• No results found

The effect of compulsory and voluntary deductibles on health care demand in the Netherlands

N/A
N/A
Protected

Academic year: 2021

Share "The effect of compulsory and voluntary deductibles on health care demand in the Netherlands"

Copied!
79
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The Effect of Compulsory and Voluntary Deductibles on

Health Care Demand in the Netherlands

Abstract – This study analyzes the effect of the yearly increase of the compulsory deductible and the choice for a voluntary deductible on the number of visits to Medical Specialists at a hospital, using data from 2007 until 2016. In 2006, the Dutch government implemented a health care insurance system reformation in an attempt to enhance public awareness about the costs involved in health care and encourage rational use of health care, thereby reducing the effects of adverse selection and moral hazard. In this empirical investigation, the Arellano-Bond difference GMM estimator and the Blundell-Bond system GMM estimator are employed, which take possible endogeneity and predeterminedness of regressors into account. The model is dealing with an identification problem regarding the effect of the compulsory deductible on health care demand. No significant effect of the choice for a voluntary deductible on the number of visits to Medical Specialists at a hospital can be found. This is remarkable, since the reversed adverse selection phenomenon predicts that individuals who opt for a voluntary deductible will make less use of medical services. However, the conclusions drawn in this research should be interpreted with caution, since the results are heavily influenced by choices made by the investigator.

Keywords: health care demand, doctor visits, compulsory deductible, voluntary deductible, moral hazard, adverse selection, panel data, autoregressive distributed lag model, GMM methods, Arellano-Bond estimator, Blundell-Bond estimator.

Thesis MSc Econometrics

Author:

Marlou Beringer (10439048) Date of final version: 15-01-2018

Supervisor:

Prof. dr. J.F. Kiviet Second reader: Dr. M.J.G. Bun

(2)

0

Statement of Originality

This document is written by Marlou Beringer who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no other sources other than those mentioned in the text and its references have been used in creating it. The faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1

Preface

I would like to thank my supervisor Professor Jan Kiviet for his time, supportive feedback and patience during the process of writing this thesis. He has not only assisted me in making decisions on essential econometric issues, but has also helped me with small details and fine-tuning this paper. I appreciate the fact that he supported me during the entire process by giving me the time and space I needed, but at times also pushing me to the limit. Moreover, I respect the fact that he has kept me motivated and positive throughout the entire process. He has always been a pleasure to work with and I feel privileged to have had the opportunity to be supervised by him.

(4)

2

Table of Contents

1. Introduction ... 4

2. Economic Considerations and Previous Research ... 8

2.1 The Dutch health care system ... 8

2.2 Previous research ... 11

2.2.1 Moral hazard and adverse selection ... 11

2.2.2 Research on Dutch data after the health care reformation ... 13

2.2.3 Choice for voluntary deductible ... 15

2.2.4 Determinants for the number of doctor visits ... 16

2.3 Approach of this study ... 18

3. Data ... 20

3.1 Construction of the dataset ... 21

3.2 Dependent variable ... 23

3.3 Explanatory variables ... 23

3.4 Summary statistics ... 25

4. Methodology ... 27

4.1 Panel data ... 27

4.2 Suspected endogeneity and predeterminedness ... 28

4.3 GMM techniques ... 29

4.4 Autoregressive Distributed Lag model... 32

4.5 Specification tests ... 33

4.6 Time-specific effects ... 34

4.7 Estimated model ... 35

4.8 Constructing the set of instruments ... 37

4.9 Coefficient interpretation ... 39

5. Results and Analysis ... 42

5.1 Main results ... 42

5.1.1 Arellano-Bond difference GMM estimator ... 43

5.1.2 Blundell-Bond system GMM estimator ... 46

5.2 Results for subset 2007-2013 ... 49

6. Summary and Conclusions ... 51

(5)

3

References ... 55

Appendix A. Initial set of variables ... 60

Appendix B. Constructing the final set of explanatory variables ... 62

Appendix C. General structures of default and collapsed instrument matrices ... 65

