1
The Impact of Policy Reforms Concerning Student Financial Aid on Parental Subsidies
Thijs Willem Veelen1
Thesis MSc. Economics (EBM877A20) Supervisor: Dr. Carolina Laureti
05-06-2020 14262 words
Abstract
This paper studies the impact of a contractionary Dutch student financial aid reform in 2015 on the subsidy parents give to their studying children. To answer this question, data from the Dutch Household Survey of De Nederlandsche Bank (DHS) is used in a difference in difference approach. Richer parents are the treatment group and poorer parents are the control group. Interestingly, the results indicate that richer parents have increased their subsidy for students following the Dutch policy reform. On average, the richer parents increased their subsidy two times as much as poor parents did. Especially the parents who are male, younger than 50, or higher educated and belong to the treatment group tend to have increased their subsidy follow-ing the policy reform, compared with the control group of poorer parents.
Keywords: DNB Dutch Household Survey, panel data, student support, policy evaluation. JEL classification: D14, D19, I22, I28,
1 University of Groningen, Faculty of Economics and Business Student number: s2969793
2
I. INTRODUCTION
In the Netherlands, all students attending university or higher vocational education used to receive a monthly subsidy from the government. Students living with their parents received an amount of €108, and the students living on their own used to receive €300 monthly. In 2015 a policy reform was implemented which abolished this subsidy: only the students whose parents’ income is below a threshold continued to receive a monthly allowance. This Dutch policy reform caused that students did not receive a substantial part of their living costs anymore, thus those who still wanted to go to college had to seek for other sources of money. One of these sources is the introduced ‘social loan system’ (sociaal leenstelsel in Dutch), which facilitated beneficial borrowing conditions for the duped students (Rutte & Samsom, 2012). Another essential source of money for Dutch students is a subsidy given from parents to their studying children (henceforth: parental subsidy).
This study investigates whether parental subsidy has increased in the aftermath of the Dutch reform. Examining the impact Dutch policy reform’s effect on the parental subsidy is important for at least two reasons. First, a subsidy from either the government or parents has a positive influence on the educational performance of the students. For instance: drop-out rates from universities and colleges correlate negatively with financial aid for students (Bettinger, 2004; Glocker, 2011); the probability to graduate correlates positively with financial aid (Dynarski, 2003; Glocker, 2011); the average grades of students are increased by receiving more subsidy (Churaman, 1992a ; Rock, Centra, & Linn, 1970). If the Dutch students whose governmental allowance has been forfeited do not receive additional pa-rental subsidies, their performances and delight of studying are likely to decrease. This study’s results are therefore relevant for the interests of the Dutch students.
Although students from poorer households continue to receive a governmental subsidy, the amount these students receive is reduced following the policy reform (SCP, 2013). Baumgartner and Steiner (2006) showed that a comparable policy reform in Germany lead to a decrease in low-income student enrolment. Further research by Callender and Jackson (2005) suggested that students, espe-cially those from poorer households, are averse to borrow money for educational purposes. Therefore, especially poor students are more prone to forgo their educational aspirations when their subsidy is cut off.
A similar argument is valid for non-white students: black parents, for example, are less likely to contribute to their children’s education (Addo, Houle, & Simon, 2016). Students whose parents are black are being financially supported to a minor extent. These students could thus be demotivated to go to college.
3 prone to face anxiety and sleeping problems at the end of their studies and at the beginning of their career.
Student belonging to the cohort starting college on or after 1 September 2015 misses a maxi-mum of a monthly €300 allowance, which covered a substantial part of the expenditures. Thereby are the prices for tuition, books, living, and housing ever increasing. In the case that the absence of the governmental subsidy would entirely be filled by borrowing money, the students’ debt could increase to €15,000 (SCP, 2013). Being discouraged by this debt, the Dutch reform thus could have a conse-quential impact on (prospective) students, unless parental subsidies have increased. The research ques-tion for this study therefore is:
Has the parental subsidy increased due to the Dutch policy reform?
This paper is related to the body of literature discussing reforms in educational financial aid. Van den Berg (2019) investigated the impact of the same Dutch reform on the students’ willingness to move out of their parental house to follow a study program. Since the abolition of the governmental subsidy, less students in the Netherlands have moved out of their family’s home to live on their own. The Dutch reform thus has a noticeable effect on Dutch students. Further research on similar Dutch policy reforms found that a shortening of the period students receives governmental subsidies led to an improvement of educational results (Belot, Canton, & Webbink, 2006). Dutch students are, apparently, quite affectable to governmental reforms, which ratifies the relevance of this study.
Another part of the literature that discusses educational finances focusses on the relationship between household finances and children attending college. The height of the allowance is mostly determined by the age, race, income, and educational level of the head of the household. The marital status of the parents and the number of children attending college affect subsidies as well (Churaman, 1992a; Huston, 1995; Lee, Hanna, & Siregar, 1997; Yilmazer, 2008). Literature focussing on the rela-tionship between student aid policy reforms and college expenditures of households is lacking. This study thus bridges the existing gap in the literature.
The data that is used in this study are extracted from De Nederlandsche Bank (DNB) Dutch Household Survey, constituted by CentERdata. The DNB Household Survey yearly collects data from 2000 people and has done this since 1993. The areas of interest that are captured by the survey are general information about the household; employment; living conditions and mortgages; health and income; possessions and loans; and various economic and psychological concepts2. The data used in-cludes observations from the years 2004 until 2019. In total, there are 1,013 observations consisting of
4 239 households. The panel data is unbalanced: the number of responses of the households ranges from two to thirteen years.
To investigate whether the ceasing of the fixed governmental allowance influenced parental subsidies I make use of a ‘difference-in-differences’ (DiD) method. For using this method, I assigned the poorer households as a control group and the richer parents as the treatment group. The students coming from poorer households continue to receive a governmental subsidy following the policy re-form, while subsidy of students with richer parents has been ceased. The second difference is time-specific: observations after 2015 belong to the ‘after’ group and those recorded in or before 2015 to the ‘before’ group. The model is controlled for the abovementioned determinants of the height of the parental subsidy. Making use of these control variables and the two differences in the data I can observe whether the policy reform has affected the Dutch households.
The paper is organised as follows. Section II reviews the relevant literature on policy reforms and household finances concerning college expenditures. This section also examines the Dutch policy reform in more detail and is followed by the hypotheses. Section III further explains the methodology. Section IV discusses the dataset used and provides descriptive statistics. Section V discusses the results on whether the Dutch policy reform lead to alterations in college expenditures by the households and section VI concludes.
II. LITERATURE REVIEW
This section discusses the literature written about the effects of student aid policy reforms on student behaviour. For the interest of the study of the Dutch policy reform, it is relevant to review the following aspects of the existing literature. The first subsection discusses how the college costs en-countered by students affect the finances of households. The second subsection focusses on how the student’s behaviour is affected by changes in financial subsidies, both governmental as well as parental subsidies. Specifically, the change in student performances and enrolment rates after a subsidy increase or –decrease is discussed. These results are similarly important, since they prognosticate the behaviour of post-reform students in the Netherlands. This section finishes with the made hypotheses for this study. Subsequently, the Dutch policy reform of 2015 is extensively discussed.
II.i College Expenditures and Household Finance
Parents are inclined to transfer money to their studying children, since it is an excellent way to contribute to their social and financial advancement (Churaman, 1992b). Therefore, the economic and financial consequences for parents when their child attends college have been researched by several scholars, which are discussed in this subsection.
5 Especially parents with a lower socioeconomic status are likely to encounter complication when their child starts to attend college. These parents are less likely to have attended college them-selves and are less aware of the importance of attending college. Therefore, they are less likely to sponsor their studying children (Hossler & Vesper, 1993; Swartz, 2008). Furthermore, parents from lower socioeconomic status (SES) insufficiently estimate the costs of college. Their saving behaviour reflects this: low SES parents tend to undersave for higher education, since they initially thought the costs would turn out to be lower (Grodsky & Jones, 2007).
Children from lower income households are more likely to move out of their parental house when they start studying. Further are children from these households, if they continue to live with their parents, more often asked to pay for room and board. Both measures are taken because these parents find it hard to provide for their studying children (Choy & Berker, 2003).
To surmount the ever-incremental college costs, about seven percent of the U.S. households borrowed money for educational purposes. The family’s willingness to borrow for educational expenditures depends negatively on the age of the student, the cash, and savings the household possesses, and the student’s income. Borrowing households with more children are less likely to financially support the students, since lenders see those households as a riskier borrower (Cha, Weagley, & Reynolds, 2005).
Lastly, Churaman (1992b) did find that single parents, and especially single mothers are at disadvantage in two ways. Prior to the start of the child’s college career, single parents have a lower ability to save for future college expenses. While the child is following a program in higher education, the single mothers’ income is only half of the income of a two-parent household, which results in less money available for the child’s study and expenses.
II.ii Student Aid and College Behaviour
The consequences of student aid policy reforms on the behaviour of college students have likewise been researched. This body of literature can be subdivided. First, this subsection discusses how changes in governmental subsidies affect students. These changes in behaviour are consequences of policy reforms affecting either fluctuations in tuition fees or student financial aid. Subsequently, I discuss how alterations of parental subsidies affect student behaviour.
As previously mentioned, this section provides a prognosis of how Dutch students behave after the reform and is therefore profoundly relevant.
II.ii.i Policy Reforms
6 out of 43 studies. The main result of this study is that males are more prone to forgo higher education if tuition rises. In addition, European students are less elastic to tuition changes than their U.S. coun-terparts. This subsection commences with an in-depth discussing of how student aid policy reforms affect enrolment rates and academic presentations of several countries which have been researched. Subsequently, I focus on the impact of Dutch policy reforms on enrolment rates and academic presen-tations of Dutch students.
In the United States enrolment rates are decidedly negatively related to tuition fees and posi-tively associated with financial aid from the government (Leslie & Brinkman, 1986). Dynarski (2000) measures the consequences of a new American scholarship called HOPE. She compares states where HOPE is introduced with states where it is not available. She did find that the enrolment– and the lecture attendance rate have increased significantly due to HOPE. Another U.S. policy regarding al-lowances for poorer students, the introduction of Pell Grants, significantly increased children– and adult enrolment rates in higher education (Seftor & Turner, 2002).
Choy (2001) shows a negative relationship between college tuition and enrolment rates in the U.S. She concludes that the elasticity is higher for students whose parental education is low. Further is the college dropout rate negatively related with the number of students receiving financial aid (Bettinger, 2004). Not only for new students is financial aid for education beneficial: Bound and Turner (1999) conclude that the allowance for U.S. veterans of the Second World War increased college com-pletion and educational attainment.
The effect of a governmental subsidy or an increase in tuition fees on the student’s performance and enrolment rates has been researched in European countries. Glocker (2011) did find that in Ger-many the introduction of governmental subsidies for students from poorer households lets those stu-dents elongate their educational duration. Furthermore, due to this introduction the drop-out rate of these students decreased significantly, and it leads to a major increase in the probability to graduate.
Following the implementation of another policy reform, the German government increased the amount of financial aid the students could receive by ten percent. However, the enrolment rates were hardly affected (Baumgartner & Steiner, 2006; Steiner & Wrohlich, 2012). The introduction and abol-ishment of tuition fees in German universities similarly affects enrolment rates minimally (Baier & Helbig, 2014). Likewise, in Belgium the price elasticity of university enrolment is close to zero. Eligi-ble higher vocational education students are, however, more susceptiEligi-ble to enrol if tuitions are lower (Duchesne & Nonneman, 1998). Furthermore, a large incremental Danish reform in financial aid for students reduced drop-out rates and raised academic completion rates, especially for students from poorer households (Arendt, 2013). The same reform caused an increase in enrolment rates for Danish colleges and universities (Nielsen et al., 2010).
7 financial aid was increased. Murphy et al. (2017) explore this change in financial administration and did find that university enrolment had increased considerably. Further did the trend of an increasing gap between the number of poor and rich students in college stabilise due to the reforms and did the quality of education increase. Dearden, Fitzsimons, and Wyness (2014) did find an enrolment rate increase in the U.K. of four percentage points per thousand British pound student aid.
The reaction of students to policy reforms concerning student finances differs internationally. Solely looking at the literature of foreign countries therefore does not yield consistent forecasts of the situation in the Netherlands. Since this study researches the aftermath of a Dutch reform, researching the alterations of student behaviour consequently to preceding Dutch policy reforms is helpful.
Huijsman et al. (1986) discuss the male and female enrolment rates in the post-war era until the 1980s in the Netherlands. Male enrolment in higher education has a positive relation with govern-mental financial aid and a negative relationship with foregone earnings and, although insignificant, tuition fees. Female enrolment in universities was developing during the studied period, causing in-consequential estimates. Sterken (1995) further analysed the effect of tuition on enrolment and found a significant negative relationship. Encompassing more contemporary data, the gender specific effects of tuition on enrolment rates become indistinguishable. Canton and De Jong (2005) found that tuition fees do not affect enrolment rates for both males and females. However, an increase in financial aid increases the enrolment rates of higher education.
A Dutch policy reform in the nineties shortened the period where students could receive gov-ernmental subsidy with one year, causing the cost of higher education to increase between €700 and €4,400. Belot et al. (2006) investigated this reform and did find that students switched less to other programs and first-year students obtained higher grades. This indicates that students prefer to finish their program faster than to study longer without governmental support.
A recent study by the Dutch Bureau for Economic Policy Analysis (CPB) discusses the effects of the 2015 Dutch policy reform regarding student aid, the same policy reform treated in this study. The CPB did find that the enrolment rates, for both students from lower– and higher-income house-holds, were hardly affected by the ceasing of the governmental subsidy. Further did the CPB conclude that the percentage of students who took out a loan following the research has increased from seventeen to fifty-one. This study has not researched the change in parental subsidies following the Dutch reform.
8 II.ii.ii Parental Subsidies
An early research by Rock et al. (1970) concluded that colleges where the students receive more financial aid achieve better results compared to colleges with ‘poorer’ students. Dynarski (2003) measures the level of attainment of college students, controlled for the financial aid the student receives from his or her parents. When the students’ parental subsidy is enlarged, their college attendance and educational attainment both increases. Further does the study show that financial aid stimulates youth to continue schooling later in life; the initial and fixed college costs are a hurdle for new students.
Churaman (1992a) researched the differences between two-parent and single-parent households. She did find that in both cases the children’s study course is affected by finances and especially children with only a mother indicated that their educational career is impacted by the lack of parental subsidies.
Keane and Wolpin (2001) found that in the United States, the parental subsidies increase the average educational attainment with one year (i.e. the students who receive parental subsidy obtain a higher degree). Besides did the authors find that parental subsidies increase the number of children having at least little college experience with seventeen percent.
The differences in parental subsidies further cause distortions between white and black stu-dents. The latter group is less likely to be financially aided by parents. Consequently, the enrolment rates of black students are lower, and the black students are more in debt after they have earned their degree. The differences in parental subsidy are the causes of the existing racial gap between black and white students (Addo et al., 2016; Charles, Roscigno, & Torres, 2007). This ethnic disparity is not the consequence of the wealth gap between white and black households: even after controlling for parental income, the black students have accumulated a higher college debt after graduation than their white counterparts (Jackson & Reynolds, 2013). This debt emanates from the parental subsidy deficiency of black students in higher education.
Furthermore, hours worked at a part-time job are negatively related to parental subsidies and positively related to college costs. Hours worked at a part-time job are, in turn, negatively related to the educational achievements of the students. Therefore, children who receive less parental subsidy, allocate less time to their studies, causing a decrease in performance (Kalenkoski & Pabilonia, 2010).
9 II.iii Hypotheses
Based on the above discussion of the relevant literature for this study, this subsection provides the hypotheses for this study.
As mentioned before, the research question is ‘has the parental subsidy increased in the after-math of the Dutch policy reform?’ In order to know whether the parental subsidies have increased, it is good to look at the motives for parental subsidies. Intergenerational transactions are critical, as they increase educational attainment (Churaman, 1992a; Bound and Turner, 1999; Dynarski, 2003) and de-crease the drop-out rate (Bettinger, 2004; Glocker, 2011), thus parents of children in higher education should be willing to increase their educational expenditures after governmental allowances decrease.
For a preliminary investigation, the Dutch Institute for Social Research (in Dutch: Sociaal Cultureel Planbureau) interviewed Dutch high-schoolers, parents, and students about their opinions on the 2015 policy reform. High-schoolers indicated to have minimal price-sensitivity, meaning that they are just as willing to go to university as they were before the policy change. Parents indicated to be willing to pay for the education of their children, as they believe that their children’s education is important for their later career. All this should result in higher parental subsidies (SCP, 2013). Further-more, Callender and Jackson (2005) demonstrated the borrow-aversion of students. This aversion could also play a role in the parents’ decision to contribute to their studying children, as they would rather transfer some of their current income than let their children go into debt.
Taking the abovementioned arguments into account, the following hypotheses are composed.
h0 = Parents have not increased their parental subsidies after the Dutch policy reform. h1 = Parents have increased their parental subsidies after the Dutch policy reform.
II.iv The Policy Reform
In 2015, a policy reform was implemented in the Netherlands that affected students from higher vocational colleges (in Dutch: HBO) and universities. This policy reform ceased the fixed part of a subsidy these Dutch students used to receive from the government. The reason for this ceasing was to save funds for developing and improving the educational sector in the Netherlands. By decreasing the allowances for the students, the government saves money which is reinvested in higher education and academic research (Rutte & Samsom, 2012). In total 85,301 students of higher vocational education and 45,956 university students were affected in the first year of the new policy3.
3 These numbers are subtracted from the website of the Education Executive agency of the Dutch Ministry of Education, Culture and Science (in Dutch: Dienst Uitvoering Onderwijs). Retrieved from:
10 Prior to the change in administration, a student that started his or her educational career by either attending university or vocational college could receive financial aid from the government. This financial aid consisted of four categories: the basic subsidy, the supplementary subsidy, the student loan, and the free public transportation card.
The amount of the basic subsidy differed for students living with their parents and students living away from home, where the difference amounted €179. The means-tested additional allowance depends on the parental income, the number of children attending higher education, and the living-situation of the student. The annual income-threshold for the means-tested governmental subsidy was in 2015 €35,000. Eligible children received a supplemental allowance from the government of €238 – €258 per month, depending on whether they lived with their parents or on themselves. The student loan was available for all students, with a maximum of about €500 per month. The travel pass provided free transportation via busses, trains, trams, and metros during either weekdays or weekends. In total, a student living in their parental home received €100, and potentially €338 per month. A student living on their own received €279, and €537 potentially per month.
Students of cohort 2015 and subsequent cohorts received only the possible additional allow-ance, the rights for the student loan, and the free public transportation card. The additional allowance has increased such that students who are eligible can receive €378. It should be stated that the students that are eligible for an additional allowance and lived on their own are also duped from this policy reform. Before, in total these students would receive €537 per month. Post-reform, this total amount decreased with €159. The amount students could borrow has increased to €1,071. The ceased basic subsidies became loans after the policy reform, and the total amount available for the students has increased. All alterations in the governmental subsidies are described in table 1. The old situation covers the amounts the students could receive from the government in 2014, while the new situation depicts the amounts in the first year after the reform was implemented, 2015. The threshold of parental income below which children receive a governmental subsidy was in both 2014 and 2015 €35,000.
Table 1. The governmental subsidy: old (2014) and new (2015) situation (in €)
Parental income < €35,000 Parental income > €35,000
Living with parents Living alone Living with parents Living alone
11 The Dutch policy reform further relaxed the conditions for the student loan. The students have 35 years to repay their debt and are obliged to repay only 4% of their monthly income. Thereby has the interest rate decreased substantially. Because this new policy only applies to the cohort starting in September 2015, it does not affect students who started at a facility for higher education earlier, they continue to receive the fixed allowance.
III. METHODOLOGY
This section describes the research methodology of this research. First, I determine which econometric model is used in this study, followed by the equation of the model. Subsequently, all variables and control variables are explained. Finally, I discuss the shortcomings of the model.
III.i Model Specification
In order to specify the econometric model, the Lagrange Multiplier test, designed by Breusch and Pagan (1980), is executed to find the appropriate model. This test helps to decide whether a (pooled) ordinary least squares (OLS) or a random-effect regression of panel data is pertinent. The null hypothesis of this test is that the individuals are significantly similar, thus an insignificant outcome would suggest using OLS regression. Executing this test yields a χ2 equal to 117.89 (P < .001), indicating that there exists significant variation between respondents. It is thus advantageous to use a panel regression.
Furthermore, the model is tested for robustness. According to the test for heteroskedasticity designed by Breusch and Pagan (1980), the data in this study suffers from heteroskedastic standard errors (χ2 = 105.799, P < .001)4. This is reaffirmed by performing a likelihood-ratio test between an iterated generalised least squares (iGLS) regression with panel-level heteroskedasticity and an iGLS regression without heteroskedasticity specified. This likelihood-ratio test evaluates whether the heteroskedastic model is nested in the non-heteroskedastic model5. Having performed the test, it spec-ifies that heteroskedasticity is present in the model (χ2 = 1,301.46, P < .001). Therefore, to prevent that the model suffers from heteroskedasticity, I continue to use a feasible generalised least squares regres-sion to estimate the model (GLS) with a heteroskedastic error structure.
Further have Bertrand, Duflo, and Mullainathan (2004) found that difference-in-difference models are likely to suffer from autocorrelation. The best way to assess whether autocorrelation is present in the model is by performing the Wooldridge test (Drukker, 2003). This test, designed by Wooldridge (2002), has no autocorrelation as its null hypothesis. Applied to the model of this study, I
4 The heteroskedasticity of the data’s standard errors similarly seen graphically. See the scatterplot between the error terms and the fitted values of the parental subsidy model displayed in figure A2 in the appendix.
5 This additional test for heteroskedasticity is recommended by Vince Wiggins and Brian Poi, both employees of the company supplying and developing the statistical software Stata. Retrieved from:
12 do not reject this null hypothesis (F = 1.286, P > .10). Therefore, I do assume that there is no autocor-relation present in the model.
Subsequently, I use a regression with random-effects. A large disadvantage of the random-effects model is that it, to be valid, assumes that the explanatory variables are exogeneous from the error term. I cannot with certainty declare that there are no unobserved variables that both affect the parental sub-sidy and the households themselves. According to the Hausman (1978) specification test, a random-effects model is more suitable for this research than a fixed-random-effects model is (χ2 = 29.00, P > .05). Thereby would a fixed-effects model discard all time-invariant variables. The larger part of the control variables used in this study are time invariant (e.g. education, gender, region, occupation), thus the model would not be correctly controlled when using a fixed-effects model. Therefore, I continue to use a random-effects model with time– and region fixed-effects included.
Summarising this subsection, I use a feasible GLS regression for estimating the model. The model is specified to have a heteroskedastic but uncorrelated error structure. Further is the regression specified with random-effects, but with dummy variables controlling for year– and region specific fixed-effects.
III.ii Research Methodology
The testing of the hypotheses is done by investigating the difference between two differences present in the data (i.e. difference-in-differences approach). Using a GLS model for this approach is possible when the model is correctly specified (Hausman & Kuersteiner, 2008).
The first difference can be found in the time dimension. The policy reform came into operation on 1 September 2015 and thus affected students starting their college career at this date. Therefore, it is treated as the breakpoint in this study. Initially, the ‘before’ group consists of households that had studying children before 2015 and the ‘after’ group of the other households with children in higher education that started after the breakpoint. The students in the after group receive less money from governmental sources, thus they must seek other sources of money. The students before the reform experienced no change in their governmental subsidy.
13 households with students before the reform are perceived as baseline observations; their governmental allowance has not been affected by the reform. With the following DiD-estimation I can observe whether parental subsidies have increased:
𝑆𝑢𝑏𝑖𝑡= 𝛽0+ 𝛽1𝐼𝑛𝑐𝑖𝑡+ 𝛽2𝑿𝑖𝑡+ 𝛽3𝒀𝑖𝑡+ 𝛽4𝑌𝑒𝑎𝑟𝑖𝑡+ 𝛾{𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖𝑡× 𝑌𝑒𝑎𝑟𝑖𝑡} + 𝑣𝑖𝑡, (1)
with 𝑣𝑖𝑡 = 𝑢𝑖𝑡 + 𝜀𝑖𝑡
where the variable Sub indicates the logarithmic transformation of the amount of parental subsidy for an individual household at a certain year. The logarithmic transformation of the variable reduces the skewness of this variable. Inc represents the household’s annual income and is subdivided into 15 levels of €5,000 difference per level. Again, this subdivision reduces skewness in the data. Please refer to table A2 in the appendix for an overview of the exact subdivision of the household’s income. The error term consists of two parts: u captures the individual-specific error term and ε the idiosyncratic error term.
The vector X controls the model for head-of-the-household-specific characteristics. Vector Y controls for the year– and region specific fixed-effects. Please refer to section III.iii for an overview of the control variables used. The variables in these vectors are important for making the estimations for the different households better comparable. Furthermore, because I use a random-effects model, adding control variables reduces the omitted variable distortion in the model.
The variable Year accounts for the before-aftereffects of the policy. If the observation was taken in 2016 or after, the policy reform has affected the household’s student. In that case, the Year-variable takes the value one. If the observation is from before the policy reform, the Year-variable equals zero. Similarly, Treatment denotes the treatment effect. As is explained above, the households with an income above the threshold belong to the treatment group: their studying children receive no longer a governmental subsidy. If a household has a mutual income above this threshold, treatment equals one and zero otherwise. γ denotes the DiD-estimator that captures the policy and estimates its impact. The coefficient belongs to the two dummy variables described above, multiplied to each other. Following the hypotheses made in the previous section II.iv, the null hypothesis expects that γ equals zero, while the alternative hypothesis suggests that γ is larger than 0.
There still are concerns present within the equation. For example, would the enrolment rates not have been decreased following the Dutch policy reform, and affect the parental subsidy that way? It is shown in earlier literature discussed in section II that Dutch students are not that price sensitive, thus that problem would not alter the result to a large extend.
14 distort the results. This and possible other problems concerning the model are in-depth discussed in the concluding section.
III.iii Control Variables
This subsection discusses which control variables are used and why these are included. To determine these control variables, earlier literature that researched the factors influencing the height of parental subsidy is used.
Huston (1995) attributed the education level of the head of the household as one of the main factors of the height of the parental subsidy. A higher education of the parents results in a higher pa-rental subsidy. Single parents, especially single mothers, contribute less to their children’s secondary education compared to two-parent families (Churaman, 1992a). Yilmazer (2008) argues that the height of the parental subsidy is positively correlated with the age of the head of the household.
Lee et al. (1997) did find that Hispanics and Asian parents save more money earmarked for education than parents of other ethnicities. Black parents contribute less to studying children in com-parison with non-black parents (Addo et al., 2016; Charles et al., 2007).
Further face parents a trade-off parents between saving for retirement or financially supporting their children during studies. Parents thus have to decide whether they spend their money on their studying children or invest this money into their own retirement (Browning & Lusardi, 1996).
The parents’ financial literacy is an important explainer of parental subsidy. Hossler and Vesper (1993) have found that parents that have more knowledge of college costs and financial matters save more money earmarked for their children’s higher education. Families with great financial knowledge invest more in their children’s education, since they are more familiar with the positive consequences of higher education (Swartz, 2008). Furthermore, socioeconomic disadvantaged parents underestimate the costs of college and, consequentially, are contributing less to their studying children (Grodsky & Jones, 2007).
At last is the height of the costs of education is an important factor too; increasing college expenditures lead to a higher percentage of parents that borrow money to finance their child’s education (Cha et al., 2005). Please refer to table A4 for an overview of the tuition fees used in the Netherlands in the years 2004 to 2019.
15
IV. DATA
This section defines the dataset used and provides the descriptive statistics. First, I explain how the data is gathered and which variables are used. Afterwards, figures and tables depicting and describ-ing the used data are displayed and explained. Lastly, the limitations of this dataset are discussed.
IV.i Dataset
I use survey data of the De Nederlandsche Bank Dutch Household Survey, collected and con-stituted by CentERdata. The survey has been actively updated since 1993. CentERdata annually sur-veys about 2000 Dutch households and publishes an overview containing over 1800 different variables and 4600 observation. All the members of the household that are over 16 years old are invited to fill in the questionnaire. If it happens that a household wants to drop out of the survey a new household with similar characteristics is sought to replace the dropped household (Teppa & Vis, 2012). This prevents attrition of the dataset.
The obtained variables are subdivided into six surveys which question the following aspects: general information about the household (1); work (2); accommodations and mortgages (3); health and income (4); assets and liabilities (5); and several psychological en economical concepts (6). For this study I have gathered data from the first, second, fourth, fifth, and sixth survey. These surveys are used to compose a dataset containing information about parental subsidies, income levels and the before-mentioned control variables (cf. Section III.iii). Please refer to table A1 in the appendix for an extensive description of all the variables used and the survey questions underlying these variables.
16 The income and subsidy variables are lagged one year. The amounts reported actually are the amounts of the previous year. For example, in the survey taken in 2018, the household report their income of 2017. Because the Dutch policy reform affected students in 2015 and consecutive cohorts, the surveys of 2016, 2017, 2018 and 2019 contain post-policy information. For the surveys taken before 2016, the students of the concerning households still received a fixed governmental subsidy. The total sample includes the most current information available. For this study I have gathered data from the years 2004-2019. There are 206 recorded observations after and 807 observations before the Dutch policy reform. In total, there are 1,013 observations and 239 households.
In the case where the household responded with don´t know to the question how much they have given their studying children, a missing value is generated. I have replaced these missing values with the mean of subsidies given of the income group the household belonged to. For example, a household with income level 7 (€35,000 – €40,000) did not know exactly how much they had subsi-dised their children. The average subsidy given in that income level is €3,705.85, thus the household in the example is expected to give €3,705.85 to their studying children.
Because I use a difference in differences method for determining the effect of the policy reform, I use only panel data. This makes it necessary to have observations that at least have responded twice. Households that only respondent to the survey once were deleted from the sample. The total sample consisted of 50 to 80 heads of the households per year. The panel data is unbalanced: the num-ber of responses from a distinct household fluctuates from two to thirteen. On average, each household has responded to the survey 4.2 times.
IV.ii Descriptive Statistics
Table 2 provides summary statistics for the used variables for the household dataset. As can be seen was the average income of the households before 2015 about €40,000 and after the reform about €47,000. The parental subsidy given was about €3,450 before the reform and about €3,100 after. The total subsidy therefore seems the have dropped substantially. However, this number takes the observa-tions of households who did not give subsidy to studying children into account.
The percentage of households that does not subsidise students is larger post-reform (17.31% before and 36.41% after the reform), highly influencing the average subsidy. When solely looking at nonzero subsidy amounts, the average parental subsidy given is approximately €4,200 before and €4,900 after the reform, which is respectively 10.45% and 10.38% of the average households’ income. This indicates that the Dutch policy reform has not decreased the average subsidy given. The reform only decreased the number of households that have given subsidy to their studying children.
17 and 40% of the Dutch students received a supplementary subsidy from the government, thus the figure in the sample corresponds to the Dutch population (CPB, 2020).
Table 2. Summary statistics of the dependent variable and its explanatory variables
Variable Notation Pre-reform Post-reform
Parental subsidy in € 3,456.05 3,111.74
Did give subsidy to studying children if yes: 1; if no: 0 0.8269 0.6359 Parental subsidy if household has given subsidy in € 4,179.53 4,893.26
Income household in € 40,012.13 47,146.00
‘Poor’ households % 37.86 32.04
Background
Gender If female: 0; if male: 1 0.6538 0.6262
Age 56.81 58.30
No. children belonging to the household 0.7385 0.7282 No. children not belonging to the household 1.9601 1.7476
Single parent household % 16.56 14.08
Highest educational attainment 1-5 3.6476 3.6650
Occupation 1-4 1.7870 1.6893
Knowledge of finances 1-4 2.2352 2.2316
Health 1-5 3.8792 3.4951
Illness if yes: 1; if no: 0 0.2802 0.2136
Degree of urbanisation 1-5 2.8356 2.8738
Additional pension plan if yes: 1; if no: 0 0.5131 0.4175 Region fixed-effects
Three largest cities if yes: 1; if no: 0 0.1644 0.1067
Other West if yes: 1; if no: 0 0.2814 0.3010
South if yes: 1; if no: 0 0.2316 0.1990
East if yes: 1; if no: 0 0.2291 0.2379
North if yes: 1; if no: 0 0.0934 0.1553
Saving reasons
Finances children 1-7 4.9714 5.1805
Supplement pension 1-7 4.9801 5.5293
Observations 807 206
Number of households 213 79
There are more masculine heads of the households, but after the reform the number of feminine heads of the households increased. Their age has increased as well: was it on average 56 years before, it is 58 after the reform. Few of the households are ran by a single parent. On average is there less than one child present in the household. Further are less children no longer a part of the household after the household. This could indicate that students stay longer at their parent’s house to save money, as Van der Berg (2019) and the CPB (2020) have observed as well.
18 education (basisonderwijs/VMBO); higher secondary education (HAVO / VWO); lower vocational education (MBO); higher vocational education (HBO); and university (WO). The mean education at-taint by the head of the households is approximately 3.6, indicating an average education between lower–and higher vocational education. The occupations are divided into four levels as well: employed; self-employed; unemployed; and retired or disabled. The average level in has decreased between the two periods with about 0.1, indicating that after the reform more households received their income either via their employer, or via their own business and less households were unemployed, disabled, or retired.
The self-reported knowledge of financial matters remains unchanged after the policy reform. The respondents could indicate whether they thought of themselves as being not knowledgeable (1) to very knowledgeable (4) about financial matters. The Dutch head of the households consider themselves relatively knowledgeable about financial matters.
Furthermore, about half of the household made additional arrangement for their pensions. This could have either have been done via annuities; through whole life policies; by buying extra pension rights via their employer; through extra periodical payments via employer; or by other means. These plans are thus additional to the regular pension plans of the employer. About half of the households had made such plans before the reform, and about 40% after the reform.
Before the policy reform, the majority of the respondents lived in one of the three largest cities in the Netherlands or else in the west of the country. Few respondents lived in the northern region. After the policy reform, the living region of the respondents was scattered to a larger extent. Still, the majority still lived in the western region of the Netherlands, but more northern households were in-cluded present in the sample. This could suggest that more northern households were invited to participated in the DNB Household Survey, or more northern households started subsidising their studying children.
The households are also asked how important they thought the following reasons for saving are: to help their children if they have financial difficulties; and to supplement their general old-age pension. On a scale between 1-7 they could indicate whether these reasons are very unimportant (1), very important (7), or somewhere in between. Both reasons for saving became more important to the households after the reform.
19 parental subsidy drops afterwards, thus so far there is no trend of increased subsidy visible. In 2019, the parental subsidy seems to have increased to pre-reform levels.
Figure 1. The average parental subsidy (left axis) and the average income of the households (right axis) over the period studied.
The graphs in figure 1 only depict mean subsidy levels of both ‘rich’ and ‘poor’ households. For the interest of this study, it is relevant to subdivide this graph and observe whether there is a dif-ference visible between the two subgroups.
Figure 2. The average subsidy given per year to students by the control group (households with an income below the governmental threshold for an additional subsidy) and by the treat-ment group (with an income above this threshold).
Figure 2 demonstrates the difference in parental subsidies between students coming from ‘poorer’ households and those from ‘richer’ households. The household is considered poor if their annual income is below the government’s threshold for determining which students receive an addi-tional subsidy and richer households have an income above this threshold.
The averages follow a comparable trend: when the subsidy from the rich increases, the subsidy from the poorer households increases as well. The policy reform took place in the middle of 2015. As
0 10.000 20.000 30.000 40.000 50.000 60.000 0 1.000 2.000 3.000 4.000 5.000 6.000 Ho u seh o ld i n co m e (in € ) P ar en tal s u b sd y ( in € )
Subsidy (left axis, in €) Income (right axis, in €)
0 1.000 2.000 3.000 4.000 5.000 6.000 P ar en tal s ub sid y (in € )
20 is explained before, the parental subsidy-amounts reported to the survey are lagged with one year. 2016 should therefore be the first year where the effect of the 2015 policy effect is noticeable. There is substantial decrease of the parental subsidy from the poorer households visible. The richer households did increase their subsidy. As is explained in section II.iv, the policy reform has been especially disad-vantageous for students with richer parents, which explains this occurrence. In the posterior years, 2017 and 2018 specifically, the subsidy given by richer households decreases as well. In 2019, the subsidy of both groups has increased again.
Finally, figure 3 shows the average parental subsidy categorised for the income level of the households. The income of the households is divided into fifteen distinct levels. The exact division can be found in table A2 in the appendix. Overall, each level consists of a €5,000 difference in income. Level 1 captures all incomes below €10,000 annually, level 2 the incomes between €10,000 and €15,000, and so forth. The group with an income level 15 consists of all income above €75,000. On average, parents with a higher income and a better degree subsidise their studying children more.
Figure 3. Income levels (1-15) of the households and the average annual parental subsidy of household in the income level (in €).
IV.iii Limitations
As is explained above, the only usable responses from the survey are the ones that have re-ported to have given parental subsidy for studying children at least once. There are quite few house-holds that have done this. Therefore, the number of observations is relatively low. If all responses of the survey for the sixteen years used in this study would have been encompassed, the number of ob-servations is just over 30,000. Due to the stratification I had had to make, the number of obob-servations decreased to only 1,013, which might affect the results. Would there have been a more straightforward way to observe whether parents did not subsidise their studying children, the number of observations would have increased. As mentioned before, it is unfortunate that the survey does not include questions about the children that do no longer belong to the household.
21 Due to this limitation, I also cannot observe consider the number of students living away are subsidised by the households. I can therefore not observe how high the subsidy per child is. Nor can the age and level of education of the students not been taken into account. Earlier research has shown that the ethnicity, age, and level of education of the students significantly affects the height of the parental subsidy (Cha et al., 2005; Churaman, 1992b). Therefore, not being able to include these vari-ables in the model biases the results (i.e. the probability of the presence of an omitted variable bias is increased due to this limitation).
Further was the Dutch policy reform only implemented in 2015, causing to have few post-reform observations. As was shown in table 2, only 206 observations after the Dutch policy post-reform could be incorporated in the sample, which might be insufficient for yielding representative results.
V. RESULTS
As detailed in the methodology section, I start the analysis by estimating equation 1. Initially, I focus on the case where γ equals 0. This regression therefore focusses only on the difference between before and after the Dutch policy reform. The first estimates cover the subsidies given by households to their studying children. The various control variables used in the estimates are specific for the heads of the household. Further is the model controlled for time– and region specific fixed-effects. Likewise, the inclusion of these fixed-effects alternate depending on the other control variables. In the initial before-after analysis, where γ equals 0, the year specific fixed-effects cannot be included because of multicollinearity. The saving reasons are as well interchangeably used in the estimates. Subsequently, I forgo of the before-made assumption that γ equals 0. In the second subsection the results of the DiD estimations are discussed. Similarly, the control variables and the two different fixed-effects are used interchangeably. Lastly, I perform a heterogeneity analysis.
V.i Before-After Analysis
The first three columns in table 3 show the estimates of the before-after analysis, controlled for various combinations of variables. The income of the households is, as mentioned in the previous section, subdivided into fifteen levels. According to expectations does the income of the household significantly influence the subsidy given to the students. If the mutual parental income increases with €5,000, or one income level, the student can expect an increase in subsidy between 14% and 19%.
22 subsidy increases with 0.5%. After the maximum, each incremental year causes a decrease of the same amount in subsidy.
Education of the parent is a crucial factor too. Parents who have attended higher vocational education tend to give approximately 50%-30% less subsidy than parents with a university degree, and a parent that following higher vocational education gives 50%-30% more than a parent who followed lower vocational education. The occupation of the parents does not have a large nor a significant effect on the subsidy given. The parental employment status thus does not influence the subsidy given to a considerable extent.
The number of children present in the household does not affect the amount of subsidy a stu-dent receives. The number of children present outside of the household, however, are positively corre-lated with the amount of subsidy. Overall, an extra child outside of the household increases the amount of subsidy between 1.7% and 6.6%. Having a partner does not significantly relates with the amount of subsidy a student receives.
Further, the knowledge of financial matters influences the subsidy given significantly. More knowledgeable parents tend to increase their subsidy in comparison with less knowledgeable parents. Tuition costs for colleges likewise have a significant effect on the subsidy. If the tuition increases with one euro, parents increase their subsidy with 0.2% in general.
The health of the households does not seem to have a significant effect on the tuition. Both parents in a good health status as well as parents in bad health status subsidise their studying children. Similarly, it does not matter whether the parents suffer from a long-term illness.
23 Table 3. Estimation results of equation 1. The estimations in this table are made under the
as-sumption that γ equals 0. This way, the difference between subsidies before and after the Dutch policy reform. The model is controlled for various variables, saving behaviour, and region spe-cific fixed-effects. For estimating the model, a GLS method is used with heteroskedastic error structure. Before-After Analysis (1) (2) (3) Income (1-15) 0.188*** (0.017) 0.149*** (0.016) 0.147*** (0.017) Gender 0.318*** (0.100) 0.099 (0.110) Age 0.553*** (0.098) 0.519*** (0.096) Age2 -0.005*** (0.001) -0.005*** (0.001) Education (1-5) 0.515*** (0.053) 0.392*** (0.053) Occupation1 Self-employed -0.055 (0.143) Unemployed 0.276 (0.227) Retired/disabled -0.160 (0.142)
No. children in household -0.038
(0.050)
0.004 (0.055)
No. children outside household 0.066**
(0.029) 0.017 (0.031) Partner -0.023 (0.136) -0.151 (0.163) Knowledge of finances 0.129* (0.069) 0.174*** (0.066) Tuition -0.002*** (0.000)
Additional pension plan 0.203**
(0.095) Health 0.085 (0.076) Illness -0.129 (0.130) Degree of urbanisation -0.043 (0.040) Year -1.101*** (0.216) -0.827*** (0.188) -0.360 (0.229) Constant 5.737*** (0.133) -10.884*** (2.817) -6.630** (2.812) Observations 1,013 1,013 1,013 Number of households 239 239 239
Saving reasons included No No Yes
Region fixed-effects No No Yes
Note. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
24 Table 4. Estimation results of equation 1. The estimations in this table are made with taking all parameters of equation 1 into account. With this assumption, the difference-in-differences are estimated. The model is alter-natingly controlled for various control variables, saving reasons, year specific fixed-effects and region specific fixed-effects. For estimating the model, a GLS method is used with heteroskedastic error structure.
Difference in Differences Analysis
(4) (5) (6) Income (1-15) 0.158*** (0.018) 0.126*** (0.018) 0.127*** (0.018) Gender 0.003 (0.115) -0.022 (0.112) Age 0.680*** (0.112) 0.511*** (0.111) Age2 -0.006*** (0.001) -0.005*** (0.001) Education (1-5) 0.314*** (0.053) 0.314*** (0.056) Occupation1 Self-employed -0.155 (0.159) Unemployed 0.291 (0.242) Retired/disabled -0.269* (0.148)
No. children in household -0.005
(0.059)
-0.028 (0.059)
No. children outside household 0.054
(0.036) 0.012 (0.039) Partner 0.094 (0.149) -0.118 (0.167) Knowledge of finances 0.087 (0.078) 0.112 (0.079) Tuition -0.001 (0.001)
Additional pension plan 0.337***
(0.107) Health 0.516*** (0.090) Illness -0.002 (0.137) Degree of urbanisation -0.059 (0.043) Treatment×Year 2.676*** (0.446) 2.156*** (0.440) 1.796*** (0.411) Constant 6.686 (0.120) -12.764*** (3.237) -8.544** (3.481) Observations 1,013 1,013 1,013 Number of households 239 239 239
Saving reasons included No No Yes
Region fixed-effects No Yes Yes
Year fixed-effects Yes Yes Yes
Note. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
1 The different occupations are compared to the situation where the head of the household is employed on a contractual basis.
25 V.ii Difference-in-Differences Analysis
The three columns in table 4 depict the difference in differences estimates. The coefficient cohering to the variable Treatment×Year denotes the policy effect. The policy variable consists of two dummy variables, multiplied by each other. The first dummy denotes whether the observation has been made before or after the policy reform and the second dummy indicates whether the household belongs to the ‘poor’ family group. Because the students from poorer families still receive a governmental subsidy after the reform, those are treated as the baseline observations. The threshold for determining which students receive an additional subsidy from the government has changed throughout the years. Table A3 in the appendix provides an overview of these thresholds used throughout the years. House-holds with a mutual income below the threshold are treated as poor families and receive a value of 1. For example: a household has a mutual income of €40,000 in 2004. This household does not belong to the poor group and therefore it belongs to the treatment group. A household with a mutual income of €26,000 in 2004 is below the government’s threshold. The child of these parent receives a governmen-tal subsidy and is therefore treated as a poor household and is not treated by the policy reform.
The mutual income of the households remains important for determining the parental subsidy in the DiD estimations, just as the education, age, and gender of the head of the household are. The values of the control variables do not alter substantially. Therefore, these are not discussed in full detail.
The coefficient of interest in the three columns of table 4 is the policy parameter γ, which belongs to the variable Income×Year. The sign of this coefficient is positive all three columns. The coefficient also is significantly estimated in all three columns. The values are, depending on which control variables were used in the estimation, 2.676, 2.156, and 1.796.
This result is interpreted in the following way. For After the Dutch policy reform has the pa-rental subsidy of the treatment group, thus the households with an income above the threshold for governmental subsidy, increased compared to the control group. This increase in subsidy is ranges between 1.8 and 2.7. Because the depend variable subsidy is logarithmically transformed, the effect should be denoted in percentages. After the reform, the parents with an annual mutual income above the governmental threshold increased their parental subsidy approximately twofold compared to the control group. Therefore, taking this result into account, I can assume that the Dutch policy reform had a significant effect on the subsidy given by ‘richer’ parents.
V.iii Heterogeneity Analysis
26 Table 5 provides an in-depth analysis of the difference in differences of three subgroups in the sample. For these heterogeneity estimations I use the same equation as before, with the same econo-metric specifications. However, I estimated the model with only part of the total sample: the results are provided for stratifications by gender, age, and level of education. The results depicted in table 5 are the estimations of the difference-in-differences coefficient. For the complete estimations, please refer to table A5 in the appendix.
Table 5. Estimates of the DiD-coefficient stratified by separate groups. For this estimation, various control variables are used, as well as saving behaviour controls and region– and year specific fixed-effects. For estimating the model, a GLS method is used with heteroskedastic error structure.
Gender Age1 Education2
Male Female Young Older Lower Higher
DiD 2.170*** (0.535) 0.567 (0.629) 2.124*** (0.595) 1.579*** (0.544) 1.289** (0.613) 3.812*** (0.578)
Control variables included Yes Yes Yes Yes Yes Yes
Observations 655 357 441 571 357 655
Number of households 153 92 143 148 91 150
Note. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
1 Parents with an age below 50 are considered to be young, while an age above 50 is considered to be old. 2 Primary–, secondary– or lower vocational education is considered as lower education, higher vocational education or university is considered higher education.
First, in the case where the model is regressed considering heads of the households that are male, the DiD estimate is higher compared to the estimation with only female heads of the households. This indicates that children of families where the head of the household is a man experienced a greater increase in their parental subsidy, compared to fellow students whose family is ran by a woman. In other words, males are more susceptible for the policy reform than females are.
A large gap is observed between the DiD estimations of younger and older parents. The younger parents are more affected by the Dutch policy reform and have increased their subsidies to their stud-ying children. The older parents are less susceptible for the policy reforms and have increased their parental subsidy to a minor extent.
The last subgroup evaluated is the difference between lower-and higher educated parents. For this particular analysis are parents with high school– or lower vocational education considered to be lower educated. Parents with either university– or higher vocational education are considered as higher educated. The difference between both estimates is quite considerable. Parents who have followed a higher education are more affected by the policy reform compared to the lower-educated parents. The explanation might be that the parents who followed higher education themselves might know how the ceasing of the basic subsidy affects the student’s finances, as they profited from this subsidy themselves when they were students. This last result is expected by Grodsky and Jones (2007), who indicated that parents with a lower socioeconomic status, which is commonly correlated with a lower educational attainment, are relatively poor at predicting total costs of college.
27
VI. CONCLUSION
This study investigates whether the parents of studying children have increased their subsidy following the Dutch policy reform in 2015. Before this reform was implemented, students of higher education in the Netherlands received a monthly governmental allowance. The amount of which varied between €100 and €300, depending on whether the student lived on its own or with their parents. The students were also eligible to receive an additional governmental allowance, which was means-tested on the income of their parents. After the reform, the basic student allowance was ceased. Only the students with a parental income below a certain threshold continued to receive a monthly grant from the government. In this paper the policy reform is treated as an exogenous shock and used to research whether the parents filled in this gap in the student’s finances made by the ceasing of the governmental subsidy.
In order to execute this research, I used survey data from the Dutch Household Survey for the years 2004-2019. The research is conducted by implementing a difference-in-differences analysis. The control group consists of households earning a mutual income below the threshold of the government, because students from these households continued to receive a governmental subsidy. The treatment group are households with a mutual income above this threshold, because the students from this group did not receive any subsidy from governmental sources.
The results for the before-after analysis showed which the most important determinants of the subsidies are. The income of the household, the age, education, and the gender of the head of the household are contributing a large value to the subsidy the student receive from its parents. Considering all households and not making a subdivision between ‘poor’ and ‘rich’ households, the subsidy had dropped substantially following the Dutch policy reform. Subsequently, I allowed the model to consider the difference between these ‘poor’ and ‘rich’ households. This analysis, the DiD-analysis, yields positive significant results for the DiD estimator, indicating an increase in the parental subsidy after the Dutch policy reform. An additional heterogeneity analysis determined that parents who are male, younger or have attended higher education tend to be more susceptible for the policy reform. The estimation of the DiD estimation is higher when the dataset is controlled for these characteristics of the households.
This paper fills in a gap in existing literature. Previous scholars have focussed on how policy reforms similar to the Dutch policy reform in 2015 affected grades, performances, and enrolment rates (cf. Section III.ii.i). Other scholars researched the influence parental subsidies have on the student’s behaviour or how the college costs affected the finances of the households (cf. Section II.ii.ii). Hitherto, research on how policy reforms affect parental subsidies does not exist. This study, accordingly, fills the existing gap in the student aid reform literature.
28 to receive an additional governmental subsidy, while the ‘richer’ households increased their parental subsidy. Would this not have been the case, the post-reform performances of the Dutch student were likely to have deteriorated as a considerable amount of scholars have shown before (see for example Dynarski, 2000; Kalenkoski & Pabilonia, 2010).
The drawbacks of this study are that the Dutch Household Survey does not request information about the studying children themselves. The determination of age and level of education of the students was impossible, nor could I establish the number of children of the household that were enrolled in higher education. Due to this deficiency the number of observations is relatively low. Another limita-tion is that the policy reform only was implemented in 2015 and the informalimita-tion in the surveys is lagged one year. Consequently, I was only able to encompass 206 observations after the reform, which decreases the external validity of this research. A third limitation is the omitted variable bias. Because the choice of variables was limited by the questions in the survey, there is a probability of the presence of an omitted variable bias in this study. An example of a missing variable is the ethnicity of the head of the household, a large determinant of the parental subsidy (Addo et al. 2016; Lee et al.,1997).
29
REFERENCES
Addo, F. R., Houle, J. N., & Simon, D. (2016). Young, black, and (still) in the red: Parental wealth, race, and student loan debt. Race and Social Problems, 8(1), 64-76.
Arendt, J. N. (2013). The effect of public financial aid on dropout from and completion of university education: evidence from a student grant reform. Empirical Economics, 44(3), 1545-1562.
Baier, T., & Helbig, M. (2014). Much ado about €500: do tuition fees keep German students from entering university? Evidence from a natural experiment using DiD matching methods. Educational Research and Evaluation, 20(2), 98-121.
Baumgartner, H. J., & Steiner, V. (2006). Does more generous student aid increase enrolment rates into higher education? Evaluating the German student aid reform of 2001.
Belot, M., Canton, E., & Webbink, D. (2007). Does reducing student support affect scholastic perfor-mance? Evidence from a Dutch reform. Empirical Economics, 32(2-3), 261-275.
Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differ-ences estimates?. The Quarterly journal of economics, 119(1), 249-275.
Bettinger, E. (2004). How financial aid affects persistence. In College choices: The economics of where to go, when to go, and how to pay for it (pp. 207-238). University of Chicago Press.
Bodvarsson, Ö. B., & Walker, R. L. (2004). Do parental cash transfers weaken performance in college?. Economics of Education Review, 23(5), 483-495.
Bound, J., & Turner, S. (2002). Going to war and going to college: Did World War II and the GI Bill increase educational attainment for returning veterans?. Journal of labor economics, 20(4), 784-815.
Breusch, T. S., & Pagan, A. R. (1980). The Lagrange multiplier test and its applications to model spec-ification in econometrics. The review of economic studies, 47(1), 239-253.
Browning, M., & Lusardi, A. (1996). Household saving: Micro theories and micro facts. Journal of Economic literature, 34(4), 1797-1855.
Callender, C., & Jackson, J. (2005). Does the fear of debt deter students from higher education?. Jour-nal of social policy, 34(4), 509-540.
Canton, E., & De Jong, F. (2005). The demand for higher education in the Netherlands, 1950– 1999. Economics of Education Review, 24(6), 651-663.
Cha, K. W., Weagley, R. O., & Reynolds, L. (2005). Parental borrowing for dependent children’s higher education. Journal of Family and Economic Issues, 26(3), 299-321.
