Reducing health costs comes at a price?
A study investigating the policy change from the no-claim insurance
scheme to the compulsory deductible in the Netherlands and the effect
on inequality in subjective health and health consumption.
Mariëlle Passchier, s1652745 University of Leiden
Faculty Governance and Global Affairs Master Public Administration
Specialization Economics & Governance
Supervisor: E. Suari Andreu Second reader: H. Vrijburg 11/06/2019
Abstract
The Dutch insurance system switched in 2008 from a no-claim rebate scheme to a compulsory deductible. In the first eight years after its introduction, the price of the deductible increased with more than 150%. The aim of the research is to investigate whether this policy change affected inequality between socioeconomic groups in terms of health consumption and health status. In addition, the literature showed that this difference between high- and low-income groups can potentially relate to liquidity constraints. Low-income groups are more prone to be liquidity constrained.
The research uses survey data from the LISS panel. The research first looks at the years 2007 and 2008 to study the effects of the introduction of the deductible as compared to the no-claim policy using a regression discontinuity design. The results do not show a more negative effect for the lower-income group but do display some evidence for a difference between liquidity and non-liquidity constrained individuals in terms of health and health consumption. Second, the years 2008-2016 are used to establish the effect of increases in the deductible over time using an ordinary least squares regression. In this case, the results did show a negative trend in subjective health for the low-income group, but not for health consumption. Moreover, the difference between the liquidity and non-liquidity constrained group remained over the years.
While there was some evidence that suggests a difference between liquidity and non-liquidity constrained individuals, the research found no significant difference between these two groups. The results for the low-income group are even more mixed. Therefore this research does not provide evidence for an increase in inequality caused by the policy switch from a rebate to a deductible in the Netherlands.
Table of Contents
1. Introduction ... 4
Added value ... 6
2. The Dutch health care system ... 9
2.1 Determinants of expenditure increases ... 9
2.2 Policy design ... 10
3. Literature review... 12
3.1 Health inequality ... 12
3.2 Reducing health care costs ... 14
3.3 Liquidity constraints ... 15
3.4 Hypotheses ... 16
4. Research design ... 17
4.1 Method of data collection ... 17
4.2 Case selection ... 19 4.3. Operationalization of variables ... 20 4.3.1 Independent variables ... 20 4.3.2 Dependent variables ... 23 4.3.3 Control variables ... 26 4.4 Method of analysis ... 30
Common trend assumption ... 33
4.5 Validity and reliability ... 34
5. Results ... 36
5.1 Regression discontinuity model ... 36
5.2 OLS results ... 41 5.2.1 Liquidity constraints ... 41 5.2.2 Income categories ... 46 6. Conclusion ... 53 Discussion ... 55 Databases ... 57 References ... 57 Appendices ... 61
Appendix A Overview variables ... 61
Appendix B Trends health categories ... 62
List of Tables and Figures
Figure 1 Health expenditure growth between 1997 - 2017 ... 4
Figure 2 Increase in the deductible between 2008-2016 ... 20
Figure 3 Distribution of health by income group ... 23
Figure 4 Total health consumption per year ... 24
Figure 5 Distribution of age ... 26
Figure 6 Distribution of age by income category ... 26
Figure 7 Income categories by gender ... 26
Figure 8 Total unemployment per income category ... 27
Figure 9 Levels of education per income category... 28
Figure 10 Trend subjective health by Liquidity Constrained status 2009-2016 ... 43
Figure 11 Trend subjective health by Liquidity Constrained status including controls ... 43
Figure 12 Trend Health Consumption by Liquidity Constrained status 2009-2016 ... 43
Figure 13 Trend Health Consumption by Liquidity Constrained status including controls ... 45
Figure 14 Trend Subjective Health low, middle- and high-income group ... 49
Figure 15 Trend Subjective Health low, middle- and high-income group including controls ... 49
Figure 16 Trend Health Consumption by low-, middle- and high-income ... 51
Figure 17 Trend Health Consumption by low, middle and high income including controls ... 51
Table 1 Trend in deductible, inflation rate, health expenditure and GDP ... 6
Table 2 Sample response and non-response Health Waves ... 18
Table 3 Income categories ... 21
Table 4 Summary statistics of income by group ... 21
Table 5 Liquidity constrained individuals per income group ... 22
Table 6 Categories of Subjective Health ... 23
Table 7 Total health consumption by income group ... 24
Table 8 Summary statistics subjective health and health consumption by liquidity constraint status .. 25
Table 9 Distribution levels of education ... 29
Table 10 RD output by Liquidity Constrained status ... 36
Table 11 RD output by Income group ... 38
Table 12 Regression output by Liquidity Constrained status ... 41
Table 13 Regression output by Income category ... 47
Appendix Table 1A Description of all the variables used ... 61
Figure 1B Trend poor health for the different income groups………62
Figure 2B Trend moderate health for the different income groups………62
Figure 3B Trend good health for the different income groups………...62
Figure 4B Trend very good health for the different income groups ………..62
4
1.
Introduction
The distribution of health is unequal. People in lower socioeconomic groups have poorer health, tend to live shorter and have a higher chance of diseases compared to people in higher socioeconomic groups (Kröger, Pakpahan & Hoffmann, 2015). This trend increased in the last decades, despite government efforts to reduce this inequality (Mackenbach, 2012, pp.761-762). There is a broader link between health and socioeconomic variables; this is called the social gradient. The social gradient in general means “the lower an individual’s socioeconomic
position, the worse their health” (The World Health Organization [WHO], 2013). According
to Barr (2012, pp.235-236) the social gradient relates to the fact that poorer people suffer more stress, have worse diets and less good access to health care and thus more diseases than wealthier people. Moreover, Van Doorslaer & Masseria (2004) showed using survey data from 21 OECD countries for the year 2000, that richer people tend to consume more specialist medical health care compared to lower-income groups, even when controlled for equal need. Socioeconomic inequalities can cause inequality in access, which can reduce health status. Inequality in access can be amplified or reduced by the used public health care system in a country. Inequality in access relates to the distribution between private and public financing. According to Devaux & De Looper (2012, pp.28-29) a larger part of public health spending leads to lower inequality in access. Contrary, private financing has the opposite effect and creates a bigger risk of exclusion for the lower socioeconomic groups.