Tables and Figures... 66

Table 1. Monthly premium quotes for the basic package of ten insurance companies in 2018 ... 66

Table 2. Allocation of individuals across different heights of voluntary deductible in the Netherlands ... 66

Table 3. Number of individuals in the final sample opting for voluntary deductibles ... 66

Table 4. Month of data collection and number of participants in each wave of the Health Core Study ... 67

Table 5. Number of manually generated values per variable ... 67

Table 6. Representativeness of the initial and final sample ... 68

Table 7. Descriptive statistics of Medical Specialist visits ... 68

Table 8. Descriptive statistics of General Practitioner visits ... 68

Table 9. Descriptive statistics of explanatory variables ... 69

Table 10. Mean health care utilization for some subsamples ... 70

Table 11. Mean values for some variables conditional on chosen voluntary deductible ... 71

Table 12. Classification of variables ... 71

Table 13. Estimation results for robust AB1 and corrected AB2 and BB2 GMM estimators ... 72

Table 14. Results specification tests ... 74

Table 15. Classification of variables for subset 2007-2013 ... 74

Table 16. Estimation results for corrected AB2 GMM estimator for subset 2007-2013 ... 75

Figure 1. Health care expenditures as share of GDP for various countries in 2008 and 2015. 76 Figure 2. Total health care expenditures as share of GDP in the Netherlands ... 76

Figure 3. Height of the Dutch compulsory deductible from 2008 to 2018 ... 77

(6)

4

1. Introduction

Western countries have been dealing with continually rising health care expenditures in the past decades. In recent years, the annual growth in health care expenditures in the European Union has been greater than the economic growth, resulting in a rising share of GDP allocated to health. In EU countries, the average share of GDP devoted to health care increased from 8.3% in 2008 to 9.9% in 2015 (OECD, 2016). Figure 1 exhibits health care expenditures as a percentage of GDP for different western countries. From this figure it can be derived that in most countries the health care expenditures as a share of GDP increased substantially in recent years. In the Netherlands this increase was relatively high compared to other countries. The percentage of GDP devoted to health care in the Netherlands increased from 9.1% in 2008 to 10.8% in 2015.

The Dutch Bureau for Economic Policy Analysis (Centraal Planbureau) indicates that there has been a long-term upward trend in the health care expenditures as a share of GDP since 1972, however, since the beginning of this century the health care costs increased considerably further in comparison to GDP (Rapport Taskforce Beheersing Zorguitgaven, 2012). This is illustrated in Figure 2. The main reason for the rising health care expenditures is the growing consumption of health care. This growth in health care consumption can be attributed to three specific developments. First of all, the ongoing development of technological innovation in medical care contributes to the growth in expenditures. Okunade and Murthly (2002) confirm that there is a significant and stable long-run relationship between per capita health care expenditure, per capita real income and broad-based Research and Development expenditures. Secondly, developed countries have experienced a significant change in age group composition of their population over the last decades. The relative proportion of people in the geriatric age groups has steadily increased, especially the group of people over 80 years old. This phenomenon contributes to the growth of health care consumption in the way that elderly in general consume more health than younger people. The age group composition is expected to shift even more to the elderly over the next decades as a consequence of diminishing birth rates and increasing life expectancy (Zweifel et al., 1999). In EU countries, for example, life expectancy at birth reached 80.9 years on average in 2014, which was an increase of about seven years since 1990. In the Netherlands life expectancy at birth had even reached 81.8 years in 2014 (OECD, 2016). The third development that contributes to the growing health care consumption is the existence of large-scale health insurance schemes which result in a segregation between the amount of health care consumed by agents and the actual costs (Chiappori et al., 1998). The reason for this segregation is twofold. (i) People have an incentive to demand more health than actually needed in case they no longer bear the costs for this health care consumption. The consumption of unnecessary health care due to insurance is a result of moral hazard (Van Ophem and Berkhout, 2012). (ii) According to Cardon and Hendel (2001), the insurance market is likely to suffer from information asymmetry between insurance companies and the insured. Individuals are presumed to know more about their own risk profile than insurers (Rothschild and Stiglitz, 1976). Less healthy individuals tend to purchase more insurance than healthier ones and insurance companies are not able to correct for this asymmetric information in the premium of the insurance schemes. In general, less healthy

(7)

5

individuals consume more medical care, which is relatively less costly for them than for healthier ones. This phenomenon is known as adverse selection (Frank et al., 2000; Resende and Zeidan, 2010).

The rapidly rising health care expenditures are a major concern for both the government and citizens of concerning countries. For that reason, governments are constantly trying to diminish health care expenditures by adjusting the health insurance system. The first two suggested developments that contribute to the growth in health care consumption are beyond the control of the government, however, the third development is not. Therefore, this development offers the best opportunities to reduce the extensive growth in medical expenditures, especially in countries with publicly provided or financed health care systems. Bago d’Uva and Jones (2009) provide a broad overview of various methods European governments have used to diminish health care costs. These methods are often based on addressing the effects of moral hazard and adverse selection by means of law and regulations. Another popular method which is not mentioned in their paper is the introduction of deductibles (Holly et al., 1998; Van Kleef et al., 2009). Deductibles enforce public awareness about the use of health care and encourage the rational use of medical care. In addition, deductibles provide an incentive for individuals to reduce unnecessary health care consumption, as part of the risk and financial burden is shifted from the insurers to the insured (Van Kleef et al., 2009; Gechert, 2010).

In 2006, the Dutch government implemented a health care reform that changed the health care system from a publicly provided structure to a more libertarian structure (Maarse and Ter Meulen, 2006). In the first two years a no-claim of €255 was obligatory for all Dutch residents, after which a compulsory deductible was introduced in 2008. This implies that each year a certain amount of health care costs have to be paid by the consumer him or herself, before the costs will be reimbursed by health insurance companies. Besides making consumers more aware of the costs concerned with health, another purpose of introducing the compulsory deductible was to decrease the premium paid by the insured. On top of the compulsory deductible, individuals have the possibility to opt for a voluntary deductible in return for an even lower premium. This is one of the main reasons for clients of insurance companies to choose for an additional voluntary deductible, along with the argument that they expect to not make use of health care a lot (Van Der Maat & De Jong, 2010). Before 2006, the focus of the Dutch health care system was on the supply of medical services. However, with the reformation the focus shifted towards the demand (Stroosnier and Van Ophem, 2014).

The new health care system and the introduction of the compulsory deductible in specific also came along with some unfavorable side effects. In 2016, the Dutch survey agency TNS NIPO reported that in that year at least ten percent of the Dutch citizens avoided or postponed medical care because of financial reasons (Bos, 2016). This percentage has been increasing every year since the introduction of the compulsory deductible. Besides, the National Association of General Practitioners (Landelijke Vereniging van Huisartsen) reported in 2013 that General Practitioners more often than before experience the fact that patients ignore a referral to see a Medical Specialist because of the costs involved (Joosten, 2015). In the same year, two percent of the

(8)

6

Dutch citizens acknowledged to have renounced to make use of medical care because of financial reasons, even though they presumably severely needed it. This fact followed from a research conducted by Edith Schippers, the Minister of Health of the Netherlands. This study also indicated that the deductible leads to a lot of misconceptions under the Dutch insured. Nineteen percent assumed incorrectly that the costs for General Practitioner visits have to be paid from the compulsory deductible. On the other hand, 41 percent assumed incorrectly that costs arising from a visit to a Medical Specialist are covered by the basic insurance and therefore do not have to be paid from the deductible (NRC, 2013).

Since the introduction of the compulsory deductible in 2008, the height of this deductible increased every year. Figure 3 exhibits the increase of the compulsory deductible from 2008 to 2018. From this figure it can be derived that in 2018 the height of the compulsory deductible is about 2.5 times higher than at the time of its introduction. This study addresses the question whether the yearly increase of the compulsory deductible has had an effect on the way individuals make use of health care services. In her study, Vluggen (2015) analyzed the effect of the yearly increase of the compulsory deductible on the number of visits to a General Practitioner. Her main argument to model the number of General Practitioner visits and not the number of Medical Specialist visits is the fact that visiting a General Practitioner is an independent choice of the individual, while the decision to see a Medical Specialist has to be made in accordance with the General Practitioner. She concluded that there is no significant effect of the compulsory deductible on the number of yearly General Practitioner visits. This is a remarkable result, since the phenomenon of moral hazard predicts a reduction in health care demand when the financial burden increases. A possible explanation for this notable result is that the General Practitioner is excluded from the compulsory deductible, which cancels out the negative effect caused by moral hazard. Building forth on the study of Vluggen (2015), this research investigates the effect of the yearly increase of the compulsory deductible on the number of visits to Medical Specialists at a hospital, as the services of Medical Specialists are included in the deductible. Moreover, in this study it is examined how the choice for a voluntary deductible effects individual demand for medical care. Using Dutch data from 2007, Van Ophem and Berkhout (2009) investigated the decision to accept a voluntary deductible or not. They conclude that the main reason for the Dutch government to introduce a deductible - that is, making individuals more aware of the costs involved in health care - appears to be irrelevant. However, as stated before, their study used cross-sectional data of 2007. Since the compulsory deductible currently has already been in operation for almost ten years, more data is available to reinvestigate the effect of the voluntary deductible on the demand for health care. In this paper use is made of data of the LISS (Longitudinal Internet Studies for the Social sciences) panel administered by CentERdata (Tilburg University, The Netherlands).

This paper is structured as follows. A discussion of related previous research and the economic considerations is provided in Section 2. In this section the Dutch health care system is also briefly reviewed. Section 3 contains information on the data used and in Section 4 the econometric model that is used to model the yearly number of visits to Medical Specialists is presented. The

(9)

7

estimation results and their implications are discussed in Section 5, after which some conclusions are provided in Section 6. The last section contains possible pitfalls concerning this investigation and discusses some suggestions for further research.

(10)

8

2. Economic Considerations and Previous Research

The increase of the compulsory deductible has been an important topic in politics in the Netherlands in the past years. Not only Dutch citizens are affected by the developments concerning the compulsory deductible, the Dutch government and insurers are too. At this moment, limited previous research is available on the effect of deductibles on the demand for health care, however, it has expanded in recent years. Before reviewing the existing literature extensively, this section first provides an introduction on the Dutch health care system and the reformation that was implemented in 2006. Lastly it is discussed in which way this paper contributes to existing literature.

2.1 The Dutch health care system

The relation between deductibles in health insurance and the demand for health care is obvious, at least in theory (Van Ophem and Berkhout, 2009). Deductibles reduce the effect of moral hazard: individuals are believed to consciously think about their utilization of medical care due to the financial consequences involved, which diminishes the use of unnecessary health care (Van Ophem and Berkhout, 2012). Policy makers assume that in practice individuals are willing to actively manipulate their health care demand if this is advantageous for them. With this in mind, the Dutch government introduced a new health care insurance system in 2006. This reformation shifted the health care system from a publicly provided one to a more progressive structure (Maarse and Ter Meulen, 2006).

On January 1st 2006, a no-claim of €255 per year was introduced for every citizen who was 18 years or older at that time. This implicated that every individual without health insurance claims during a certain year received €255 at the end of the concerning year. If the total amount claimed was positive but less than €255, the individual was refunded the difference between €255 and their claim. Individuals who claimed more than €255 in the concerning year received nothing1. The General Practitioner was excluded from the no-claim regulation, i.e. individuals were allowed to visit a General Practitioner without getting a reduction on their no-claim refund. The no-claim regulation was replaced by a compulsory deductible of initially €150 on January 1st 2008. From this moment on, everyone has to pay a certain amount out of their own pocket each year before their health care costs are refunded by health insurance companies. Like with the no-claim regulation, visits to a General Practitioner are excluded from the compulsory deductible. On top of the compulsory deductible, insureds have the possibility to influence their financial risk in return for a lower or higher monthly health insurance premium (Stroosnier and Van Ophem, 2014). On the one hand, individuals can opt for an additional voluntary deductible, which offers a trade-off between a lower premium and a higher financial risk. On the other hand, individuals

1 A no-claim regulation does not differ fundamentally from a compulsory deductible, except that in the case of a deductible individuals are directly aware of the costs involved, while in the case of a no-claim they become aware of the costs at the end of the concerning year.

(11)

9

have the option to purchase complementary insurance to reduce the risk, while the premium increases. Both deductibles and complementary insurance are methods to shift part of the risk and costs involved with health care from the insurers to the insured (Van Kleef et al., 2009; Gechert, 2010).

Due to the health care system reformation, the role of the insured, insurance companies, health care providers and the government fundamentally changed. Like before the reformation, every Dutch citizen is obliged to purchase at least basic health insurance. From 2006 onwards, health insurance is only offered by private insurance companies which compete for clients. The insurers are in turn obliged to accept all applicants for basic health insurance and are not allowed to differentiate the insurance premium based on gender, age or health status of the insured. In this way, patients are able to join a health insurance policy which fits their needs and preferences well. Insurance companies have to negotiate with medical care providers on price and quality of the provided services, and the role of the government changed from directly managing the system to supervise the correct functioning of the health market (Schäfer et al., 2010). In the new situation, the government sets the guidelines and the insurance companies make the best out of it within the imposed rules. Another method implemented by the Dutch government to diminish unnecessary health care utilization is the fact that the General Practitioner serves as a gatekeeper to further medical care. Patients are required to see a General Practitioner, before getting access to other medical services like Medical Specialists at a hospital.

As stated before, all Dutch residents are required to obtain basic health insurance from a self-chosen insurance company. This basic insurance covers all costs for health services included in the basic package. The content of the basic package is determined by the government and therefore does not differ across insurers. The premium that has to be paid for this basic package, however, differs considerably across insurance companies. The variation in price is caused by differences in insurance policies, (price) agreements between insurers and health care providers, and the height of the financial reserves of the insurance companies. Table 1 provides an overview of the monthly premiums for the basic package of ten arbitrarily chosen Dutch health insurance companies in 2018. The height of the basic premium within each insurance company changes yearly, which is partly caused by the fact that the contents of the basic package is slightly modified each year. In 2018, the following elements were included in the basic package (Independer, 2017):

• Medical care provided by General Practitioners, consultants and obstetricians; • Prescription medication;

• Hospital stay; • Ambulance services;

• Primary mental health care, including treatment by a psychologist and online support; • Secondary mental health care, including psychiatric treatment;

• Dental surgery and dentures; • Maternity care;

(12)

10

• Physiotherapy and remedial therapy as from the 21st session for certain chronic conditions;

• Speech therapy and occupational therapy; • Up to three sessions with a dietitian; • Up to three IVF treatments;

• Dyslexia therapy;

• Smoking cessation programs; • Temporary stay in a care facility; • Artificial teeth up to the age of 23; • Reimbursement of click teeth;

• Treatment for severe chronic constipation; • Breast reconstruction and blepharoplasty; • Medically necessary circumcision.

In case the basic insurance does not suffice in the needs and wishes of an individual, insurance companies provide the possibility to opt for complementary health insurance that covers costs of health care that is not included in the basic package. Health insurers are not obliged to accept everyone who applies for complementary insurance, however, if they do accept applicants, these individuals have to be charged the same premium for the complementary insurance. Complementary health insurance packages differ considerably among insurance companies, but also within particular insurers different complementary packages are offered, which vary in both coverage and premium.

As stated earlier, since 2008 all Dutch residents also have to deal with a compulsory deductible in health care insurance. As mentioned before, visits to a General Practitioner are covered by the basic insurance and therefore do not need to be paid from the deductible. Visits to a Medical Specialist, however, do need to be paid from the deductible. Healthy individuals who do not expect to make use of paid medical services a lot are inclined to opt for a voluntary deductible of €100, €200, €300, €400 or €500, in return for a reduction in insurance premium. The amount of discount depends on the height of the voluntary deductible and differs substantially among insurers. Table 1 lists the monthly premiums of ten health insurance companies in case an individual opts for a voluntary deductible of €500. From this table it can be derived that the average monthly premium reduction received for a voluntary deductible of €500 compared to a contract without a voluntary deductible is €18.44. Therefore, the actual financial risk in case of a voluntary deductible of €500 is much lower than the €500 suggests, i.e. only around €279 on average. Because of the limited amount of money involved, individuals may be inclined to take the risk of a voluntary deductible, in particular individuals who expect to make limited use of health care and who have sufficient financial reserves (Stroosnier and Van Ophem, 2014). Nevertheless, only a relatively small fraction of the Dutch individuals opts for an additional voluntary deductible. In the past years the percentage of insured opting for a voluntary deductible increased from 6.9% in 2012 to 12.3% in 2017 (Zorgwijzer, 2017). Figure 4 displays the increase in the fraction of individuals opting for a voluntary deductible. Table 2 shows the allocation of

(13)

11

individuals across different heights of voluntary deductible. From this table it can be derived that most individuals who opt for a voluntary deductible choose to select the maximum amount of €500.

2.2 Previous research

This section provides a comprehensive overview of previous research concerning health care demand. First, some literature concerning the relation between health care demand and moral hazard and adverse selection is analyzed. Next, a closer look is taken at some papers which make use of Dutch data and focus on the period after the health care reformation in 2006. Thereafter a closer look is taken at some papers which consider the choice for a voluntary deductible. Finally, it is reviewed which factors affect the number of doctor visits according to existing literature.

2.2.1 Moral hazard and adverse selection

The major argument of the Dutch government for the implementation of the compulsory deductible was to make people more aware of the costs involved in medical care and thereby diminish the unnecessary utilization of health care. According to Van Ophem and Berkhout (2009, 2012), the relation between deductibles in health care insurance and the demand for medical care is evident. If the use of health care is free, the demand will be higher. This phenomenon is known as moral hazard. Moral hazard suggests that not all health care demand is strictly necessary, as individuals will demand more medical care in case that no costs are engaged with this health care consumption. Deductibles will, at least in theory, provide an incentive to make conscious decisions about health care utilization, as it involves direct financial consequences. The fundamental argument of policy makers is the believe that individuals can intentionally manipulate their demand for health care and are also willing to do so. Provided that this is true, the introduction of the compulsory deductible will diminish the effect of moral hazard and will reduce the growth of health care expenditures.

Along with moral hazard, adverse selection is another phenomenon that must be taken into consideration when investigating the effect of insurance choice on the demand for medical care (Frank et al., 2000). Insurance markets are prone to suffer from information asymmetry between insurers and insured, as individuals know more about their own health risk profile than insurers (Cardon and Hendel, 2001). Less healthy people are inclined to purchase more complementary insurance than healthier individuals. Moreover, healthier individuals are more likely to obtain an additional voluntary deductible on top of the compulsory deductible. In general, less healthy individuals consume more health than healthier ones, which is relatively less costly for them. Adverse selection is a major issue for insurance companies, since they are not able to correct for the asymmetric information in the premium of insurance schemes.

(14)

12

As stated before, the available previous research on the effect of deductibles on health care utilization is not very extensive. Nevertheless, a considerable wide range of literature exists on the relation between health care demand and moral hazard and adverse selection. The question arises whether the phenomena of moral hazard and adverse selection exists in practice, and if they do exist, what their effects are on the consumption of health care. Some studies on the relation between health insurance coverage and health care demand can be found in Arrow (1963), Cameron et al. (1988), Ellis (1989), Cameron and Trivedi (1991), Hurd and McGarry (1997), Chiappori et al. (1998), Holly et al. (1998), Frank et al. (2000), Schellhorn (2001), Beaulieu (2002), Riphahn et al. (2003), Sapelli and Vial (2003), Deb et al. (2006), Finkelstein and McGarry (2006), Barros et al. (2008), Cutler et al. (2008), Resende and Zeidan (2010), and Schokkaert et al. (2010).

Very few studies focus only on the effect of moral hazard on health care demand. In early research, Arrow (1963) acknowledges the existence and presence of moral hazard in the relation between health insurance and the demand for medical care. The more recent study of Chiappori et al. (1998), who use longitudinal data from a French natural experiment, also focuses mainly on the moral hazard phenomenon. Their analysis indicates that no evidence of moral hazard can be found in the demand for physician office visits, however, they do find evidence of moral hazard in the demand for home physician visits.

A larger share of literature intents to investigate only the effect of adverse selection in the relation between health insurance choice and demand for health care. The evidence regarding the impact of adverse selection, however, is not straightforward. Among others, Ellis (1989), Frank et al. (2000), Beaulieu (2002), and Deb et al. (2006) conclude that adverse selection is strongly present in the purchase of health insurance, while Cameron and Trivedi (1991) find weak evidence of adverse selection. Nevertheless, some other studies, such as Finkelstein and McGarry (2006) and Resende and Zeidan (2010), observe no adverse selection effects. Hurd and McGarry (1997) find that those individuals who are most heavily insured demand the most medical care. However, they dedicate this relation to the effect of the incentives embodied in insurance rather than to adverse selection in purchase of health care insurance. Although some studies do not find adverse selection effects, most investigations seem to find some evidence for adverse selection, which corresponds to microeconomic theory. Cutler et al. (2008) provide an explanation for the seemingly limited effect of adverse selection, namely individual behavioral differences towards risk. Hence, estimating the effect of adverse selection is only possible if the applied model allows for these behavioral differences.

The vast majority of studies aim to discuss moral hazard and adverse selection together. The general belief in these papers is that both phenomena exist, but that moral hazard and adverse selection are interrelated and hard or even impossible to distinguish (Chiappori et al., 1998; Van Ophem and Berkhout, 2012). Like Hurd and McGarry (1997), Chiappori et al. (1998) encounter a positive correlation between insurance coverage and health care demand. However, they state that it is ambiguous whether the increase in health care consumption is a reaction on the extent to which the individual is insured, or conversely, that an individual’s insurance choice is

(15)

13

influenced by the expected health care demand in the near future. This affirms the interrelated nature of moral hazard and adverse selection. Among others, Cameron et al. (1988), Schellhorn (2001), Sapelli and Vial (2003), and Barros et al. (2008) find strong evidence of adverse selection in the purchase of health insurance and, in addition, small moral hazard effects. Only Barros et al. (2008) find no effects of moral hazard for dentist visits. Schokkaert et al. (2010) aim to analyze the Belgium health care system, which has a structure similar to the Dutch reformed health care system, i.e. with a compulsory deductible and the possibility to opt for a voluntary deductible and complementary insurance. Their results show no evidence of both adverse selection and moral hazard. In each of the foregoing studies, a model based on cross-sectional data is employed. Riphahn et al. (2003) use longitudinal data from the German Socioeconomic Panel which surveys a representative sample of German households. They aim to approach moral hazard and adverse selection separately, by stating that adverse selection affects the decision on insurance coverage and that moral hazard influences health care demand conditional on existing insurance coverage. A bivariate count estimator is applied for the number of visits to a General Practitioner and Medical Specialist in the hospital. In this way, account is taken for the possible correlation between General Practitioner visits and Medical Specialist visits. Their results indicate that an individual’s health care demand is not influenced by insurance type, which suggests that effects of moral hazard are absent. However, Riphahn et al. (2003) state that high-risk individuals are indeed more likely to obtain more insurance, which implies that effects of adverse selection are present. They argue that their results are in line with the conclusions of Holly et al. (1998), who confirm the endogeneity between the insurance coverage choice and subsequent health care demand by using cross-sectional data from the Swiss Health Survey.

Since the evidence with respect to adverse selection in previous investigations is mixed, this investigation aims to take the potential effects of adverse selection into account.

2.2.2 Research on Dutch data after the health care reformation

In 2006 the Netherlands experienced a health care system reformation. Initially a no-claim regulation was introduced, which was replaced by a compulsory deductible in 2008. Since the reformation, only limited research has been done on Dutch data regarding the effects of the new regulations. Some papers which address this topic are discussed in this subsection.

Van Ophem and Berkhout (2009) analyze the decision to opt for a deductible on top of the no-claim of €255 in the case of health insurance in the Netherlands. Dutch data from the DNB Household Survey 2007 is used to estimate a simultaneous model, specifying both the choice for an additional deductible and the number of doctor visits. They use a Poisson and Negative Binomial count model to model the number of doctor visits, which account for unobserved heterogeneity. Their results indicate that the health status of individuals and the expected demand do not affect the choice for an additional deductible. Therefore, they conclude that the main argument of the Dutch government for introducing deductibles, i.e. making individuals more aware of the costs involved in health care, seems not to be relevant. Since the no-claim

(16)

14

regulation was replaced by a compulsory deductible in 2008, Van Ophem and Berkhout (2012) reinvestigate the decision to accept a voluntary deductible using data of the LISS panel from 2010. The results from this investigation suggest that the choice for a voluntary deductible does not firmly depend on the self-reported health status of individuals, but does depend on their expected demand for medical care. From these findings they conclude that adverse selection is a relatively minor issue, however, moral hazard appears to be an important determinant of deductible choice. A major deficiency of both studies of Van Ophem and Berkhout (2009, 2012) is that they do not account for endogeneity of the regressors, which is inevitably present.

Stroosnier and Van Ophem (2014) address the choice to opt for complementary health insurance in the Netherlands using data of the LISS panel from 2009. Several specifications of the switching count model are estimated, which account for the simultaneity between the choice for complementary insurance and health care utilization. The Poisson and Negative Binomial count models are used to model the number of physician visits. Although sensitive to model specification, the results indicate that the decision to obtain complementary insurance depends on the self-reported health status of individuals and the difference between expected health care demand with and without having complementary insurance. As a consequence, both moral hazard and adverse selection seem to be relevant in the choice for complementary health insurance. In this paper it is attempted to account for endogeneity, however, problems are being faced when estimating the effect of moral hazard.

All aforementioned papers are based on cross-sectional data. As the reformed health care system has already been in operation for almost ten years at this moment, more data has been gathered and therefore panel data analysis becomes feasible. Panel data analysis is preferred over cross-sectional data analysis, since it considerably improves estimation precision due to the increased number of observations. Section 4.1 elaborates on some more advantages of using panel data rather than cross-sectional data.

Vluggen (2015) makes use of panel data analysis by using LISS data from 2007 until 2013 to examine the effect of the increase of the compulsory deductible on the yearly number of visits to a General Practitioner. To guarantee decent access to basic health care in the Netherlands, visits to a General Practitioner are excluded from the compulsory deductible. As a result, the number of physician visits seems to be irrelevant in this investigation. Nonetheless, Vluggen (2015) attempts to investigate whether the General Practitioner is taking over the role of the Medical Specialist, which is included in the compulsory deductible. To do so, first the Negative Binomial count model is exploited which assumes that all included explanatory variables are strictly exogenous. As exogeneity of the regressors does not appear to be a justifiable assumption, subsequently the Arellano-Bond difference GMM estimation method is applied, which accounts for predeterminedness and endogeneity of health status indicators and choices for voluntary and complementary insurance and therefore provides more reliable results. The final results indicate that no evident significant effect of the increase of the compulsory deductible on the number of General Practitioner visits can be concluded.

(17)

15

All foregoing papers use the number of visits to a General Practitioner as a measure for health care demand. Many other studies do the same (Cameron et al., 1988; Deb and Trivedi, 1997; Chiappori et al., 1998; Holly et al., 1998; Schellhorn, 2001; