The effect of study loan debt on homeownership among young Dutch adults

(1)

The effect of study loan debt on homeownership

among young Dutch adults

Rutger-Roger Van der Vlies S3483444

Faculty of Economics & Business University of Groningen Master Thesis MSc Finance Supervisor: Dr. Carolina Laureti

Word count: 10,912 January 2020

Abstract

This paper investigates the effect of having a study loan debt on homeownership among young Dutch adults. In this study, we use data from the DNB Household Survey (DHS). To examine the impact of having a study loan debt on homeownership, we perform probit regressions and marginal effects, controlling for household demographic and socio-economic characteristics. Obtaining a more reliable estimate of the results, the logit model serves as a robustness check. In line with our expectation, we find that young adults with a study loan debt are less likely to own a home than young adults without a study loan debt.

(2)

2 1. Introduction

Study loan debt and mortgage are large debts for young Dutch households. The average debts of households in 2017 consist of 1.9% student loan debt, 86% mortgage and 12.1% remaining debts (CBS, 2019). Interestingly, homeownership and study loan debt have had an opposite trend.

On the one hand, the homeownership rate for young Dutch adults is decreasing. The number of mortgage applications to buy a home under young adults decreased by 9% in 2018 compared to the year 2017 (De Voogt, 2018). A decline in home sales is an explanation for the decreasing mortgage applications under young adults (De Voogt, 2018). Another factor that makes homeownership more difficult are the changed mortgage rules. Since the first of January 2018, the Dutch government sets new rules about getting a mortgage. An example of a rule is the so-called loan to value (LTV) rule, wherein the value of the mortgage cannot exceed the purchase value of a house (Van Veluwen, 2018). The new rules require more savings from people buying a house, so the households are able to cover the extra payments above the LTV.

On the other hand, the total student loan debt keeps on rising in the Netherlands. In 2017, compared to 2009, total study loan debt increased by €1.0 billion to a total amount of €11.2 billion, indicating an increase of 143% (Smit, 2018). The higher the study loan debt, the more difficult it becomes to save money because of debt repayments (Van Veluwen, 2018). The decreasing homeownership rate among young adults and the rising study loan debt makes it interesting to examine the impact of study loan debt on homeownership among young adults.

(3)

3 while in 2017, the percentage increased to 73% of students having a study loan debt (Schonewille et al, 2017).

The average study loan debt is growing every year. The consequence for households is higher debt repayments. The higher debt repayments are a liquidity constraint for households. Add the changed mortgage rules to this situation and there is a relevant problem for young adults who want to buy a home. This situation is a major policy concern given that young adults are important for the growth of the economy and the housing market in particular (Brown and Caldwell, 2013) and the stage of life in which young adults find themselves is a crucial time for wealth acquisition. Wealth creation is an advantage of homeownership, except when the mortgage value exceeds the value of the home, which is known as underwater (Letkiewicz and Heckman, 2018). The National Association of Realtors (2012) emphasizes the relevance of first-time buyers in the recovery of the housing market because they help homeowners to sell and trade. Despite the attention to this topic from media, empirical studies on the influence of study loan debt on homeownership are rare.

Having study loan debt is associated with both advantages and disadvantages. For example, the research of Avery and Turner (2012) argues that study loan debts motivate students to invest in their education, which is in line with the human capital theory. In contrast, having a study loan debt lowers the maximum mortgage amount households can receive (Vereniging Eigen Huis, 2019). In addition, Lewis and Van Venrooij (1995) emphasize the ignorance and underestimation of students in overseeing the consequences of having a study loan debt and the

repayment of their debt.

There are mixed findings about the impact of a study loan debt on homeownership. According to Mezza et al (2016), there is a negative relation between student debt and the likelihood of homeownership among young American adults. For example, a 10% increase in student loan debt causes a 1 to 2 percentage points decrease in the homeownership rate for student loan borrowers. In contrast, the research of Letkiewicz and Heckman (2018) does not show an impact of having a study loan debt on homeownership among young Americans.

(4)

4 Does having a study loan debt influence homeownership among young Dutch adults?

To investigate the effect of having a study loan debt on the probability of homeownership among young Dutch adults, we use data from the De Nederlandse Bank Household Survey (DHS). An advantage of using data from the DHS is the availability of detailed data collected from Dutch households. Since the number of students having study loan debt and the amount of student debt increases every year, it is interesting to focus on the most recent year. We use data from the most recent survey held, which is the year 2018.

Performing probit and logit regressions and controlling for household socio-demographic and economic characteristics. This paper finds that having study loan debt decreases the likelihood of homeownership among young adults in the Netherlands. Precisely, young households with a study loan debt are approximately 22% less likely to own a home.

This paper is relevant for policymakers because understanding the consequences of study loan debt on homeownership could be important for the housing market. The finding of this paper shows the impact of having a study loan debt on the likelihood of homeownership among young Dutch adults.

This paper continues as follows. Section 2 presents an overview of the relevant literature on study loan debt and homeownership. Section 3 presents the methodology. In section 4, we describe the dataset used in this paper. Section 5 shows the results. Finally, section 6 presents the conclusion.

2. Literature review

This section provides a literature review of the most relevant and important studies for this research.

2.1 Why education matters

(5)

5 students who graduate from higher education earn on average between one and a half to twice as much as their peers without a higher education degree (Rijksoverheid, 2014). Another benefit of education is the positive relation between education and homeownership (Segal and Sullivan, 1998).

2.2 Study loan debt

Life-cycle hypothesis and human capital theory explain why individuals choose to take a study loan. From the point of view of life cycle theories of Ando and Modigliani (1963), students belong to a temporary low income group, having high necessary expenditures resulting in a high probability to be in debt since they are at the beginning of their career (Lea et al, 1993). Throughout their life cycle, students borrow as long as the expected future earnings exceed their study loan debt. The human capital theory suggests that a person makes investments in human capital if the potential benefits exceed the costs associated with education (Schultz, 1960).

The assumption that study loan debt discourages young adults to buy a home is largely based on two trends. Namely, the decreasing mortgage applications among young adults and the rising study loan debt (Houle and Berger, 2015). Study loan debt was the only kind of consumer debt that increased during the Great Recession, and where other types of debt can be discharged in the case of bankruptcy, this is not possible for study loan debt (Atkinson, 2010).

After finishing secondary school people have to make an intertemporal choice. This choice is between working or studying. The ability to work results immediately in benefits and delayed costs. For example, benefits are a monthly wage and not creating a study loan debt. An example of a delayed cost is a possible lower future wage. The other option, studying comes together with immediate costs and delayed benefits, and the possibility of creating a study loan debt. The human capital theory is in line with the intertemporal choice of going to college when the delayed benefits exceed the immediate costs. Intertemporal choices involve trade-offs between costs and benefits occurring at different times (Frederick et al, 2002). Many students enter their student life relatively uninformed about the consequences of a student loan debt and may take large amounts of debt without thinking of the impact the debt has on future mortgage applications and the liquidity constraint due to the repayment of their study loan debt (Bowen et al, 2009).

(6)

6 attitude behavior of students against debt, at the beginning students are not in favor of debt, but once they incur a study loan debt, the attitude towards debt changes, resulting in both an increase in the amount of money borrowed per year and an increase in the number of students who use this opportunity.

There are different motives for receiving a study loan. An example is the motivation of students to invest in their education when receiving a student loan (Avery and Turner, 2012). According to Oosterbeek and Van den Broek (2009), students with a study loan debt are less likely to have a part-time job resulting from the finding that students with a study loan debt spend 36 minutes per week more on their study compared to others.

2.3 Homeownership and mortgage

Some researchers stress the advantages of homeownership for households. DiPasquale and Gleaser (1999) document that due to the willingness of homeowners to invest and improve their community, homeownership has a positive impact on the social activities of individuals. For example, DiPasquale and Gleaser (1999) find that American homeowners are 6% more likely to work together to solve local problems. In addition, homeownership also shows a positive impact on the level of happiness of households. According to Hu (2013), who studied the impact of homeownership on overall happiness in China, homeownership has a positive effect on both overall happiness and one’s housing satisfaction. Compared with renters, homeowners exhibit higher levels of life satisfaction because homeowners feel much safer, might enjoy higher social status and dignity (Hu, 2013). In addition, to the social benefits of homeownership, there is also a financial perspective that makes homeownership interesting. Lersch and Dewilde (2018) find that homeowners in Germany and the UK are financially wealthier and have higher savings than tenants. Compared with tenants, homeowners benefit from the accumulation of housing wealth. Furthermore, mortgaged homeownership requires a stable income and job guarantee (Lersch and Dewilde, 2018).

Despite the advantages of homeownership, there are also disadvantages. Coco (2005) finds that households with a large part of their wealth invested in their home are often badly diversified. Due to the investment in homeownership, younger investors have limited money available to invest in bonds and stocks, which decreases the advantages of the debt and equity market participation (Coco, 2005). Another disadvantage of homeownership is the lower mobility to move away. According to Munch et al (2008), due to the costs associated with selling and buying a home, homeowners are less flexible than tenants to move away. For this reason, homeownership provides lower labor mobility (Munch et al, 2008).

(7)

7 young adults. Houle and Berger (2015) suggest structural changes in the economy or the delayed transition to adulthood can explain the decreasing homeownership rates. For example, the increase in the number of students might explain a delay of homeownership among young adults. In addition, both marriage delays and investments in human capital activities decline the rate of homeownership among young adults (Drew, 2015). According to Drew (2015), young adults have a greater likelihood to be single, and therefore defer marriage, compared with previous generations. Furthermore, young adults choose to invest in human capital activities, such as career development, and for this reason, they delay marriage, and therefore homeownership (Drew, 2015). Another important consideration of whether to own a home or not is the risky nature of homeownership. Letkiewicz and Heckman (2018) suggest that young adults can be influenced in making their decision resulting from the housing crash in 2007, in case they think homeownership is not a stable investment as it was before the housing crash.

2.4 Study loan debt and homeownership

The literature on the relation between study loan debt and homeownership is divided into two mainstream views. On the one hand, some authors support the view that study loan debt and homeownership are negatively related. The main reason is that homebuyers need to borrow money to buy their first home (Mezza et al, 2016). Having a study loan debt has an impact on the households’ ability to obtain the desired amount of mortgage needed to buy a home. The repayments of the study loan debt have influence on households’ liquidity constraints. Liquidity constraints have influence on the amount of mortgage households can receive to buy a house. According to Hiltonsmith (2013), a dual-headed household with a student debt of $53,000 has on average a lifetime wealth loss of nearly $208,000, in which two-thirds of the loss belongs to lower retirement savings of the indebted household, and one-third comes from a lower home equity. The impact of the lifetime assets of households with a study loan debt is nearly four times the amount borrowed.

(8)

8 Other authors suggest that student debt delays the age at which households buy their first home. Houle and Berger (2015) find that study loan debt is related to a delay in buying a home compared to those without a study loan debt. Stone et al (2012) find that 40% of the American students graduating with a study loan debt delay the purchase of a home. An explanation of the home purchase delay is given in the research of Brown and Caldwell (2013), suggesting that students with a study loan debt score 24 credit points less than students without a study loan debt. Having lower credit scores makes it more difficult to gain access to mortgage markets. The rising study loan debt discourages millennials to avoid homeownership because they prefer no additional debt, or they are unable to receive a mortgage due to their poor credit scores and high study loan debt (Houle and Berger, 2015).

On the other hand, some authors do not find a negative effect of study loan debt on homeownership. According to Letkiewicz and Heckman (2018), study loan debt does not have an impact on homeownership once all control variables are included. While Houle and Berger (2015) find limited evidence of the negative effect of study loan debt on homeownership, the impact is very modest in size. Houle and Berger (2015) even suggest that the decline in homeownership among young American adults cannot be blamed on having study loan debt. They conclude that any correlation between homeownership and student loan debt derives from individual life-cycle transitions and changes in the economic environment. The Great Recession is an example of a change in the economic environment. According to Brown and Caldwell (2013), the Great Recession negatively affected the homeownership rates for all young American households, but especially for young American households with a study loan debt. In addition, Houle and Berger (2015) do not find a connection between the amount of student loan debt among young student loan debtors and homeownership.

2.5 Study loan debt and homeownership in the Netherlands

(9)

9 On the one hand, study loan debt reduces the maximum mortgage amount households can receive. On the other hand, study loan debt decreases the possibility to save money. In both cases, a lower amount of money is available due to student debt repayments (Bowen et al, 2009). Because of the student debt repayments, households are faced with liquidity constraints (Bowen et al, 2009).

The assumption from the existing literature that study loan debt might influence homeownership leads to the following hypothesis:

H1: Study loan debt has a negative effect on homeownership.

This study contributes to the existing literature by examining the effect of study loan debt on homeownership among young adults in the Netherlands. Studies such as Letkiewicz and Heckman (2018), Mezza et al (2016), and Houle and Berger (2015) focus on the impact of student debt on homeownership among American households. This study pays particular attention to the impact of study loan debt on homeownership among young Dutch adults.

3. Methodology

This study aims to discover whether study loan debt influences homeownership among young adults in the Netherlands. This section contains the methodology applied in this paper.

3.1 Model specification

While investigating the research question, it is important to use different research methods. In this way, it can be implied whether study loan debt is a strong predictor of homeownership. The study of Letkiewicz and Heckman (2018) is closely related to our work. They studied the impact of study loan debt on homeownership among young Americans using logistic regression models.

(10)

10 have to keep in mind that homeownership is binary. For this reason, the probit model will be used in this paper to measure the influence of study loan debt on homeownership. The probit model overcomes the limitation of the linear probability model, whereby it produces estimated probabilities that are negative or greater than one (Dey and Astin, 1993). The probit coefficients and standard errors are estimated by the maximum likelihood technique.

The results of a probit model only indicate whether a variable has a positive or negative effect on the dependent variable. To interpret the coefficient estimates more appropriately the marginal effects are estimated in order to determine the magnitude. The error term may suffer from heteroscedasticity. For this reason, heteroskedastic- robust standard errors are used in this study. Furthermore, the McFadden’s Pseudo R² indicates the explanatory power of the probit models. The McFadden’s Pseudo R² should be interpreted with great caution because the statistic differs from the R-square means in OLS regression. In addition, many researchers have discussed about the non-existent widely accepted Pseudo-R² for binary models and find it disadvantageous (Veall and Zimmermann, 1994). Despite, there is no widely accepted Pseudo R², the use of binary models becomes more popular (Veall and Zimmermann, 1994). When the McFadden’s Pseudo R² assumes a value of between 0.2 and 0.4, there is a very good fit of the model (Louviere et al, 2000).

Since we use the study of Letkiewicz and Heckman (2018) as an example, the logit model will be used as a robustness check.

The focus of this study is to estimate the effect of study loan debt on homeownership. Therefore, an equation is formulated for the binary dependent variable homeownership. Homeownership equals one if a household is a homeowner and zero if the household is not the owner of a home (for example a household is renting a home). A probit model of being a homeowner is estimated by maximum likelihood and the equation to measure the impact is as follows:

𝐻𝑖 = 𝛼 + 𝛽1∗ 𝑆𝐿𝐷𝑖 + 𝛽2∗ 𝐶𝑉𝑖 + 𝜀𝑖 (1)

Where 𝐻𝑖 is our dependent variable. It is a binary dependent variable that takes the value of

(11)

11 𝑆𝐿𝐷_𝑖 represents a dummy variable for having study loan debt, where one means the household has a study loan debt and zero if a household does not have a study loan debt. It is our main explanatory variable. Study loan debt is defined as money owed on a loan that was taken out to meet the financial needs of students (Van Veluwen, 2018). It equals one if a household has a study loan debt and zero otherwise. 𝐶𝑉_𝑖 represents a set of socio-demographic control variables for household i, including age, gender, marital status, education level, household size, and gross income. Finally, 𝜀_𝑖 is the error or disturbance for household i.

3.2 Control variables

Several control variables are used in the regression in order to take into account differences in households’ socio-demographic and economic characteristics. In the research of Segal and Sullivan (1998), differences in homeownership rates are observed, the variables they used in their research are race, age, gender, marital status, education level, household size, and income. The association between homeownership and study loan debt is analyzed, using the data of DHS. The DHS does not provide information about the race of the respondents or a possible proxy variable, so the variable race will not be used in this paper. The other variables are available and are used in this paper.

Age is a life cycle component to homeownership. As the age of the respondent increases, then the rate of homeownership tends to increase as well (Segal and Sullivan, 1998). After the financial crisis, the average age of households that buy a home for the first time increased to the age of 32. Compared to the average age before the financial crisis, the average age of first-time homebuyers is nowadays higher in the Netherlands (Vereniging Eigen Huis, 2016). In this study, age is expected to show a positive influence on homeownership.

(12)

12 Data from the census (U.S. Bureau of the Census, 1995) suggests the existence of inequalities. In total, 65% of the households in the U.S. own a home, but there are differences in the rate of homeownership among the marital status of a household. The percentages of the rate of homeownership are as follows: 79% of married couples hold a home, 55% of the single man, and 45% of households headed by women without husbands. Houle and Berger (2015) find that marital status is positively associated with increased homeownership rates. The expectation of the influence of marital status on homeownership is in line with the finding of Houle and Berger (2015).

With the intertemporal choice, individuals have to make between working immediately or studying after finishing secondary school, at which the option of studying increases the level of education. The choice of an individual to invest in education has several reasons. For example, higher education levels result in a potentially higher income and beneficial job prospects (Brown et al, 2014). Beneficial job prospects and potentially higher income are not the only benefits of education. According to Segal and Sullivan (1998), there is a positive relation between education and homeownership. The expectation of education level on homeownership is in line with the finding of Segal and Sullivan (1998).

Another household characteristic is the size of a household. According to Letkiewicz and Heckman (2018), children have a positive impact on homeownership. They find that households with children are significantly more likely to own a home than households without children. This finding is in line with the expectation of this paper.

Gross income is an economic characteristic that can influence whether households own a home or not. According to Segal and Sullivan (1998), the level of homeownership is 1.2 percentage points lower due to the increase in income inequalities. The rate of homeownership would be higher for the period between 1977 and 1995 if the real income growth with the same level as the economy. According to Gabriel and Rosenthal (2015), homeownership is more common for families with higher incomes. The reason for including gross income as an additional economic control variable is that an increase in the households’ gross income provides them more opportunities to receive a higher mortgage and to pay the amounts above the LTV. Higher-income is expected to increase the probability of homeownership.

4. Data

(13)

13

4.1 Sample construction

In this paper, we use data from the DNB Household Survey (DHS), which is part of the CentERpanel. The DHS consists of 2000 households in the Netherlands. The participants in this paper are randomly selected from the population, every year the participants answer questions asked by the DHS. The longitudinal survey is administered through the internet. To prevent selection bias, households without an internet connection are provided with a connection or with a set-top box for their television. The data contains information about employment, income, dwell, mortgages, pensions, health, assets, debts, and economic and psychological concepts and personal characteristics. The survey collects information annually, including individuals with an age of a least 16 years old. The questions of the survey are only asked once to every participant in the year 2018. Since the amount of student debt and the number of students having study loan debt increases every year, it is interesting to focus on the most recent year. Therefore, we take the most recent cross-sectional dataset, which is the year 2018. To get the variables that are needed for this research several datasets are merged. In this study we use information from the following datasets: Assets and Liabilities, Accommodation and Mortgages, General Information on the Household, Aggregated Data on Income, and Household and Work. The combined datasets contain 4,591 observations of respondents in the original sample.

A household is allowed to buy a home when the age of 18 years is reached in the Netherlands because from the age of 18, you are an adult. The data of DHS also consist of respondents who are underage, so younger than 18 years old. We eliminated the underage individuals because they are not able to own a home. This resulted in the elimination of a total of 736 responses. The elimination of 736 individuals seems to be a large elimination for the individuals with an age of 16 and 17, but the dataset General Information on the Household uses distracted information of all the members of the household, so also the minors. This means that the questions were not submitted to the respondent, but were derived from other variables.

In the merged dataset, children who are living at their parents’ home are recognized as a homeowner. Keeping the children living at their parents’ home in the research would give a biased estimation of the coefficients since the parents are the homeowners or tenants and not the children. The elimination of the children living at their parents’ home consists of 321 responses.

(14)

14 40 years old (Fry, 2014). Respondents with an age of 40 years and older are deleted from the sample because the focus in this research is on young adults. The age of the respondent is calculated by: 2018 less year of birth respondent. The elimination of the household members who are too old consists of 2,678 observations.

Given that the research question is whether study loan debt affects homeownership. The respondents who were not eligible to attend the required education level at which you can receive a study loan are removed from the sample. If the highest attended education level of an individual is special education, kindergarten/ primary education, other sort of education, or no education the individual was unable to qualify for receiving a student loan. Keeping the four mentioned education levels in the dataset would add noise in the analysis and so on 22 observations are eliminated.

Many observations were deleted from the original sample and the dataset is left with 834 observations.

4.2 Descriptive statistics

The control variable age is categorized using four categories. The categorization of the respondents’ age is as follows: 1: between 18 and 23, 2: between 24 and 29, 3: between 30 and 35 and the last category 4: between 36 and 39. Gender is a dummy variable and equals one if the participant is a male and zero if the participant is a female. Thirdly, the marital status of the respondents is categorized as follows: 1: married, 2: living with their partner, and 3: single. Dummies are created for the different categories. Furthermore, the categorization of the respondents’ education level is as follows: 1: low education, which exists of pre-vocational education, pre-university education and senior pre-vocational training, 2: middle education, which includes vocational colleges and the last category 3: high education, which is university education. Fifthly, household size is measured as: 0 if there are no children in the household and 1 if there is at least one child in the household. Finally, the categorization of the economic control variable gross income is as follows: 1: less than €20,000, 2: between €20,000 and <€40,000, 3: between €40,000 and <€60,000 and 4: at least €60,000.

(15)

15 young adults could also be a reason why there is a relatively low percentage of households with at least one child compared with the 64% households without children. Assuming the probability of being a parent is higher when you have an age between 36-39 than if you have an age between 18-23, see Table A1 in Appendix A for the distribution of household size and age of the household. Furthermore, 38% of the respondents attended low education, 37% attended middle education and 26% attended high education. The average age of the respondents is 32. Finally, the average gross income of the respondents is €25,961. Gross income is divided by €1,000 due to the large scale of the variable.

Table 1: Descriptive statistics

This table reports the descriptive statistics for the variables used in this paper. The data are from the 2018 DNB Household Survey. Where N is the number of observations.

Mean St.Dev min max range N

Dependent variable

Homeowner 0.64 0.48 0 1 1 834

Independent variable

Study loan debt 0.29 0.45 0 1 1 447

Control variables Age Age 32 4.98 18 39 21 834 Gender Male 0.45 0.50 0 1 1 834 Marital status Married 0.34 0.47 0 1 1 514

Living with partner 0.26 0.44 0 1 1 514

Single 0.40 0.49 0 1 1 514 Education level Low education 0.38 0.48 0 1 1 834 Middle education 0.37 0.48 0 1 1 834 High education 0.26 0.44 0 1 1 834 Household size At least 1 child 0.36 0.48 0 1 1 834 Gross income Gross income/€1,000 25.96 27.94 0 361.39 361.39 435

(16)

16 study loan debt status of homeowners of approximately 36%. Around 29% of the households have a study loan debt, at which the majority 19.46% are no homeowners. A quarter of the households are neither homeowners nor study loan debt holders.

Table 2: Cross-tabulation study loan debt by homeownership

This table presents the cross-tabulation of study loan debt by homeownership. The values are in percentages. The number of observations is 447. The data are from the 2018 DNB Household Survey. Study loan debt

Homeowner

No Yes Total

No 25.50 45.64 71.14

Yes 19.46 9.40 28.86

Total N=447 44.97 55.03 100.00

Table 3 provides an overview of the household characteristics by study loan debt and homeownership using cross-tabulations. The table is divided into a group of households that own a home and a group of households that do not own a home. Furthermore, a distinction has

been made between the study loan debt status. The first column represents the households that are no homeowners and do not have study loan debt. The second column reports the households that do not own a home but do have a study loan debt. The third column exists of households that are homeowners without study loan debt. The last column represents homeowners with study loan debt.

From Table 3, the first signs of the influence of study loan debt on homeownership become clear. As shown in the third and fourth columns, every variable presents a higher percentage of homeowners without study loan debt compared to the homeowners with study loan debt. For example, there is a difference in the group of homeowners between the ages of 30 and 35, of at least three times, 19.24% do not have study loan debt and 5.59% have study loan debt. Furthermore, 21.48% of the respondents are male who owns a home without having a study loan debt. This is quite similar to the 24.16% of respondents who are female owning a home without having study loan debt. If we compare the females who are homeowners, 24.16% do not have a study loan debt, against 4.70% with a study loan debt.

(17)

17 who own a home. Comparing the marital status of homeowners with study loan debt, we see that the single households represent 1.45% of the respondents, the percentage for the households that are living with a partner or that are married is more than twice as large as the rate of the single households. A possible explanation is that married households or households living with a partner can have dual earners. In contrast, a single household is a sole earner. Normally dual earners receive a higher mortgage amount than sole earners, which influences the homeownership rate. Study loan debt becomes more common when the level of education rises, 2.91% of the respondents attended low education and have study loan debt. Followed by 11.85% of respondents with study loan debt and middle education.

(18)

18 Table 3: Cross-tabulation of the control variables by study loan debt and homeownership This table presents the cross-tabulation of the control variables by study loan debt and homeownership. The values are in percentages. Where N is the number of observations. The number of observations can differ between the variables due to missing information. The data are from the 2018 DNB Household Survey.

Homeowner

No Yes

Study loan debt

No Yes No Yes Age N=447 Between 18 and 23 2.24 7.38 0.22 0.00 Between 24 and 29 7.38 7.83 9.17 2.24 Between 30 and 35 8.72 3.80 19.24 5.59 Between 36 and 39 7.16 0.45 17.00 1.57 Gender N=447 Female 16.33 10.07 24.16 4.70 Male 9.17 9.40 21.48 4.70 Marital status N=413 Married 4.12 1.21 23.00 3.63

Living with partner 4.12 3.39 13.08 3.39

Single 16.71 15.98 9.93 1.45 Education level N=447 Low education 13.42 2.24 20.13 0.67 Middle education 8.05 7.38 18.79 4.47 High education 4.03 9.84 6.71 4.25 Household size N=447 No children 21.48 19.02 25.50 6.71 At least 1 child 4.03 0.45 20.13 2.68 Gross income N=359 <€20,000 13.37 16.43 11.42 2.79 €20,000 and <€40,000 8.64 3.34 19.22 2.79 €40,000 and <€60,000 1.11 0.84 12.53 3.06 >=€60,000 0.84 0.00 2.79 0.84

(19)

19 Dwyer et al (2013), men are more likely to drop out of college at a lower level of study loan debt than women. In the correlation matrix of the independent variables, none of the values show a (very) strong correlation with each other. The highest value in the correlation matrix is between household size and marital status and is around -0.518, which is a moderate correlation (Taylor, 1990). Therefore, there is no serious sign of multicollinearity. The correlation matrix of the independent variables is shown in Table B1 of Appendix B.

The correlation matrix of the dependent variable homeowner and independent variable study loan debt is presented in Table B2 of Appendix B. The correlation between homeowner and study loan debt is negative, and precisely -0.288. This is in line with the expectation of our hypothesis.

To examine whether the negative effect of study loan debt on homeownership is significant multiple probit regressions are performed. In the next chapter, more relevant information about the influence of study loan debt on homeownership is available.

5. Results

In this chapter, we provide the impact of study loan debt on homeownership. The results are presented from the probit model. The logit model serves as a robustness check.

5.1 Regression analysis

Multiple probit regressions are performed to measure the influence of study loan debt on homeownership. Coefficient estimates are shown in Table C1, see Appendix C. These coefficients indicate whether a variable increase or decrease the likelihood of households to own a home. They do not show the magnitude the probabilities of a variable have on the dependent variable homeownership. For example, in column 1 of Table C1, we cannot conclude that households with a study loan debt have 81.5% lower probability to own a home. For a better understanding of the results of the probit model, marginal effects are computed.

(20)

20 with a study loan debt are around 32% less likely to own a home. The aim is to determine whether study loan debt independently influences the probability of homeownership among households. The value of the McFadden’s Pseudo R² in column 1 is below 0.2, so it is not a good fit of the model. Adding control variables increases the explanatory power of the model.

To overcome endogeneity problems control variables are added to the regression. The control variables also serve as a purpose to obtain more reliable estimates of the primary independent variable. In column 2 the socio-demographic control variables are included in the regression, and in column 3 the economic control variable gross income is added to the model. To prevent perfect multicollinearity reference groups are used for every variable in Table 4. For example, low education is omitted from the regression.

Since the inclusion of the socio-demographic control variables in column 2, the impact of the main independent variable study loan debt on homeownership weakens. However, study loan debt is still significant at a significance level of 1%. Households who have study loan debt are approximately 26% less likely to own a home.

Column 3 includes the economic control variable, so all variables in this study are reported. Just like columns 1 and 2, study loan debt still shows a negative and significant effect on homeownership in column 3. The stars behind the coefficients measure the significance level, at which more stars behind a coefficient provide a more reliable interpretation of the results. In the case of the influence of study loan debt on homeownership, the coefficients are reliable.

Adding more control variables to the regression reduces the number of observations because the model only includes observations for which the variables are accessible. Including all the variables, column 3 shows that households with a study loan debt are 22% less likely to own a home than households without study loan debt.

(21)

21 Even though the gender coefficients in column 2 and column 3 are not significant, the coefficients deserve attention. The inclusion of gross income in column 3 results in a change of the sign of likelihood for males to own a home. Where males are more likely to own a home than the reference group females in column 2. The opposite is true in column 3, the negative coefficient means that males are less likely to own a home than females. The gender coefficients must be interpreted with caution because the coefficients are insignificant. The insignificant effect of the gender coefficients on the likelihood of homeownership is in line with the finding of Allen (2002).

Married households and households that are living with a partner show a strong positive effect on homeownership. To be more specific married households and households that are living with a partner are around 41% and 33% more likely to own a home than the reference group single households. The positive influence of marital status on the likelihood of homeownership is as expected and in line with the finding of Houle and Berger (2015).

The education levels of middle education and high education show an insignificant effect on homeownership. This insignificant effect is not as expected and not in line with the finding of Segal and Sullivan (1998). A possible explanation for this insignificant effect is the relatively high percentage of respondents with a study loan debt representing the education groups middle and high education compared to the respondents with low education, see Table 3. As shown in Table 4, households with a study loan debt are less likely to own a home. This could explain the insignificant effect of education level on homeownership, especially since this study focuses on young adults.

Having at least one child in the household seems to matter since the presence of at least one child increases the likelihood of owning a home. Column 3 of Table 4 shows that households with at least one child are approximately 22% more likely to own a home than households without children. Probably, because the average age of households buying a home for the first time is 32 (Vereniging Eigen Huis, 2016). And a household within the 36 and 39 age group is more likely to have at least one child than a household with an age between 18 and 23, see the cross-tabulation in Table A1 of Appendix A. Furthermore, households without children serve as a reference group. The influence of at least one child in a household on the probability of homeownership is as expected and is in line with the finding of Letkiewicz and Heckman (2018).

(22)

22 of gross income on homeownership is not as expected, and in particular due to the highest gross income group. This study suggests an insignificant effect of the highest gross income group on homeownership. A possible explanation for the insignificant effect of gross income >=€60,000 on homeownership is the relatively low percentage of respondents with a gross income of at least €60,000, which could lead to conflicting results. As shown in Table 3, approximately 4.5% of the households represent the highest income group. All in all, the influence of a higher gross income on the probability of homeownership in this paper is not in line with the finding of Gabriel and Rosenthal (2015).

Finally, the McFadden’s Pseudo R² assumes a very good fit of the model in column 2 and column 3 of Table 4, equal to 0.3143 and 0.3972.

Providing a more reliable estimate of the impact of study loan debt on homeownership. A robustness check is performed using the logit model and the marginal effects over the logit regression, see Table D1 and Table D2 in Appendix D. The main independent variable study loan debt remains significant at any significance level. Meaning that the negative impact of study loan debt on homeownership is consistent with the results from Table 4 and Table C1.

(23)

23 Table 4: Marginal effects probit model

This table reports the marginal effects using a probit model. The data are from the 2018 DNB Household Survey. The robust standard errors are reported in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Marginal effects

(1) (2) (3)

Study loan debt -0.316*** -0.257*** -0.220***

(0.049) (0.071) (0.082)

Age (reference group: between 18 and 23) Between 24 and 29 0.463*** 0.349** (0.132) (0.147) Between 30 and 35 0.583*** 0.373** (0.128) (0.165) Between 36 and 39 0.550*** 0.400*** (0.115) (0.145) Gender (reference group: female)

Male 0.093 -0.015

(0.058) (0.071) Marital status (reference group:

single)

Living with partner 0.350*** 0.329***

(0.058) (0.066)

Married 0.425*** 0.411***

(0.060) (0.065) Education level (reference group:

low education)

Middle education 0.076 0.023

(0.071) (0.081)

High education 0.119 0.029

(0.078) (0.093) Household size (reference group: no

children)

At least 1 child 0.142** 0.219***

(0.072) (0.082) Gross income (reference group:

<€20,000)

€20,000 and <€40,000 0.219***

(24)

24 €40,000 and <€60,000 0.454*** (0.060) >=€60,000 0.233 (0.139) Observations 447 413 346 McFadden’s Pseudo R² 0.0607 0.3143 0.3972 Log likelihood -288.91 -195.15 -143.31

6. Conclusion and discussion

In this study, we analyzed whether study loan debt influences homeownership among young Dutch adults. The review of available literature such as Letkiewicz and Heckman (2018), Mezza et al (2016), and Houle and Berger (2015) showed mixed findings about the impact of study loan debt on homeownership among American households. The contribution of this paper to the existing literature is by studying the effect of study loan debt on homeownership among young Dutch adults. Since our dependent variable homeownership is a binary variable, this paper uses a probit model to regress the variables. To measure the influence of study loan debt on homeownership different household characteristics were added to the probit model. The addition of several household characteristics also serves for a more reliable estimation of the coefficients. For a better interpretation of the coefficients, marginal effects were estimated to determine the magnitude of the coefficients.

Our main prediction is confirmed by the data: the influence of study loan debt on homeownership is strongly significant in all the regressions. Precisely, study loan debt has a negative effect on homeownership. And so on, as expected, this paper finds a significant negative effect of study loan debt on homeownership after controlling for age, gender, marital status, education level, household size, and gross income. This means that households with a study loan debt are 22% less likely to own a home. This finding is in line with studies such as Mezza et al (2016), and Cooper and Wang (2014), they also find a negative impact of student debt on homeownership.

(25)

25 amounts of debt without thinking of the impact the debt has on future mortgage applications and the repayment of their study loan debt.

Other variables also show a significant effect on the likelihood of homeownership. For example, the positive and significant effect of married households on homeownership. Married households are 41.1% more likely to own a home than single households, which is a larger impact on homeownership than the influence of study loan debt on homeownership. For future research, it would be interesting to examine the influence of marital status on homeownership, and especially to focus on the likelihood of single households on homeownership. Furthermore, most respondents in this study studied before the abolition of the basic grant, so it would be interesting to compare several years before and after the abolition of the basic grant. Especially if more data is available about households with a study loan debt who studied after the abolition. This is interesting for policymakers because it shows the impact of the abolition of the basic grant on homeownership.

This study is not without some limitations. To begin with, this paper uses data from the DHS. The DHS obtains data through survey questions to respondents. The quality of the data depends on the honesty of the respondents’ answers. The answers could differ from the actual behavior of the respondents. Therefore, the use of survey data in this paper can be considered as a limitation.

Secondly, the status of the respondents in the DHS database changes from having a study loan debt to no study loan debt, after the repayment of their study loan debt. For future research, it could be interesting to investigate the impact of study loan debt on homeownership if respondents were included who paid off their study loan debt. The inclusion of households who paid off their study loan debt can provide a more reliable estimate of the long-term impact of study loan debt on homeownership.

(26)

26 References

Allen, B.L., 2002. Race and gender inequality in homeownership: does place make a difference? Rural Sociology, 67(4), 603-621.

Ando, A., Modigliani, F., 1963. The life cycle hypothesis of saving: aggregate implications and tests. American Economic Review, 53(1), 55-84.

Atkinson, A., 2010. Race, educational loans, and bankruptcy. Michigan Journal of Race and Law, 16(1), 1-31.

Avery, C., Turner, S., 2012. Student loans: do college students borrow too much or not enough. Journal of Economic Perspectives, 26(1), 165-192.

Bowen, W.G., Chingos, M.M., McPherson, M.S., 2009. Crossing the Finish Line: Completing College at America’s Public Universities. Princeton University Press, Princeton.

Brown, M., Caldwell, S., 2013. Young student loan borrowers retreat from housing and auto markets. Federal Reserve Bank of New York. Accessed on the 26th of September 2019: https://libertystreeteconomics.newyorkfed.org/2013/04/young-student-loan-borrowers-retreat-from-housing-and-auto-markets.html

Brown, M., Haughwout, A., Lee, D., Scally, J., Van der Klaauw, W., 2014. Measuring student debt and its performance. Federal Reserve Bank of New York, 668. Accessed on the 15th of September 2019: https://www.newyorkfed.org/research/staff_reports/sr668.pdf

(27)

27 Centraal Bureau Statistiek., 2019a. Studenten lenen vaker en meer. Accessed on the 6th of November 2019: https://www.cbs.nl/nl-nl/nieuws/2019/41/studenten-lenen-vaker-en-meer

Centraal Plan Bureau., 2014. Gemiddelde aflossing en inkomenseffecten sociaal leenstelsel. Accessed on the 16th of September 2019: https://www.cpb.nl/sites/default/files/ publicaties/download/cpb-notitie-22mei2014-gemiddelde-aflossing-en-inkomenseffecten-sociaal-leenstelsel.pdf

Cocco, J.F., 2005. Portfolio choice in the presence of housing. Review of Financial Studies, 18(2), 535-567.

Cooper, D., Wang, J.C., 2014. Student loan debt and economic outcomes. Federal Reserve Bank of Boston Current Policy Perspectives, 14(7).

Davies, E., Lea, S.E.G., 1995. Student attitudes to student debt. Journal of Economic Psychology, 16, 663-679.

De Voogt, S., 2018. Starters zijn de dupe op de woningmarkt. NRC, July 2018. Accessed on the 10th of October 2019: https://www.nrc.nl/nieuws/2018/07/12/starters-zijn-de-dupe-op-de-woningmarkt-a1609753

Dey, E.L., Astin, A.W., 1993. Statistical alternatives for studying college student retention: a comparative analysis of logit, probit, and linear regression. Research in Higher Education, 34(5), 569-581.

DiPasquale, D., Gleaser, E.L., 1999. Incentives and social capital: are homeowners better citizens? Journal of Urban Economics, 45, 354-384.

Drew, R.B., 2015. Effect of changing demographics on young adult homeownership rates. Joint Center for Housing Studies Harvard University. Accessed on 6th of November 2019: https://www.jchs.harvard.edu/sites/default/files/w15-2_drew.pdf

Dwyer, R.E., Hodson, R., McCloud, L., 2013. Gender, debt, and dropping out of college. Gender & Society. 27(1), 30-55.

Frederick, S., Loewenstein, G., O’Donoghue, T., 2002. Time discounting and time preference: a critical review. Journal of Economic Literature, 40(2), 351-401.

(28)

28 Gabriel, S.A., Rosenthal, S.S., 2015. The boom, the bust and the future of homeownership. Real Estate Economics, 43(2), 334-374.

Godfried, M., Oosterbeek, H., Van Tulder, F., 2001. Adverse selection and the demand for supplementary dental insurance. De Economist, 149(2), 177-190.

Hiltonsmith, R., 2013. At what cost? How student debt reduces lifetime wealth. Demos Debt for Diploma Series. Accessed on the 16th of September 2019: http://www.demos.org/sites/default/files/imce/AtWhatCostFinal.pdf

Houle, J.N., Berger, L., 2015. Is student loan debt discouraging homeownership among young adults. Social Service Review, 89(4), 589-621.

Hu, F., 2013. Homeownership and subjective wellbeing in urban China: does owning a house make you happier? Social Indicators Research, 110(3), 951-971.

Lea, S.E.G., Levine, R.M., Webley, P., 1993. The economic psychology of consumer debt. Journal of Economic Psychology, 14(1), 85-119.

Lersch, P.M., Dewilde, C., 2018. Homeownership, saving and financial wealth: a comparative and longitudinal analysis. Journal of Housing Studies, 33(8), 1175-1206.

Letkiewicz, J.C., Heckman, S.J., 2018. Homeownership among young Americans: a look at student loan debt and behavioral factors. Journal of Consumer Affairs, 52(1), 88-114.

Lewis, A., Van Venrooij, M., 1995. A note on the perceptions of loan duration and repayment. Journal of Economic Psychology, 16(1), 161-168.

Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods Analysis and Application. Cambridge University Press, Cambridge.

Mankiw, N.G., Romer, D., Weil, D.N., 1992. A contribution to the empirics of economic growth. Quarterly Journal of Economics, 107(2), 407-437.

(29)

29 Moretti, E., 2004. Estimating the social return to higher education: evidence from longitudinal and repeated cross-sectional data. Journal of Econometrics, 121(1), 175-212.

Munch, J.R., Rosholm, M., Svarer, M., 2008. Home ownership, job duration, and wages. Journal of Urban Economics, 63(1), 130-145.

National Association of Realtors., 2012. 2012 profile of home buyers and sellers. Accessed on the 15th of October 2019: https://www.realtor.org/topics/profile-of-home-buyers-and-sellers.

Oosterbeek, H, Van den Broek, A., 2009. An empirical analysis of borrowing behaviour of higher education students in the Netherlands. Economics of Education Review, 28, 170-177. Rijksoverheid., 2014. Studievoorschot maakt investering tot €1 miljard mogelijk. Accessed on the 24th of October 2019: https://www.rijksoverheid.nl/onderwerpen/studiefinanciering/ nieuws/2014/05/28/studievoorschot-maakt-investering-tot-1-miljard-mogelijk

Schonewille, G., Stoof, R., Van der Werf, M., 2017. Studentenonderzoek 2017. Nibud, September 2017: https://www.nibud.nl/wp-content/uploads/Nibud-Studentenonderzoek-2017.pdf

Schultz, T.W., 1960. Investment in human capital. American Economic Review, 51(1), 1-17.

Segal, L.M., Sullivan, D.G., 1998. Trends in homeownership: race, demographics, and income. Journal of Economic Perspectives, 22(2), 53-72.

Smit, V., 2018. Nationale studieschuld stijgt verder: nu 11,2 miljard euro. Accessed on the 17th of September 2019: https://nos.nl/artikel/2248123-nationale-studieschuld-stijgt-verder-nu-11-2-miljard-euro.html

Stone, C., Van Horn, C., Zukin, C., 2012. Chasing the American dream: recent college graduates and the Great Recession. New Brunswick, NJ: Center for Workforce Development. Assessed on the 26th of September 2019: https://www.heldrich.rutgers.edu/sites/default/files/ products/uploads/Chasing_American_Dream_Report.pdf

Taylor, R., 1990. Interpretation of the correlation coefficient: a basic review. Journal of Diagnostic Medical Sonography, 1, 35-39.

(30)

30 U.S. Bureau of the Census., 1995. Housing market news. Washington, DC: Government Printing Office. Accessed on the 26th of September 2019.

Van Veluwen, M., 2018. Hypotheek en studieschuld: de invloed van een studieschuld op de hoogte van een hypotheek. Interstedelijk Studenten Overleg (ISO), Augustus 2018. Accessed on the 16th of September 2019: https://www.iso.nl/wp-content/uploads/2018/08/Hypotheek-Studieschuld-ISO-Augustus-2018.pdf

Veall, M.R., Zimmermann, K.F., 1994. Evaluating pseudo-R²’s for binary probit models. Quality and Quantity, 28(2), 151-164.

Vereniging Eigen Huis., 2019. Regels hypotheekverstrekking. Accessed on the 16th of October 2019: https://www.eigenhuis.nl/hypotheken/regels-hypotheekverstrekking#/

(31)

31 Appendix

Appendix A. Distribution by age groups

Table A1: Distribution age of household and children

This table presents the distribution of the age of the respondents and the household size. The values are in absolute numbers. The data are from the 2018 DNB Household Survey.

Age of household Household size No children At least 1 child Total Between 18 and 23 57 0 57 Between 24 and 29 164 24 188 Between 30 and 35 208 130 338 Between 36 and 39 104 147 251 Total 533 301 834

Table A2 tabulates the respondents in the sample by age group for homeownership. Table A2 shows that 63.55% of the respondents own a home. Furthermore, the older the age group, the higher the homeownership rate among young adults. For example, 5.26% of the respondents with an age between 18 and 23 years old are homeowners. Compared to the oldest respondents who are between 36 and 39 years old, 82,07% own a home.

Table A2: Homeownership rates by age groups

The values are in percentages. The data are from the 2018 DNB Household Survey. Age of household

Household own home

(32)

32 The primary independent variable in this paper is study loan debt. Study loan debt is a dummy variable based on the question whether the respondent has a study loan debt or not on 31 December 2017. The dummy variable equals one if a household has a study loan debt and zero otherwise. To receive a student loan, the respondent has attended at least low education. Table A3 tabulates the respondents in the sample by age group for study loan debt. Table A3 shows that around 29% of the respondents have a study loan debt. The older the age group, the lower the proportion of respondents with a study loan debt. For example, 75% of the respondents with an age between 18 and 23 years old have a study loan debt. Compared to the oldest respondents 7.69% have a study loan debt. Probably, because older respondents have a higher wage and they have had more time to repay their study loan debt.

Table A3: Study loan debt by age groups

The values are in percentages. The data are from the 2018 DNB Household Survey. Age of household

Study loan debt

No Yes Between 18 and 23 25.00 75.00 Between 24 and 29 62.18 37.82 Between 30 and 35 74.85 25.15 Between 36 and 39 92.31 7.69 All ages 71.14 28.86

Appendix B. Correlation matrix

Table B1: Correlation matrix independent variable and control variables

This table presents the correlation matrix of the independent variable and control variables. The data are from the 2018 DNB Household Survey.

Variables Study loan debt Gender Marital status Household size Education level Age Gross income Study loan debt 1.000

Gender 0.057 1.000 Marital status 0.216 -0.051 1.000 Household size -0.227 0.009 -0.518 1.000 Education level 0.376 0.003 0.070 -0.009 1.000 Age -0.388 0.076 -0.378 0.410 -0.182 1.000 Gross income -0.218 0.237 -0.290 0.292 -0.015 0.432 1.000

Table B2: Correlation matrix homeowner and study loan debt

This table presents the correlation between homeownership and study loan debt. The data are from the 2018 DNB Household Survey.

Variables Homeowner Study loan debt Homeowner 1.000

(33)

33 Appendix C. Probit regression

Table C1: Probit regression

This table reports the results using a probit model. The dependent variable homeownership is a dummy and equals one if the household owns a home. The data are from the 2018 DNB Household Survey. The robust standard errors are reported in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Dependent variable: homeownership

(1) (2) (3)

Study loan debt -0.815*** -0.657*** -0.560***

(0.135) (0.188) (0.212)

Age (reference group: between 18 and 23) Between 24 and 29 1.337*** 0.994** (0.472) (0.481) Between 30 and 35 1.685*** 1.020** (0.473) (0.492) Between 36 and 39 1.705*** 1.178** (0.494) (0.517) Gender (reference group: female)

Male 0.234 -0.039

single)

(0.181) (0.216)

Married 1.167*** 1.168***

low education)

(0.180) (0.208)

(0.201) (0.238) Household size (reference group:

no children)

At least 1 child 0.363* 0.585**

(34)

34 Gross income (reference group:

<€20,000) €20,000 and <€40,000 0.578*** (0.185) €40,000 and <€60,000 1.516*** (0.340) >=€60,000 0.635 (0.417) Constant 0.362*** -2.107*** -1.850*** (0.072) (0.491) (0.496) Observations 447 413 346 McFadden’s Pseudo R² 0.0607 0.3143 0.3972 Log likelihood -288.91 -195.15 -143.31

Appendix D. Robustness check

Table D1: Logit regression

This table reports the results using a logit model. The dependent variable homeownership is a dummy and equals one if the household owns a home. The data are from the 2018 DNB Household Survey. The robust standard errors are reported in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Dependent variable: homeownership

(1) (2) (3)

Study loan debt -1.310*** -1.082*** -0.997***

(0.222) (0.325) (0.384)

Age (reference group: between 18 and 23) Between 24 and 29 2.651** 1.945* (1.079) (1.097) Between 30 and 35 3.249*** 1.982* (1.079) (1.106) Between 36 and 39 3.235*** 2.278** (1.111) (1.145) Gender (reference group: female)

Male 0.419 -0.077

single)

(0.309) (0.390)

Married 1.967*** 1.978***

low education)

(0.317) (0.375)

(0.346) (0.420) Household size (reference group:

no children)

At least 1 child 0.587* 0.967**

(35)

35 Gross income (reference group:

<€20,000) €20,000 and <€40,000 1.009*** (0.319) €40,000 and <€60,000 2.762*** (0.642) >=€60,000 1.081 (0.865) Constant 0.582*** -3.929*** -3.415*** (0.117) (1.111) (1.117) Observations 447 413 346 McFadden’s Pseudo R² 0.0607 0.3133 0.3983 Log likelihood -288.91 -195.45 -143.04

Table D2: Marginal effects logit model

This table reports the marginal effects using a logit model. The data are from the 2018 DNB Household Survey. The robust standard errors are reported in parentheses *** p<0.01, ** p<0.05, * p<0.1.

Marginal effects

(1) (2) (3)

Study loan debt -0.316*** -0.263*** -0.244***

(0.049) (0.075) (0.091)

Age (reference group: between 18 and 23) Between 24 and 29 0.530*** 0.400** (0.157) (0.185) Between 30 and 35 0.650*** 0.433** (0.149) (0.211) Between 36 and 39 0.599*** 0.447*** (0.136) (0.174) Gender (reference group: female)

Male 0.104 -0.019

single)

(0.060) (0.068)

Married 0.438*** 0.420***

low education)

(0.078) (0.091)

(0.083) (0.102) Household size (reference group: no

children)

At least 1 child 0.144* 0.223***

(36)