Figure 1 shows that in the past decade OECD countries increased their spending on health care. This includes both public and private expenditures. According to an OECD rapport (2015, pp.24-26), the rise in health expenditures exceeded both the inflation and economic growth rate in the years before the economic crisis in 2008. The economic crisis resulted in years of flat or even negative health expenditure rates in most countries. However, fiscal sustainability remains
0 5 10 15 20 AUS AUT BEL CA N C ZEC H D EN FIN F R A G ER G R E HUN IC E IR E ISR ITA JA P KOR LU X N ET N EW NOR PO L P O R S LO V A K SPI SWE SWI TU R U N ITED UK
Health expenditure growth as percentage GDP 1997 2017
5
an issue. Models predict that health expenditures will continue to rise in the future due to institutional characteristics, demographic changes (aging of the population), new expensive technologies and rising incomes (OECD, 2015). Since health is a luxury good, the willingness and ability to spend money on health will increase as income increases (Van der Horst, Van Erp,
& De Jong, 2011). Governments are facing a dilemma; on the one hand health care expenditures are putting fiscal pressure on government budgets. There is a need to reduce spending on health care for the government budget to remain sustainable in the future. On the other hand, reducing public health spending can increase the already existing health inequalities between higher and lower-income groups. This research will investigate the Dutch case of approaching this problem of increasing health expenditures – and how the Dutch solution has affected equality in health status and health consumption.
In the years 2005-2007 the Dutch health care system used a no-claim reimbursement scheme. This scheme benefited people who did not make any health care costs. Individuals could receive up to a maximum of 255 euros when they did not incur health care costs. If you incurred health care costs, but less than 255 euros, you received the remaining part of the no-claim after the end of the insurance year. Visits to the general practitioner (GP), care for children under the age of 18, obstetric care and maternity care were excluded from the no-claim. People could choose to have a voluntary deductible, to reduce their monthly premium but only five percent of the insured choose this option in 2007 (Goudriaan, Lalloesing & Vemer, 2007, pp.26-27). However, to reduce the increasing health care costs the system changed in 2008 with the introduction of an obligatory deductible. For the basic insurance in 2008, every adult had a compulsory deductible of 150 euros. Everyone needed to pay the first 150 euros of health care costs themselves. After this, the insurer paid the costs (Goudriaan et al., 2007, p.9).
The price of the deductible was coupled to the health care expenditures. When the costs of health care increased, so did the deductible. This changed in 2018 with the implementation of a new law that freezes the price of the deductible on 385 euros until 2022 (Rijksoverheid, 2018). However, in the first eight years after its introduction, the price of the deductible increased with more than 150%. As can be seen in Table 1 on the following page, the yearly growth in the deductible exceeded not only increases in health expenditure, but also the Dutch inflation rate and the economic growth.
6
The aim is to investigate whether this policy change affected inequality between socioeconomic groups in terms of health consumption and health status. The research question is “What are
the effects of the introduction of the compulsory deductible in the Dutch health care system between 2008-2016 on inequality in the use of specialized health care and perceived health?”.
The goal of this research is twofold. The first aim is to estimate the effect of changing the no-claim policy to a compulsory deductible on inequality in health and health consumption. This will be done by comparing the years 2007 and 2008 using a regression discontinuity model. The second aim is to estimate the effects of increases in the deductible on inequality by looking at the years 2008-2016 using an ordinary least squares regression
Decreasing the total health expenditure is only one side of the coin. The policy can increase inequality if it affects lower-income groups more negatively compared to higher-incomes. One of the aims of the Dutch government is to reduce inequality and to make health care accessible for everyone (Rijksoverheid, 2017). Estimating the effect of the policy change on the effects for health consumption and health perception can provide relevant information for society and policymakers about the effect of the policy change on inequality.
Added value
Research has been published on the effect of the no-claim policy (Holland, Van Exel, Schut & Brouwer, 2009; Goudriaan et al., 2007) and on the effects of compulsory deductibles in the Netherlands (Van Der Maat & De Jong, 2010). Only one recent study has really explored the effect of the change from the no-claim policy to the compulsory deductibles. Remmerswaal, Boone, Bijlsma & Douven (2019) used a difference-in-difference design to make inferences about the effect of the policy change. People in the Netherlands only start to pay for their health insurance when they reach the age of eighteen. They compared the effect of the no-claim policy
1Source: Remmerswaal et al. (2009, p.5)
2 Source: OECD statistics (indicator: annual inflation) 3 Source: CBS Statline (indicators: BBP and Zorguitgaven) Table 1
Trend in deductible, inflation rate, health expenditure and GDP in the Netherlands 2009-2016
2009 2010 2011 2012 2013 2014 2015 2016 % Increase deductible 1 3,33 6,45 3,03 29,41 59,09 4,29 2,74 2,6
Inflation rate2 1,2 1,3 2,3 2,5 2,5 0,6 0,6 0,3
Increase health expenditure as % of
GDP3 0,3 1,1 0,3 0,5 -0,1 -0,4 0 0
7
kicking in at the age of 18 between 2006-2007 to the effect of the deductible kicking in for eighteen-year-old individuals between 2007-2008, to look if the effect of the different cost-sharing schemes is significantly different. They found that a one-euro increase of the deductible reduces health care spending 18 eurocents more than a euro increase of the no-claim rebate.
Although extensive research has been carried out on the effect of the policy change on health expenditure, the researchers have not looked at the effect on inequality in much detail. This research will add to the scientific knowledge in three ways. First, by focusing specifically on the effect of the policy change on health inequality. This research draws from the empirical strategy by Remmerswaal et al. (2019) to estimate the effect of the policy change for the health consumption for individuals in different income categories. If the effect turns out to be significantly different, this can provide useful information on how policy can affect health inequality. Second, not only the differential effects on health consumption are important, but health in general as well. This research provides a new perspective by also looking at self-reported health. This can be used as an indicator of the health status of an individual. Research has shown that self-reported health is a good and consistent predictor of morbidity and mortality (Idler & Benyamini, 1999). It is interesting to look if the policy change increases the differences between higher and lower-incomes in terms of health status and health consumption. A final contribution is that this research also focusses on the effect of the increasingly higher compulsory deductible over time.
The research uses survey data from the LISS panel. The results are varied. The results from the regression discontinuity model looking at the introduction of the deductible in 2008 do not show a more negative effect for the lower-income group. When the years 2008-2016 are also considered in the OLS regression a negative trend can be seen in subjective health, but not for health consumption. However, based on the literature a further distinction was made that differences in health outcomes are not just related to having a lower-income, but relate more specifically to liquidity constraints. Individuals with a lower-income are more prone to be liquidity constrained. Instead of a potential gain at the end of the calendar year by receiving the no-claim refund, individuals now face an additional cost when they consume health care. When they face financial hardship to pay for the deductible, they might choose to not consume health, which can lower their health in general. The results show some evidence that this is the case. The introduction of the deductible caused a decrease in health and health consumption for liquidity constrained individuals compared to non-constrained individuals and displayed that this difference remained over most of the years between 2008-2016. Yet, all the interaction
8
effects turned out to be insignificant. Implying that the found differences could potentially be caused by chance. The research found no significant evidence for an increase in inequality caused by the introducing of the compulsory deductible.
