHG-norm. During the height of the housing crisis, approximately 4000 NHG-losses where covered.

Social renting

Besides focusing on homeownership, the Dutch government is also promoting the social renting sector. Besides implicitly subsidizing this sector as seen before, the Dutch govern-ment also provides an explicit subsidy in the form of renting allowance, to make housing more affordable for households with low or no income.

3.2.3 Arguments for housing policies

The Dutch government thus influences the functioning of the housing market with several interventions. There are several motives as to why, which mostly have to do with market failures such as externalities. First, as stated, spatial planning cannot be left to the market. Expected is that this would lead to segregation, an uneven construction of houses and a market equilibrium with too high prices and too few houses. Second, as argument for the housing subsidies, it is often put forward that in the absence of these subsidies people would not spend enough money for proper housing. This may lead to impoverishments of housing districts and more homeless people. In light of this argument, housing can be seen as a merit good: a commodity that requires public financing for its needs and positive externalities. Third, the government uses housing policies to reduce income inequalities (Donders et al., 2010). However, estimated is that only half of the costs of these policies go to the lower income group of Dutch households (Romijn & Besseling, 2008).

(19)

3.3 Summary

The Dutch housing market is unique: high debts and high prices accompanied with low levels of homeownership and few foreclosures. The prevalent housing policies of the gov-ernment have shaped the division of the housing stock as it is now. Both homeownership (via the MID) and social renting (via housing corporations and renting allowance) are being subsidized, where private rental is not. This explains why only around 6 per cent of all dwellings are rented privately: it is fiscally beneficial to mortgage a house and if you have a low income, you will get subsidized in the social rental sector. This division of the housing stock and the accompanying subsidies is important since they affect the housing tenure and wealth of Dutch households. In the next chapter (Chapter 4) we will dive further in the housing position of Dutch households when we construct our theoretical framework.

(20)

4 Theoretical framework and hypotheses

This chapter describes our theoretical framework and corresponding hypotheses. The framework is constructed in four parts: we first theorize how housing wealth and wel-fare state support connect using the permanent income hypothesis. We then show how political parties consider welfare state and housing issues and aggregate them to policy outcomes. Third, using public choice theory we theorize how households use housing to make rational voting decisions. Lastly, we argue how housing tenure and political saliency might help to explain party choice. We also provide our hypotheses in this last part.

4.1 Housing wealth and permanent income

The literature review of Chapter 2 shows that economic insecurity regarding housing wealth is important. Homeownership, which is viewed as the primary source of house-hold wealth, is unequally distributed among social groups. Like other sources of wealth, housing wealth could influence life by facilitating consumption smoothing, longer time horizons and intergenerational transfers (Zavisca & Gerber, 2016: 350). Homeownership could amplify the stratifying effects of wealth that have been identified in domains such as education (Torche & Spilerman, 2009), living standards (Dimova & Wolff, 2008) and happiness (Headey, Muffels, & Wooden, 2008). Furthermore, Castles (1998) argues that homeownership is inversely related to the welfare state. This is especially true for elderly, who may use homeownership to complement their public pension, since it often leads to reduced housing costs in old age: when mortgages are paid off the house becomes an important asset. Castles also shows that settler societies with high levels of home-ownership prior to the emergence of a welfare state were least likely to develop robust public pensions, because freehold ownership of housing substantially reduced the income requirements of the home-owning elderly. In light of this, Kemeny (2005) and Schwartz (2012) argue that the deregulation and erosion of the welfare state in the 1990s and 2000s made housing increasingly important as it was able to act as a substitute for the welfare state.

Housing wealth thus is able to changing the economic behaviour of households. We con-nect this housing wealth with the permanent income hypothesis. The permanent income hypothesis, dating back to Modigliani and Brumberg (1954) and Friedman (1957), states that an individual’s present consumption depends not on their present income, which may be affected by transitory fluctuations, but rather on their permanent income across their life, which includes their wealth. Housing wealth thus adds to an individual’s perma-nent income: a rise in the value of the house raises permaperma-nent income and can therefore increase consumption. Citizens can sell their house or borrow against it as collateral in order to sustain consumption during periods of lower labour market income,

(21)

includ-ing retirement (Ansell, 2014: 3). Carroll (1997) uses the permanent income hypothesis and notes that households may face liquidity constraints when out of the labour market and are unable to borrow against future income. This creates an incentive to engage in precautionary saving in order to build a buffer of assets to ensure consumption during periods of transitory income loss (Carroll et al., 2012). Housing is therefore also able to hedge against uncertainty in the labour market.

The price of housing is paramount using the permanent income hypothesis. A rise in the value of the house may boost a homeowner’s permanent income, even when labour market income decreases. Contrary, decreases in house prices reduce both permanent income and the ability to use housing as a buffer stock (Ansell, 2014: 3). We can now see why housing insecurity is important. Besides leading to stress, anxiety and physical health problems (Nettleton & Burrows, 1998), it reduces an individual’s ability to value their permanent income. This also explains the connection of housing and the welfare state as seen above. The ability to use housing as a means to smooth consumption when unemployed and as a way to fund retirement provides citizens with steady levels of consumption during periods of lower income and can thus complement or substitute the social transfers (e.g. unemployment insurance, pensions) of the welfare state. Housing wealth hence acts as a form of private insurance relative to the social insurance of the welfare state. Higher house prices means more private insurance and reduces the demand of homeowners for social insurance. In contrast, decreasing house prices increase homeowners’ support for social insurance since the value of their private insurance declines (Ansell, 2014: 4). How-ever, the riskiness of the asset determines the demand for social insurance: if the sale of a house is uncertain, individuals are expected to demand more social insurance to hedge against the possible income loss from selling their house (Iversen & Soskice, 2001). We discuss three important qualifications regarding housing wealth. First, individuals do not see the realized value of their asset unless they sell their house. Therefore, even though citizens typically do not know the precise value of their house ex ante, they will use current housing conditions to shape their price estimate by using information extracted from the local or national housing market. However, these housing conditions do not always give the right price. For example, many homeowners, politicians and banks dur-ing the 2004–2006 housdur-ing bubble believed that real estate prices would continue to rise in value (Case, Shiller, & Thompson, 2015; Shiller, 2015). Second, housing is relatively illiquid and has high sale transactions costs, reducing its value as a hedge. Therefore, it is likely that the effectiveness of housing as a cushion against income loss is conditional on the time frame in which the property is sold. Whereas retirements are planned over a long horizon, unemployment is often an unexpected event. It is thus likely that hous-ing wealth reduces demand for social insurance programs for old age more so than for

(22)

programs targeted at unemployment. Lastly, many homeowners hold mortgages on their house and thus hold relatively little housing wealth. Accordingly, individuals with more housing wealth show the least support for redistribution and social insurance since they face less risk of losing their house and they have accumulated enough wealth to substitute social insurance (Ansell, 2014: 4).

4.2 Political parties, housing and the welfare state

In this part we describe how the above microfoundations of housing and welfare state support aggregate to the national political party-level. We classify political parties on housing issues and the welfare state to make a gross distinction between three sorts of parties. The first group consists of political parties that are strong supporters of home-ownership and a small welfare state. This group largely corresponds with right-wing parties. These parties are generally in favour of a small welfare state and therefore dis-proportionately represent homeowners (Kingston, Thompson, & Eichar, 1984; Verberg, 2000). As seen above, homeowners with housing wealth prefer private insurance in oppo-site to the social insurance of the welfare state. When house prices rise, housing wealth increases and more homeowners prefer less social security transfers. The political parties in this group channel the preferences of this constituency for lower insurance spending. Rising house prices create an opportunity for right-wing parties to cut social spending as their electoral base becomes more inclined towards that ideological end (Ansell, 2014: 5). We classify this group as ’pro-ownership’.

The second group, which is classified as ’pro-welfare’, consists of political parties that are in favour of a larger welfare state and thus more social insurance spending. Consequently, this means that these parties are less supportive of private insurance, and therefore the housing wealth that accompanies homeownership. The political parties in this group are mainly left-wing ones, with a constituency that is in favour or in need of social security, and usually consists or renters in stead of homeowners. As shown by Margalit (2013) and Ansell (2014), left-wing parties are primed to resist cuts to social spending regardless of economically derived preferences. Ideology thereby operates as a filter on economic changes such as house prices. So, when house prices increase, pro-ownership parties de-crease social spending due mainly to their constituency, where pro-welfare parties inde-crease it, on both ideological grounds and due to their constituency.

The last group is more ’balanced’ or ’centrist’ which consists of political parties that strive for a form of equilibrium between social and private insurance. These parties will not excessively scale down the welfare state but will also not enlarge it at the expense of homeownership, and are usually in favour of the status quo.

(23)

We can depict this distinction of political parties in the following political spectrum: Figure 5: Political party spectrum

Pro-ownership Pro-welfare

Balanced

More ownership More welfare

This figure shows the position of our three political party groups. We have used the two axes on which we distinguish political parties: the x-axis describes the level of support for homeownership (private insurance) where the y-axis describes the level of support for the welfare state (social insurance).

4.3 Rational voting

We use public choice theory in order to hypothesize how households use their housing position to make rational decisions regarding their voting behaviour. We argue that households are economic agents that try to maximize their welfare (Caporaso & Levine, 1992: 79–86). This welfare maximisation is achieved by accumulating a high as possi-ble permanent income. For homeowners this means that they want to secure as much housing wealth as possible, while for renters this means that they want to pay as low as possible rent and get high welfare transfers. Households are not able to secure their welfare maximization objectives through market exchange, but have to see their interests heard in the political realm (Buchanan, 1987): households have to vote for political par-ties that look after their interests. For homeowners this means that they want to vote for pro-ownership parties while renters benefit from voting for pro-welfare parties. The above theory is subject to multiple assumptions. First, it is fundamental that house-holds have perfect information. They need to know how the housing market works and in which state it is in order to know if housing prices will raise or decrease. Regarding the welfare state they need to know which transfers are present and what the level of these transfers are. They further need to know which political parties can help them to maximize their welfare (and thus permanent income). Second, there exist ’transaction

(24)

costs’ in order for households to let their preferences known: they have to ascertain which parties promise to pursue households’ interests and subsequently vote in elections. We assume that the benefits of voting are higher than the costs, and that households are aware of it. Third, households are rational: they use the information they have to make rational decisions about their housing position and voting behaviour.

4.4 Housing tenure and voting

in this paragraph we connect the above theory of rational voting with our distinction of political parties and the theory of housing wealth and permanent income, and apply it to the Dutch housing market. We divide Dutch households into four housing groups with different levels of housing tenure, housing wealth and welfare state support. Using this division as framework we are able to construct hypotheses for each individual household group.

4.4.1 Homeowners with positive equity

The first group consists of owner-occupiers that have positive housing wealth. This group has an equal or lower mortgage debt relative to the value of their house, leading to a positive amount of equity. It could be that the households in this group have either bought their home without a mortgage, have only financed their home with a small mortgage, or have seen the price of their house rise above their mortgage value. Due to their positive wealth, this group has privately insured themselves against unemployment and retirement. They therefore do not require the social insurance provided by the welfare state and thus prefer less redistribution from the government through the welfare state (Andr´e & Dewilde, 2016). It is not in the best interest of households in this group to vote for a party that is not supportive of homeownership. For the Netherlands this largely means parties that want to dismantle the fiscal benefits (the MID) for homeowners. Dismantling the MID will make the financing of a house more expensive which cuts demand for houses, leading to lower house prices, which in turn leads to less housing wealth and lower permanent income. Furthermore, reducing the MID will lead to higher taxation costs which will reduce permanent income threatening the permanent income of this group even more. We therefore expect that homeowners with positive equity will vote for at least a balanced political party or one that is pro-ownership.

Hypothesis 1: Homeowners with positive equity are more likely to vote for balanced or pro-ownership parties.

(25)

4.4.2 Homeowners with negative equity

Besides homeowners with positive equity there is a group of homeowners that consists of households that are in negative equity. The households in this group have a mortgage debt that is larger than the value of the house. They have financed their home with a mortgage larger than its buying value or have seen the price of their home decline over time. The households in this group are therefore ’underwater’ and hold negative housing wealth. Since the households in this group are unable to insure themselves against labour market risks and retirement, they rely on the welfare state and its social insurance policies. We can expect two reactions of this group. First, they are expected to vote for a party that is aiming to stabilize the housing market, increasing sales and increase real estate prices in order to get out of negative equity, decrease economic insecurity, raise housing wealth and thus their permanent income. As previously mentioned, if the MID would be reduced, this would lead to lower house prices. This is not in the interest of people who are in negative equity since they see their debt grow and their permanent income decrease. So, through housing insecurity, we expect that households with negative equity are more likely to vote for a pro-ownership party. An alternative reaction could be that homeowners vote for a political party that is supportive of the welfare state as these promise more generous social transfers in the case of income losses by negative live events. If these homeowners think that they are unable to get out of negative equity, are unable to sell their home and have no NHG-coverage, they are expected to vote for pro-welfare parties. These parties are generally supportive of more welfare transfers and a large social rental sector which people could use to fall back on should they lose their home (Andr´e et al., 2016).

Hypothesis 2: Homeowners with negative equity are more likely to vote for a political party that is pro-ownership or pro-welfare.

4.4.3 Social renters

This group consists of renters and therefore hold per definition no housing wealth. The households in this group depend on the implicit and explicit subsidies provided by the gov-ernment. Housing corporations’ subsidies to decrease rental prices implicitly contribute to the permanent income of the households in this group. Furthermore, social renters receive renting allowance, which increase permanent income as well. The households in this group are therefore incentivized to support political parties that will protect or even extend these subsidies. Pro-welfare parties are most likely to do this: they oppose the private insurance of homeowners which do not benefit social renters and they support the social insurance from which these renters profit.

(26)

4.4.4 Private renters

The final group, the private sector renters, is expected to be the most diverse one. This group can be grossly divided in two subgroups. The first consists of renters that want to buy a house but are currently not able to. By buying a house they want to profit from raising housing prices to accumulate housing wealth and thereby raise their permanent income. This subgroup is expected to vote for parties that are supportive of homeown-ership since they make homeownhomeown-ership more afforable. The individuals in this group are therefore expected to vote for balanced or pro-ownership parties. The second subgroup consists of renters that want to be renting. Despite the fact that this form of housing is not subsidized in the Netherlands, it is possible that these households still want to rent due to, for example, the flexibility that comes with it. This group is expected to not vote for parties that are supportive of homeownership since these parties will not pursue their interests. These parties are more likely to raise housing prices (e.g. by holding on to the MID) which will likely cause raises in rental prices as well and thus lower the permanent income of these households. We therefore expect that the households in this subgroup are more likely to vote for a balanced or pro-welfare party. All in all, we expect that most private renters will vote for a balanced party.

Hypothesis 4: Private renters are expected to vote for a balanced political party.

4.4.5 Political and societal saliency

An important aspect that is expected to influence voting behaviour is political and societal salience. For example, the GFC and following housing crisis decreased the prices of houses in the Netherlands (see Figure 4) which increase saliency. As previously mentioned, homeowners may not be aware of their housing wealth until they sell their homes. During the crisis many homeowners were unable to sell their house, saw the value of their house decline and were therefore made conscious of their wealth. Housing thus became societal salient: more homeowners became aware of their housing wealth. This reflected in the political realm, where pro-ownership parties powerfully stated that they want to keep the MID and restore the housing market for homeowners, while pro-welfare parties wanted to focus more on the welfare state. Housing therefore became politically salient as well. We argue that this societal and political saliency makes homeowners aware of their decreasing wealth, and want political parties to restore the market for them. We therefore expect homeowners to vote more for pro-ownership parties during the housing crisis.

Hypothesis 5: Homeowners are expected to vote more for pro-ownership parties during the housing crisis (2010 and 2012 elections).

(27)

5 Empirical set-up

This chapter consists of two paragraphs. In the first paragraph we explain the use of our data and how we operationalize our dependent and independent variables. In the second we describe our methodology and how we apply the data to check our hypotheses and answer our research question.

5.1 Data and operationalization

In this research we use the Longitudinal Internet Studies for the Social sciences (LISS) panel dataset that contains a representative sample of Dutch households. All respon-dents provide information on their background, politics and values, and their economic situation, which includes income, assets and housing (CenterData, 2017). The two most important questions for our research are which political party respondents voted for dur-ing the last national election and what their current housdur-ing situation is.

A common problem with these sorts of surveys is missing or incorrect data. Not all respondents want or are able to fill in all questions correctly. Since we need to connect voting behaviour with housing tenure status it is important that we have correct informa-tion regarding voting and housing. We omit all respondents who had no informainforma-tion on housing and we cross-checked the housing information of respondents with other survey rounds in case of large inconsistencies.

We will look at four election rounds. The 2006 election predates the financial crisis, the 2010 and 2012 ones took place during the crisis, and the 2017 election took place after the crisis. Each election round (wave) includes up to 12000 respondents, except the 2017 one which is still in development and therefore includes around 10000. We first analyse the complete sample after which we look at each election round separately. Overall, we combine the results of the complete sample and the four election rounds to see how the relation between voting behaviour and housing change over time.

5.1.1 Dependent variable

Our dependent variable in this study is party choice. We classify Dutch political parties on homeownership and welfare state policies in order to divide the parties in three groups. The first group can be classified as pro-welfare and want to limit or abolish the MID completely, defend or extend the social rental sector and are in favour of large welfare state benefits. The second group is more balanced and want an equilibrium between the social rental sector and homeownership. They will not excessively scale down the benefits of homeownership at the expense of the welfare state. The last group is classified as pro-ownership and wants to maintain the MID or change it minimally, with more stringent

(28)

rules for social renting and a smaller welfare state. The distinction in these groups is similar to a left-centre-right distinction, but due to the focus on housing there are some small differences. We have summarized the parties’ position of the largest Dutch political parties into the three groups in Table 1.12

Table 1: Party positions of Dutch political parties

Pro-welfare Balanced Pro-ownership Large welfare state benefits

and social insurance, large social rental sector, limit or abolish MID

Average welfare state benefits, average social rental sector, moderately cut back on MID

Small welfare state with more private insurance, focus on homeownership, maintain MID or limited changes

Groenlinks (green-lefts) PvdA (labour party) VVD (right-wing liberals) SP (socialist party) D66 (social liberals)

50Plus (elderly party) CDA (Christian-democrats) PvdD (animal party) CU (moderate Christians)

SGP (orthodox Christians) PVV (freedom party)

Source: Andr´e et al. (2016)

5.1.2 Independent variables

Here we will list our variables that are expected to influence party choice. As stated, we mainly look at the housing position of households, but we also control for a list of background variables included in the LISS panel. A summary of these variables and how they are defined can be found in Appendix A2.

Housing tenure: The main independent variable of interest is housing tenure. As described in our theoretical background we divide households into four groups: home-owners with positive equity (mortgage debt equal or smaller than value of the house), homeowners with negative equity (mortgage debt larger than value of the house), social renters (rents and receives renting allowance) and private renters (rents and receives no renting allowance). We dummify these four groups and use homeowners with missing housing wealth as our reference category since they proxy for the average household. Background variables: To fully control for the effects of housing, we also control for households financial wealth, which consists of the amount of savings and investments held by the household, and households income, which is usually labour market wage.

1_{In order to not overcomplicate this study, we assume that political parties’ position on housing is}

constant over time. This is probably not the case, but since our focus lies on the household level we want to keep our analysis of political parties straightforward. In chapter 7 we provide some recommendations for further research that focus on political parties and housing.

(29)

As previously mentioned, we argue that housing is able to affect voting irrespective of households’ income and financial wealth. We therefore include these variables to isolate the effect of housing on voting behaviour. Furthermore, as argued by Brooks and Brady (1999), income and wealth have a stable effect on party preferences and a political impact on the outcomes of elections. Households with more wealth and income are expected to vote more right-wing, which largely corresponds to pro-ownership parties.

We add several other background variables to our models that are explanatory of voting behaviour. The first background variable is gender. As argued by Inglehart and Norris (2000), women are expected to vote more left-wing. This is confirmed by Abendsch¨on and Steinmetz (2014), who researched this difference in eighteen European countries, including the Netherlands. A vast literature exists on the effects of age on voting pat-terns (Beck, Dalton, Greene, & Huckfeldt, 2002; Bromley & Curtice, 2002; Goerres, 2008; Harris, Lock, Dermody, Hanmer-Lloyd, & Scullion, 2010). They argue that the voting pattern of people can be influenced among others by their life cycle, generation, social contexts and personal attitudes, which are captured by age. We also use age as a proxy for year of residency. Households who just bought a home (usually younger citizens) are more likely to be in negative wealth while elderly are more likely to have positive housing wealth. Employment status is another important factor. In light of economic voting, where citizens vote based on the economic performance of their governments (Lewis-Beck & Paldam, 2000), voters can punish the incumbent government when they become un-employed. This was clear during the GFC, where voters who were negatively affected by the crisis, through job losses or lower wages, punished mainstream parties and turned to challenger parties (Hobolt & Tilley, 2016). Education is another important variable that we will include. Education increases citizens’ knowledge of political issues such as housing, leading them to make more rational choices regarding these issues (Hanushek, 2002; Milligan, Moretti, & Oreopoulos, 2004). We will also include the urban status as explanatory factor, since some regions see more fluctuations in the housing market than others (see Dobbs et al. (2015) and Figure 4). We follow the division in the LISS-panel and make five dummies: extremely urban, very urban, moderately urban, slightly urban and rural. Additionally, we include political trust. As argued by Hobolt (2006), voters who have a low level of trust in the government, have a lower turnout or are expected to vote for anti-establishment parties. Political trust is the average of a 11-point scale for the level of trust in the government, parliament, politicians and parties.34

3_{Information that in unfortunately unavailable in the LISS panel is the amount of pensions a household}

has accrued. As argued by Castles (1998), pensions are able to substitute housing as a nest egg, and can therefore influence our results.

4_{We are unfortunately unable to control for all factors affecting voting behaviour. For example, citizens}

can vote based on gut feeling or other factors that are not included in the LISS panel. Other method-ologies or ways of research such as interviews are better able to take these into account, but lack the general scope we are trying to achieve in this study.

(30)

5.2 Methodology

In this paragraph we will describe the choice of our methodology and how we take several issues into account. For our research we use a multinomial logistic regression method. Since we have a dependent variable that is divided in three categories, we are unable to use traditional (binomial) logistic regressions. We will estimate versions of the following linear logit equation:

Logit(P arty Choicei) = β0+ β1Housing T enurei+ β2Controlsi+ i (1)

where Logit(P arty Choicei) is the logit effect of a certain household i that votes for a

certain political party. Our main variable of interest is Housing T enure and its coeffi-cient β1. Controls is a vector that contains the background variables mentioned in the

previous paragraph. The is the error term or random disturbance. This disturbance contains measurements errors, omitted variables and other factors. In order to correct for heteroskedasticity and to produce unbiased and consistent standard errors of the es-timators, we use robust standard errors.

We report the logit effects of the above model for our full sample and our four election rounds. We show the estimated coefficients in regression tables, together with the results of chi-squared tests that signify how significant housing tenure is in our models. Due to the complexity in explaining and comparing the estimated logit effects (Mood, 2010), we also calculate the predictive probabilities for each category of housing tenure. These probabilities reflect the likelihood a certain household group is likely to vote for a given party group and are calculated using the logistic function (Wooldridge, 2010: 442). To project the probability a certain household group votes for a certain party group using the coefficients in the above function gives:

P r(P arty Choice) = e

β0+β1Housing T enure

1 + eβ0+β1Housing T enure

(2) Besides providing the estimated coefficients in our tables we provide the predictive prob-abilities in graphical form to aid us in our analysis.

A potential issue for this research, and with regression analysis in general, is omitted variable bias (OVB). OVB occurs when important variables are left out of the regression function. This can inflate or deflate certain coefficients and thus lead to inaccurate results. In our case this can be applied to factors that influence the voting behaviour of house-holds, but which are not included in the panel. We hope by including our background variables we can reduce OVB as much as possible, but we are unable to completely avoid it.

(31)

6 Results and analysis

In this chapter we analyse if and how housing tenure is able to influence voting behaviour. We start with general (baseline) models for each election year, with we expand with our background variables afterwards. We first provide the estimated coefficients of our regression models in table form after which we calculate the predictive possibilities which we depict graphically. We then check our hypotheses and answer our research question. Afterwards, we perform multiple robustness checks to gauge whether our results hold up under different circumstances.

6.1 Baseline models

In order to interpret and explain our results, we start with baseline (uncontrolled) models using the complete sample (all election years included) first. We then split the sample for each separate election year and apply the same model. We use housing tenure as the only explanatory variable in these models and therefore omit all of our background variables. We later add these variables to our extended (controlled) models. The results of the complete sample baseline model can be found in Table 2.

Table 2: Baseline model (complete sample) Complete sample

Pro-ownership Pro-welfare

vs Balanced vs Balanced

Housing tenure: Reference = Missing wealth

Positive .0360 -.0034 equity (.0555) (.0598) Negative .4175*** .2858** equity (.1077) (.1188) Social -1.3043*** .6646*** renters (.1644) (.0863) Private -.7271*** .3938*** renters (.0638) (.0494) Constant -.9092*** -1.0833*** (.0267) (.0284) Obs. = 15301 χ2 _{= 413.03} Prob. χ2 _{= 0.0000}

Note: Robust std. errors given in parentheses. *** p<0.01, ** p<0.05, * p<0.1

The first column of the table shows our four housing tenure groups. As stated, we use a fifth category as reference category: homeowners with missing housing wealth. In the sec-ond and third columns we show our logit coefficients and correspsec-onding robust standard

(32)

errors. The coefficients in the second column show the effects of housing tenure on voting for a pro-ownership party, relative to voting for a balanced party. All housing tenure co-efficients are significant except the ones for homeowners with positive equity, which shows that housing tenure is a good predictor of voting behaviour. This is also confirmed by the chi-squared test which shows that housing tenure (on the whole) is significant in the model. The interpretation of the estimated coefficients is not straightforward, since they are estimated with regards to the references (balanced parties and homeowners where housing wealth is missing). They can be analysed roughly as follows: a negative coeffi-cient implies that that household group is less likely to vote for that party group, while a positive coefficient implies that it is more likely to vote for that party group, relative to the other groups. Looking at social renters, we see that they are less likely to vote for a pro-ownership party, and more likely to vote for a pro-welfare party, relative to the others.

Because of the difficulty interpreting the estimated coefficients, we calculate and depict the predictive probabilities. As previously mentioned, the calculation of these probabili-ties depend on the logistic function. We can simply fill in the coefficients in this function in order to get the predictive probabilities. For example, for homeowners with positive equity we can calculate the likelihood that they vote for a pro-ownership party as follows:

P r(P ositive equity|P ro − ownership) = e

−.9092+.0360

1 + e−.9092+.0360_{+ e}−1.0833−.0034 = 0.238

So, the predicted likelihood that a homeowner with positive equity votes for a pro-ownership party is 23.8 per cent. Similarly, we can calculate the likelihood that these homeowners vote for a pro-welfare party:

P r(P ositive equity|P ro − welf are) = e

−1.0833−.0034

1 + e−.9092+.0360_{+ e}−1.0833−.0034 = 0.192

which equals to 19.2 per cent. Since we have divided political parties in three groups, we know that the likelihood of voting for a balanced party should equal to 57 per cent. In order to make sure, we calculate:

P r(P ositive equity|Balanced) = 1

1 + e−.9092+.0360_{+ e}−1.0833−.0034 = 0.570

which indeed equals to 57 per cent. These percentages show that homeowners with positive equity are most likely to vote for a balanced party, and are largely indifferent between pro-homeownership and pro-welfare parties. However, we have to take into account that the estimated coefficients are insignificant and we have not yet added our background variables.

The political economy of the Dutch housing market : housing tenure and voting