The research is structured as follows. The next chapter examines the institutional context of the Dutch health insurance. The third chapter gives an overview of the existing literature on health inequality, health care costs and liquidity constraints. The fourth chapter presents the methodology used in this study. The research proceeds by analyzing the findings of the different models in chapter five. The last chapter shows the final conclusions and presents a discussion on the limitations and recommendations for future research.
9
2.
The Dutch health care system
In this chapter, the institutional context of the Dutch health care system is analyzed. First, the legal aspects of the Health Insurance Act are discussed. Second, an overview of the causes for the increase in health care costs in the Netherlands are presented. Finally, this chapter takes a closer look at the two policy schemes: the rebate and the compulsory deductible.
The Dutch health care system is based on the “Health Insurance Act” (Zvw) that is introduced in January 2006. This law abolished the distinction between private and public insurance. Since 2006, there is one basic package of health insurance that is compulsory for every person who lives in the Netherlands and is above the age of 18. In addition to this basic package, people can choose to have additional insurance. Insurance companies are not allowed to adhere to risk-selection and differentiate between premiums. The insurer needs to ask everyone the same premium for the same coverage. People have a free choice between insurance organizations and can switch once a year between insurer and insurance plan if preferred. This provides freedom of choice and improves competition between insurance companies. This increased competition can lower the costs and improve the quality of health care (De Jong, Van den Brink-Muinen, & Groenewegen, 2008, pp.1-3)
2.1 Determinants of expenditure increases
There has been an increase in the Dutch health care costs in the past decades. Van der Horst et al. (2011, pp.4-6) show that the Dutch nominal health costs increased with an average of 6.7% each year in the period 2001-2010. This increase can be explained by both increases in price and volume.
Increases in price can explain 2.3% of the total increase in nominal health costs. This increase in price is caused by multiple factors. First, inflation must be considered in the price of health care. In addition, the quality of health care has gone up. The more advanced treatments are more effective, but these new technologies are also more expensive. Moreover, technological innovations are less effective in increasing the productivity in jobs which are characterized by non-routine human activities. It is therefore difficult to reduce the costs in the health sector due to the high labor intensity. Furthermore, the demand for health is inelastic. When the wages in other parts of the economy increase, the wages in the health sector must increase as well in order to attract and keep people working in the health sector to fulfill demand. This problem is known as ‘Baumol’s disease’ where wages increase more than the labor productivity, making health care relatively more expensive (Van der Horst et al., 2011, p.6).
10
A second component is the increase in volume, which can explain 4.4% of the increase in nominal health costs. Demography can explain 0.7% of the increase in costs. This is linked to an increase in the total population and life expectancy. The aging of the population causes an increase in long-term care costs. Another determinant of health care cost is income, since health is a luxury good. Income can explain 2.0% of the increase in expenditures, due to an increased demand. Van der Horst et al. (2011, pp.4-6) subtracted the increase in demography and income from the total increase in volume and showed that 1.7% of this increase remains unexplained This implies that there are other unknown factors at play as well (Van der Horst et al., 2011, pp.4-6).
2.2 Policy design
The Dutch government introduced a no-claim refund scheme in January 2005. This scheme enabled people to receive up to 255 euros when they used zero to little health care. This applied to all insured persons from the age of 18. The aim of this scheme was to make citizens more aware of their health consumption and reduce moral hazard and health costs (Goudriaan et al, 2017, p.9). Barr (2012, pp.92-94) shows that moral hazard in health care can lead to incentives for overconsumption, because consumers do not face the actual price of health care due to insurance. Governments can introduce incentive mechanisms to share the costs between the individual and the insurer. Goudriaan et al (2017) evaluated the effect of the no-claim scheme using survey data from insured persons, general practitioners and health insurers plus additional information from the Dutch Health Insurance Board. They concluded that while the aim of the no-claim rebate was to reduce health care costs, it had little effect on the consumption of health.
To combat the increasing costs of health care, the Dutch government introduced another scheme in 2008. In the new scheme the rebate was replaced with a compulsory deductible. The Dutch government simultaneously lowered the average premium. In the end, the average insurance costs remained the same for the citizens. Not all-medical care counts for the deductible. The same types of care as with the no-claim policy are excluded from the deductible. The compulsory deductible reduced the total expenditure on health care more than the rebate (Remmerswaal et al., 2019, p.2).
11
According to Remmerswaal et al. (2019, p.2) low-income groups did not lower their health consumption under the no-claim rebate but did respond to the compulsory deductible. The explanation that they gave for this phenomenon relates to the liquidity constraints households face. If people pay a lower monthly premium, but do not save this money in order to be able to pay for possible deductibles in the future, they might face liquidity problems. The situation might occur that they need health care, but they do not have the money to pay for the deductibles. This was no issue in the no-claim scheme, because people already paid this cost in the course of the year through a higher premium. This can lead to the idea that heath care has become more expensive, affecting more the lower-income households who face a tighter budget constraint (Remmerswaal et al., 2019, pp.19-20).
12
3.
Literature review
The focus of this research is on inequality in health. Health inequality is defined by the World Health Organization as “[…] differences in health status or in the distribution of health
determinants between different population groups” (2010a). A lot of different factors affect
people’s health status. Among these determinants of health are the social and economic environment, institutional factors and individual characteristics and behaviors (WHO,2010b). Socioeconomic differences are perceived as one of the main determinants of inequality in health status. This research looks specifically at the difference in health status and health consumption between higher and lower-income groups. The first section elaborates on the general persistence of inequality in health outcomes. The next section covers different theories for reducing health care costs, with an extra emphasis on liquidity constraints and the effect of prices on the consumption of health care. The final part presents the hypotheses.
3.1 Health inequality
The welfare state is established to increase the welfare of people who are poor, weak and vulnerable (Barr, 2012, p.8). Nevertheless, inequalities in health remain a problem despite a lot of effort from government schemes to reduce this inequality. This failure is illustrated by the fact that inequalities in mortality and morbidity between socioeconomic groups did not decrease but increase in the last decades. This trend is the same for European countries with a more generous welfare system (Mackenbach, 2012; Balaj, McNamara, Eikemo, & Bambra, 2017). According to a rapport from the OECD/EU in 2018 (p.12) the average life expectancy has increased from 2001-2011 with two to three years. This rise in life expectancy has slowed down in recent years, but large health inequalities remain persistent. Not only gender creates differences in life expectancy, with women living on average five years longer than men, but also socioeconomic status is a huge determinant of life expectancy. The OECD/EU rapport (2018) stated that “[…] on average across EU-countries, 30-year-old men with a low education
level can expect to live about 8 years less than those with a university degree” (p.81). For
women this difference was only four years. These differences in life-expectancy are similar for gaps in income. Low-income individuals live shorter compared to higher-income individuals. Kalwij, Alessie, & Knoef (2012) looked at health inequalities in the Netherlands, with a focus on the link between individual income and life expectancy at the statutory retirement age. They found that the life expectancy was 2.5 years less for lower-income groups in comparison to high-income groups. Moreover, the OECD/EU rapport (2018) shows that lower-income groups also report 60% of the time to be in good health, while for the highest income group this percentage lies around 80%.
13
According to Balaj et al. (2017) inequalities in health and life expectancy arise from “[…]
inequalities in the social conditions in which people from different social economic backgrounds live and work”. (p.107). There are two dominant explanations for this in the
literature. First, the social causation explanation argues that people in a higher socioeconomic position tend to have more resources, knowledge, support and behavior that are beneficial to health. Nevertheless, it can also be the case that not the socioeconomic position determines health, but that health is a determinant of socioeconomic status. When you have a relative better health, you are more productive and thus can have a higher chance of job opportunities. This is called the health selection perspective (Kröger et al., 2015; Mackenbach, 2012). The cultural explanation also supports the social causation perspective. According to the cultural explanation, health inequalities arise because lower-income groups often have an unhealthier lifestyle. This is also the case for lower educational groups. Cutler & Lleras-Muney (2010) examined the relation between education and health behaviors and found that better-educated people have less chance of being obese, smokers and heavy drinkers. They are more likely to drive safely, use preventive care and live in a safe house. According to Cutler & Lleras-Muney (2010) education influences cogitative ability, which in turn leads to more healthy behavior.
Health inequalities in consumption and health status can be reduced by horizontal equity. Horizontal equity is defined by a minimum standard for certain goods and services, equal access to them and equality of opportunity (Barr, 2012, p.69). Van Doorslaer & Masseria (2004) describe this as equal access for equal need, more specifically: “[…] people in equal need of
care are treated equally, irrespective of characteristics such as income, place of residence, race, etc.” (p.9). However, equality in access is also affected by socioeconomic status and
institutional factors. Lower-income groups are five times more likely to have problems with access to health care compared to higher-income groups (OECD/EU, 2018, p.12). Other determinants of accessibility are affordability, waiting time and distance to the closest health facility. There is evidence that higher-income groups tend to overconsume health care, while lower-income groups facing accessibility issues tend to under consume health care. Problems with access are mostly related to financial reasons. The affordability of health is reduced by high out-of-pocket payments (OECD/EU, 2018 pp.169-170; Van Doorslaer & Masseria, 2004).
14
3.2 Reducing health care costs
To combat the rise in health expenditures, different policies can be proposed to create incentives on either the demand side or the supply side. Supply side incentives focus on restricting access to technologies and services. On the other hand, policies on the demand side try to reduce costs by exposing consumers to the actual costs of their health consumption. There are different mechanisms to do this: premium differentiation (higher premiums for more frequent claimants), deductibles or coinsurance. The effect of these policies depends on how consumers respond to cost-sharing incentives and the price-elasticity of demand (Brot-Goldberg, Chandra, Handel, & Kolstad, 2017).
Brot-Goldberg et al. (2007, p.1261) investigated the effect of the prices of medical care on the consumption of health. They did this by performing a natural experiment at a large self-insured firm. The firm made all its workers switch from a full coverage insured package, to a nonlinear high-deductible plan. It is interesting to see how the introduction of the high-deductible plan affected the health consumption of the employees and the causal mechanisms that underly this. The results show that the switch to the deductible decreased spending between 11.8% and 13.8%. They give three possible explanations for this reduction: price shopping, quantity reductions and quantity substitutions.
When consumers face higher costs, they can choose to go price shopping for cheaper providers. Price shopping depends on multiple factors, namely search effort, consumer provider preferences and information about prices (Brot-Goldberg et al., 2007, p.1265). Other responses are quantity substitutions to lower-cost procedures and quantity reductions. These responses can be an efficient means to reduce costs and improve welfare if the assumption applies that consumers are full-informed and rational. According to Brot-Goldberg et al (2007), this improvement in welfare is attained because “[…] consumers apparently value the forgone care
at less than the total cost” (p.1265). In the case of constraints, lack of information or
non-rationality the result can be a welfare loss, since consumers do not only reduce or substitute wasteful services, but also valuable services. While Brot-Goldberg et al (2007, pp.1288-1298) find an effect of quantity reductions, quantity substitutions and price shopping seem to matter less when it comes to reductions in spending.
15
3.3 Liquidity constraints
The previous section emphasized that the way in which health care policies are structured is important, since these policies can have negative side effects on welfare. As mentioned by Brot-Goldberg et al. (2007), when individuals face constraints in their optimal choice of health care, this can lead to a loss in welfare, since they can also reduce valuable services. One of the constraints individuals face in their choice of consumption are liquidity constraints. Individuals are liquidity constrained when they are not able to finance their present optimal consumption. A person may wish to discount his or her future income in order to finance present consumption. When individuals are unable to borrow the necessary amount of money, they cannot reach their preferred optimal consumption (Beccarini, 2014).
Both higher- and lower-income groups can face liquidity constraints. Liquidity constraints are often used as the explanation of the differences between theoretical optimal and actual behavior (Chah, Ramey, & Starr, 1995, p.272). In the case of health insurance, lower-income groups can face more financial hardship to pay out-of-pocket payments, especially when these are large compared to their income. Poorer people are therefore more prone to be liquidity constrained.
The effect of a deductible differs from a rebate. In the case of the Dutch rebate, an individual pays a higher monthly premium. The insurer directly pays for all the incurred health costs but when an individual does not use health care, he or she receives a rebate in the next year. The deductible works the other way around, the individual pays a lower premium, but firsts needs to pay a predetermined amount of health costs itself before the health insurer covers the costs. While in the Netherlands the average costs remained the same for the rebate and the deductible, the way in which the policies influence households differ. Under the deductible scheme it might happen that an individual needs health care but does not have the discrete amount of money to pay for the deductible. This creates the situation where consumers reduce their health consumption because they face a liquidity constraint (Brot-Goldberg et al. 2007, pp.1313-1314). This will more likely affect lower-income groups. This can result in less access to health care, increased health and socioeconomic inequalities, undermine health in general and increases the level of poverty. There is clear evidence between the level of financial protection and the health inequalities of the lower-income groups (OECD/EU, 2018, p.172)
16
3.4 Hypotheses
This research focusses primarily on the policy change from the rebate to the deductible in the Netherlands. While introducing a compulsory deductible might be a good strategy to reduce the demand side of health care, the research examples show that it can also have negative effects on inequality. One of the reasons for this is the liquidity constraint that lower-income groups face. This can lead to further socioeconomic inequalities in health by reducing the health consumption for lower-income groups disproportionate to higher-income groups. In addition, if the health consumption of lower-income groups decreases relatively more than the health consumption of other income groups because of the policy change, the expectation is that it increases the gap in health between socioeconomic groups even more. From this assumption, the following hypotheses are derived:
H1: the introduction of the compulsory deductible reduces the utilization of health care
relatively more for lower-income groups compared to higher-income groups.
H2: the change from the rebate policy to the deductible reduces the health of
lower-income groups relatively more compared to higher-lower-income groups.
H3: the differences in health and health consumption between the lower and
high-income groups followed by the policy change are caused by liquidity constraints households face.
17
4.
Research design
4.1 Method of data collectionThe research uses survey data from the LISS (Longitudinal Internet Studies for the Social sciences) panel administered by CentERdata (Tilburg University, The Netherlands) to test the hypotheses. This dataset contains 4500 households, which consist of roughly 7000 individuals in the Netherlands. The data collection started in October 2007. The members of the panel fill in online questionnaires every month, with the aim of collecting information on changes in the life course and living conditions of panel members. According to the LISS website, “The panel
is based on a true probability sample of households drawn from the population register by Statistics Netherlands” ("About the Panel", n.d.). The research uses the LISS data on health,
which consists of different waves. The research uses the first nine waves for the data collection. This consists of the years 2007 – 2016. Unfortunately, there is no data collected on health in the year 2014. Consequently, the increase in the deductible from 2013 to 2014 will not be included. While it is unfortunate that there is no data on the year 2014, because less data implies less precision, this is not insurmountable.
A drawback is that the LISS panel started in October 2007. Only information for the months November and December are included in the dataset. Consequently, the year 2007 has less information and datapoints compared to the other years. The year 2007 is an important year, since it is the only year in the dataset in which the no-claim policy was active. This can create issues in comparability to the other years of interest, especially in the comparison between 2007 and 2008 with regards to the policy change. However, the questions in the health survey look back to the whole year under investigation. The fact that data collection started in 2007 should not be a problem because the health survey asks questions regarding the whole year 2007, not only for the months November and December. Nevertheless, comparability issues remain. This should be kept in mind when looking at the results. A different option would have been to use another comparable panel dataset like the DNB Household Survey (DHS), that collects data since 1993. The reason that this research favors the LISS dataset relates to the relevant questions in the surveys. The LISS panel contains information on both health perception and health consumption, while the DHS data only collected information on health perception
Next to the data on health, the research uses different datasets on the economic situation and background characteristics of the panel members in order to gather information about age, income and other characteristics of interest. Another difficulty is that the questions on health and income are based on yearly data, while the background questions are based on monthly
18
information. To be able to merge the data accordingly, the research either uses an average for the value of background variables that can differ monthly or generates a new yearly variable by adding the values of the monthly variables together. This will be further explained in section 4.3 that deals with the operationalization of the variables.
In the process of merging the background variables with the health variables 50.52% of the data is not matched. The explanation for this relates to the decrease in the number of panel members during the years under observation and to the amount of non-response in the health surveys. The health sample consists of panel members aged 16 years and older who are send questionnaires and need to fill this in individually.
Table 2 gives an overview of the amount of selected household members and response for each year for the health waves. As can be deducted from Table 2, the number of household members participating decreased approximately 25% between 2007 and 2016. In combination with the non-response, the total respondents in each sample decreased. Further, because of issues with non-respondence four refreshment samples are conducted for the LISS Panel4. Especially the
years 2008 and 2009 have a very high rate of non-response. The response rate improved in 2010 after the first refreshment sample. This gives a possible explanation for the large amount of unmatched observations, because new panel members were added, while some other panel members were excluded due to a lack of response. For the background variables non-response is not as big of an issue, because only the contact person of the household needs to fill in the monthly changes (LISS Panel Data Archive).
(Source: LISS Panel Data Archive)
4 The refreshment samples were carried out between June - December 2009, October 2011 - May 2012, November
2013 - June 2014 and between November 2016 - June 2017. This was done using a stratified sample, with the intention to improve the representativeness of the panel for certain groups. Three variables were used for this stratification: household type, age and ethnicity ("Sample and recruitment”, n.d.)
Table 2
Sample response and non-response Health Waves
Year 2007 2008 2009 2010 2011 2012 2013 2015 2016 Number of household members 8478 8280 9170 7364 6533 6769 6217 7126 6336 Response in percentages 78,9 72 66,7 77,6 77,6 85,5 86,5 84,3 85,4 Non-response in percentages 21,1 28 33,3 22,4 22,4 14,6 13,5 15,7 14,6 Total respondents 6698 5961 6119 5718 5072 5780 5379 6009 5408
19
The uniqueness of the responses of the panel members are checked and verified that the number of the panel member encrypted can be uniquely matched to a specific wave and response. While the amount of unmatched observations gives room for concern, this can be partly explained by the decrease in respondents over the years and the refreshment samples. The remaining observations provide a large enough sample and are used in the analysis.
4.2 Case selection
The research uses micro data on individuals from the LISS panel to infer an effect of the introduction of the deductible on health and health perception. Two different groups will be used for the different analyses. First, for the regression discontinuity analysis, the sample exists of individuals from the age of 18 in the years 2007 and 2008. Only people who are 18 years or older are affected by the policies. Second, the sample for OLS regression using time dummies includes individuals aged ≥ 18 in 2008. People below the age of 18 in 2008 are not affected by the deductible. Because the aim is to establish an effect of the effect of the increasing deductibles, only people who are above the age of 18 in the reference year are selected. The sample also includes the years 2008-2016.
Some respondents are excluded from the sample. All individuals who use a voluntary deductible are excluded from this research, but this is only a small portion of the sample. In 2007 only five percent of the total insured had a voluntary deductible (Goudriaan et al., 2007, pp.26-27). The focus on the research is on the effect of the introducing of the deductible on equality. Like Remmerswaal et al (2019, p.8) also mention, including people who have a voluntary deductible can compromise the results by effects of moral hazard and adverse selection. Further, when there are missing values for a certain respondent on the variables of interest, this respondent is excluded from the research.
20
4.3 Operationalization of variables
Information from the LISS panel database is used to create the variables in this research. Appendix A provides an overview of all the variables used.
4.3.1 Independent variables
The phenomenon of interest in the first analysis is the introduction of the deductible – and how this has affected the lower-income groups in comparison to higher-income groups. Figure 2 represents the increase in deductibles over the past 9 years. This increase will be used to infer what the effect is of the deductible on health perception and health consumption. The average inflation in the Netherlands between 2007-2016 was 1,61% (OECD, n.d.), thus the increase in the deductible exceeds this inflation rate.
A categorical variable income is used as a measure of the household income. The LISS dataset contains information on the monthly net income of all household members combined. In addition to this, the income variable is corrected for the number of members living in a household by dividing the household income by the square root of the number of household members. A small inconvenience was the fact that there was no data on the household income for the year 2007. For the months November and December in 2007 this is calculated manually by taking the sum of the net incomes from the background dataset for the individuals in one household. The Dutch Central Bureau for Statistics (CBS, 2018) uses five different income categories based on yearly income5. This research merges these categories into three categories, to make a clearer distinction between low, middle and high-income. Since the LISS data on
5 The income categories used by the CBS in 2018 are low < €1000; low-middle €1001-€2000;
middle €2000-€3000; high-middle €3001-€4000; and high >€4001.
€150 155 €165 €170 €220
€350 €365 €375 €385
2008 2009 2010 2011 2012 2013 2014 2015 2016
Deductibles 2008-2016
21
income uses the monthly net household income in euros, the categories are altered to monthly income to suit the data. The different categories can be seen in Table 3
Table 3 Income categories Category 1 Low-income 2 Middle-income 3 High income
Household Income in euros 0 - 1500 1501– 3000 >3001
There are a few outliers in the data. There are 37 respondents with a monthly net household income that is higher than 17000 euros. These 37 respondents are not included in the sample. The maximum household income in the sample after excluding these outliers is 11384,42 euros per month. As can be seen in Table 4, the largest part of the sample consists of individuals in the middle-income group. The mean income of the whole sample is 1857,51 euros, this is close to the mean of the middle-income group. However, the low-income group receives around 30% less than the mean income, while the highest income group receives more than 50% above the mean income. The total net household income is not evenly distributed among the different income groups, this is in line with the theories around the persistence of income inequality
Table 4
Summary statistics of income by group
Income categories
N % Freq. Mean Sd P25 P50 P75 Min Max
Low 173 651 37,12 1106,36 314,90 939,15 1183,50 1347,01 0 1500
Middle 256 732 54,87 2093,50 400,42 1750 2020,76 2400 1500,40 3000
High 37 476 8,01 3721,35 884,67 3181,98 3432,81 3889,09 3001,07 11384,42
22
In addition to the three-income categories, a dummy variable liquidity constraint (LC) will be included. The operationalization of liquidity constraint draws from the research of French (2018) who explored three ways in which financial strain could occur. One of these is that
“permanent income is insufficient to meet basic needs, putting the household under financial strain” (French, p.166, 2018). He showed that the subjective financial well-being is what
matters most for the decision-making of individuals. For this reason, respondents who have answered ‘yes’ to the question of having trouble making ends meet are considered liquidity constraint (1=yes; 0=no). Table 5 gives an overview of the individuals in the sample who are considered liquidity constraint separated in the different income categories. Individuals from all income categories can potentially be liquidity constraint, but in line with the expectation, this mostly seems to occur in the lower-income groups.
Table 5
Liquidity constrained individuals per income group
Income categories
Low Middle High Total
Liquidity constrained respondents 29 768 13 241 560 43 569
LC in % 68.32 30.39 1.29 100
23
4.3.2 Dependent variables
To measure the general health, the concept of subjective health is used. This concept measures a person’s perception of his or her health by asking the question “How would you describe your
health, generally speaking?”. The answers are put into categories, which are listed in Table 6. Table 6
Categories of Subjective Health
Category 1 2 3 4 5
Perception Poor Moderate Good Very good Excellent
Figure 3 shows the distribution of health perception in the different income categories. For all the three income categories most respondents answered to be in good health. Interesting is that as health increases, the representation of the highest income group increases as well. For the lower-income group this is the other way around: in the poorest health category they are most represented, as health increases, their representation decreases. This supports the literature that higher-income groups have on average better health. In Appendix B a visual representation is given for the trends over the years.
Figure 3 Distribution of health by income group
2,16 18,88 58,67 16,15 4,14 0,98 13,9 60,3 19,78 5,04 0,68 10,72 55,56 25,97 7,07 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Poor Moderate Good Very good Excellent
Distribution of Health
24
Visits to the general practitioner are excluded from both the no-claim and the deductible. Because of this, a variable measuring the visits to the GP will not be a good reflection of the effects of the policy change. The expectation is that it will affect the use of specialized care, because in the new policy specialized care is only covered after people pay the compulsory deductible. To be able to estimate the difference in health utilization before and after the policy change, one needs to measure the use of specialized health care. First, the variable medical
specialist is used, which represents the amount of times a respondent visited a medical specialist
at a hospital per year, with a possible range from 1 to 999. The variable specialist is also included, to look at the effects of the policy change on the decision to retain from using specialist care. This is measured by a categorical variable, measuring different types of specialist care. The category of interest is category 11, which represents ‘no specialist use’, that forms the zero in the medical specialist variable. These two variables are put together measuring
health consumption.
Most observations of the variable health consumption are in the lower frequency zone. As can be seen in Figure 4, 60.95% of the individuals in the sample do not visit any medical specialist, 13.88% only visited once and 9.37% twice. Beyond this, the frequencies are quite low, Moreover, as can be seen in Table 7, the middle-income group is the highest consumer of health. Against the expectations, the high-income group seems to consume the least health care. The percentages in Figure 4 and Table 7 differ slightly because Table 7 does not include more than five hospital visits per year.
Table 7 Total health consumption by income group Visits hospital Low income Middle-income High-income Total 0 24,02 35,3 5,08 64,4 1 4,99 8,37 1,29 14,66 2 3,42 5,55 0,93 9,9 3 1,95 2,71 0,37 5,03 4 1,41 2,02 0,27 3,7 5 0,95 1,22 0,15 2,32 Total 36,75 55,17 8,09 100 60,95 13,88 9,37 4,76 3,5 2,19 5,35
Total visits to a medical specialist
0 1 2 3 4 5 > 5
Figure 4 Total health consumption per year
Figure 5 Distribution of ageFigure 6 Total health consumption
25
Table 8 gives an overview of some summary statistics for the dependent variables per year separated by liquidity constrained status. The mean of the variable subjective health does not vary much between the years for the different groups. Table 8 shows that individuals who are liquidity constrained do have on average a lower health compared to those who are not liquidity constrained. This is in line with the expectations. However, this difference is very small. In addition, a small decrease in health can be seen for both groups over the years but this decrease is more pronounced for the liquidity constrained group. The standard deviation for subjective health varies for both groups between 0.75 and 0.84 over the years.
In the case of health consumption, the standard deviation does show more fluctuations over the years and between the two groups. Moreover, the standard deviations are also quite high compared to the means. The reason for these high standard deviations is unknown, but it can be related to the fact that if an individual is sick or has a disease, their visits to a medical specialist will increase highly, while healthy individuals will often not visit a medical specialist. This creates big differences between the sick and the healthy, resulting in high standard deviation. This is also supported by the big differences between the minimum and the maximum visits to a hospital, which also differ a lot over the years. Interesting is that both groups show a large drop in their average health consumption in 2008 compared to 2007. The deductible was implemented in the year 2008. This big drop in health consumption hints that there was an effect of the policy change. The health consumption also remained at this lower level in the following years. Contrary to the expectation, Table 8 shows that the average health consumption for the liquidity constrained group is higher in comparison to the non-constrained group. However, the large standard deviation should be kept in mind. This difference might turn out to not be significant. This will be investigated further in the result section.
Table 8
Summary statistics subjective health and health consumption by liquidity constraint status
Variables 2007 2008 2009 2010 2011 2012 2013 2015 2016 Subjective
health
Liquidity constrained Mean 2.91 2.87 2.75 2.73 2.75 2.79 2.81 2.68 2.66
Non-Liquidity constrained Mean 3.18 3.21 3.15 3.10 3.08 3.11 3.08 3.09 3.09
Health consumption
Liquidity constrained Mean 4.0 1.97 3.02 2.01 1.86 1.92 1.81 1.93 2.32
Std. Dev 5.18 4.54 11.65 4.28 4.01 4.74 3.81 3.84 8.03
Min 1 0 0 0 0 0 0 0 0
Max 57 52 200 50 30 55 25 40 130
Non-Liquidity constrained Mean 3.03 1.11 1.23 1.35 1.28 1.22 1.22 1.33 1.30
Std. Dev 3.74 2.92 2.59 3.8 2.95 2.75 2.67 8.97 3.48
Min 1 0 0 0 0 0 0 0 0
26
4.3.3 Control variables
It is important to control for variables that could potentially influence both the health outcomes and the increase in the deductible to prevent confounding variable bias. Inflation and economic growth can influence the purchasing power of individuals, which can influence health care consumption. This can increase the government expenditures on health as well, which can affect the price of the deductible, since the price of the deductible is coupled to the health expenditures of the Dutch government. Time dummies will be included to control for these aggregate effects. These dummies capture all effects that affect everyone at the same time. In addition, there are a lot of differences between the three income groups and between liquidity and non-liquidity constrained individuals unrelated to the increase in the deductible, that can possibly affect health outcomes. Multiple control variables are included in the regressions to control for these differences, to be able to estimate on the effect of the deductible more in isolation. The lowest income category is correlated with being liquidity constrained. 68.32% of the liquidity constrained group belongs to the low-income category. The estimated Pierson correlation coefficient is 0.2450, which indicates a weak positive correlation. While this correlation is not strong, to prevent repetition this section primarily focusses on the distribution of the control variables by the different income categories. Appendix C provides additional information on general summary statistics by liquidity constrained status.
First, the variable age is included. Research shows that when people are becoming older their health decreases, which results in more health care consumption and higher expenditures (Wouterse, Huisman, Meijboom, Deeg, & Polder, 2013). Figure 5 shows that there are relatively more individuals who are older in the sample. From the six age categories, 38% are in the three lower age categories, while 62% are in the three highest age categories. Moreover,
11% 11% 16% 18% 21% 23% Distribution of age 14 - 24 25 - 34 35 - 44 45 - 54 55 - 64 ≥ 65 0 5 10 15 20 25 30 14 - 24 25 - 34 35 - 44 45 - 54 55 - 64 ≥ 65 Age separated by income
Low income Middle income High income
Figure 6 Distribution of age by income category
Figure 43 Income categories by genderFigure 44
Distribution of age by income category
Figure 5 Distribution of age
Figure 97 Distribution of age by income
27
age is also not evenly distributed among the different income categories. Figure 6 shows that individuals in the age categories 14-24 and 35-44 more often have a low income. Contrary, individuals between 55-64 more often have a high-income. It is thus important to control for age. Since it can influence the health outcomes and differs by income position.
Previous literature also shows that marital status influences health. According to Lillard & Panis (1996, p.313) married individuals have lower mortality rates compared to unmarried individuals. Both men and women show to have a significant reduction in mortality rate, but this reduction is higher for men. A dummy variable named married is included (0= unmarried, 1=married) to control for this influence of marital status. 56.91% of the individuals in the sample are married. There is a slight difference between income groups. In the low-income group 52.58% is married, while this is 58.98% for the middle-income group and 62.75% for the high-income group.
As mentioned before, gender can play a role in health differences. Women live on average five year longer in comparison to men (OECD/WHO, 2018). To control for these gender differences, a dummy variable male is included (1=male; 0=female). 53.17% of the individuals in the panel are female and 46.83% male. This equal division between male and female respondents changes a little bit when the sample is divided by income. Figure 7 shows that individuals who are male more often have a higher-income, while female respondents are more often in the low-income category. Gender thus does not only create differences in health, but there is also a gender differences related to the dependent variable income.
Figure 7 Income categories by gender
Figure 151 Total unemployment per income categoryFigure 152 Income
categories by gender
Figure 153 Total unemployment per income categoryFigure 154 Income
categories by gender 0 10 20 30 40 50 60
Low-income Middle-income High-income
Income distibution and gender
28
Further, 31.43% of the individuals in the panel suffer from any kind of long-standing disease, affliction, handicap or consequences from an accident. Research shows that people with a handicap generally use more health care (WHO, 2018). The variable handicap will be included to control for this influence (1=yes; 0=no). The variable handicap is equally represented at each income category.
In addition, the fourth category of the variable ‘belbezig’ is used, which represents the occupation ‘job seekers following job loss’. The variable ‘unemployed’ is included that equals one for job seekers. This variable will be used to control for economic effects that could possibly affect lower-income groups more than the higher-income groups. Lower-income groups more often have lower-skilled jobs with higher job insecurity (Walter, 2010). Only 2.41% of the sample is unemployed. As Figure 8 shows, most individuals who are unemployed are either in the low-income or middle-income category. Only 3.74% of the unemployed individuals are in the highest income category. The income categories are based on household income, which can explain that individuals in the higher-income categories can still be unemployed. In addition, these numbers do not give information about the duration of the unemployment and possible unemployment benefits, which can reduce the effects of unemployment on income.
Next, a continuous variable is used to control for the influence of the healthcare allowance in the Netherlands called the “zorgtoeslag”. People below a certain income receive this benefit. The amount of money a person receives is income dependent; as income increases, the health care allowance reduces. The healthcare allowance can influence the costs of health insurance for poorer people (Belastingdienst, 2019). The variable zorgtoeslag will control for this influence. There are some outliers in the data. The maximum amount a person could receive in health benefit in the year 2016 is €1.905. This will be taken as the maximum possible amount,
55,56 40,7
3,74
Total unemployment by income category
Low-income Middle-income High-income
Figure 8 Total unemployment per income category
Figure 205 Levels of education per income categoryFigure 206 Total
unemployment per income category
Figure 207 Levels of education per income categoryFigure 208 Total
unemployment per income category
Figure 209 Levels of education per income categoryFigure 210 Total
unemployment per income category
29
responses above this number are not included. The original data is in years, this is divided by twelve to show the monthly received health benefit. 11.81% of the individuals in the sample receive this benefit.
Finally, higher-educated people often have more healthy behavior (Cutler & Lleras-Muney, 2010). For this reason, a categorical variable measuring the level of education is included in the sample. Table 9 shows the education levels, which are based on the Dutch education system.
The level of education also shows differences per income category. Individuals in the lowest income categories are more highly represented in the lowest two education levels compared to the other income categories. On the other hand, the two highest education levels have a higher representation of individuals in the highest income category. Education does not only influence health outcomes, but also income potential. Higher education often results in a higher-income, this is also called the education premium (Weistanner, 2018).
Table 9
Distribution levels of education
Category Level of education Frequency Percentage Cum.
1 Primary school 45 569 9,76 9,76
2 Vmbo
Intermediate secondary education, 115 139 24,66 34,43
3 Havo/vwo
Higher secondary education/ preparatory university education 51 011 10,93 45,35
4 Mbo
Intermediate vocational education, 106 897 22,9 68,25
5 Hbo
Higher vocational education 106 789 22,88 91,13
6 Wo University 41 415 8,87 100 0 5 10 15 20 25 30 35 40 Primary school
Vmbo Havo/vwo Mbo Hbo Wo
Level of education
Low income Middle income High income
Figure 9 Levels of education per income category
Figure 259 Trend Subjective Health low, middle- and high-income groupFigure 260 Levels of education per
income category
30
4.4 Method of analysis
This research has a deductive approach and follows a quantitative data analysis. The research is based on a Large-N, time-series comparative analysis. The introduction of the deductible is studied by comparing the years 2007 and 2008. An additional analysis will look at the effects of the increase in deductibles over the last nine years, from 2008 to 2016. The research draws from the empirical strategy of Remmerswaal et al (2019). They apply a difference-in-difference design to make inferences about the effect of the policy change. There is a clear treatment group, because the cost sharing in the Dutch healthcare system only applies to persons from the age of 18. They use two difference-in-difference models. The first model looks at the effect of turning 18 under the rebate policy by looking at the years 2006 and 2007. The second models look at the effect of turning 18 under the deductible using the years 2007 and 2008. The treatment group in both models are the individuals who are 17 in the first year but turn 18 in the following year under investigation. The control group are the individuals below the age of 18 in both years. They compared the output of the two models to establish whether the different cost-sharing policies provide significantly different results.
While this is an interesting approach, due to data constraints it is not practically feasible because the dataset does not contain information on the year 2006. For this reason, this research will use a regression discontinuity (RD) model to look at the effect of the introduction of the deductible. The aim is to establish the effect of treatment at the cutoff, which is the introduction of the deductible in 2008. This will be done by comparing individuals of the same age under the rebate in 2007 to individuals under the new policy in 2008. There are two groups of interest. The first RD focusses on the differences for liquidity constrained individuals and non-constrained individuals, while the second RD looks at differences between the three income groups at the cutoff. The first regression equation looks as follows:
𝑌𝑖𝑡 = 𝛼 + 𝛽2008𝑡+ 𝜗 𝐿𝐶𝑖𝑡+ 𝛿𝑡(2008𝑡∗ 𝐿𝐶𝑖𝑡)) + ζ𝑋𝑖𝑡 + 𝜀𝑖𝑡
The dependent variable of interest (Y) is either subjective health or health consumption. i indicates the individual member of the LISS panel; t indicates the time period. The alpha α represents the constant term. The 2008 variable represents the running variable. In the case of the introduction of the deductible, time is the running variable since this determines treatment. The variable 2008 is a dummy variable that equals to one in 2008. This is the moment the new policy is implemented and the treatment starts. The treatment effect is the difference in health consumption and subjective health under the old policy in 2007, compared to the new policy in
31
2008. This is measured by 𝛽 for the non-liquidity constrained group. 𝛿 measures the interaction effect, the difference for the liquidity constrained group compared to the non-constrained group. The effect of 𝛽 and 𝛿 are added to create the treatment effect for the liquidity constrained group. The interaction effect shows if the difference between the two groups is significant. The regression compares different individuals of the same age in 2007 and 2008 to test this. The underlying idea is that individuals of the same age in 2007 and 2008 do not differ very much in characteristics. The only difference is whether the rebate policy is active or the deductible. The variable liquidity constraint (LC) is the dummy variable for the category of individuals who face difficulties making ends meet. Parameter zeta ζ measures the impact of the control variables on the outcome, ε measures the regression residuals, the unexplained parts.
The regression equation stays mostly the same for the RD analysis on the effect for the different income groups. The only difference is that the LC variable is replaced by the variable income, a dummy variable that measures the income position (low, middle or high) for an individual in the sample. The equation then will look as follows:
𝑌𝑖𝑡 = 𝛼 + 𝛽2008𝑡+ 𝜗 𝐼𝑛𝑐𝑜𝑚𝑒𝑖𝑡+ 𝛿𝑡(2008𝑡∗ 𝐼𝑛𝑐𝑜𝑚𝑒𝑖𝑡)) + ζ𝑋𝑖𝑡 + 𝜀𝑖𝑡
The bandwidth of the RD can be expanded by including more years in the analysis. Increasing the bandwidth has the benefit of more datapoints, which increases the precision of the estimates. It would be interesting to see whether the results would differ when the window around the cutoff is expanded by including more years in the bandwidth. However, in the following years the deductible has increased in price. By including more years after 2008, the risk is that not the effect of the introduction is estimated, but also the effect of the increase in the deductible. This would create biased estimates. The bandwidth can also not be expanded in the years before the 2007, since the dataset does not contain information before this period. For this reason, only one bandwidth is used, by only looking at 2007 and 2008 in isolation. The analysis on the increase in the deductible will take the other years into consideration.
The effect of the increase of the deductible between 2008 and 2016 on health consumption and subjective health is investigated by performing an Ordinary Least Squares regression (OLS) including year dummies. The first section looks again at the two groups separated by liquidity constrained status with the following regression equation:
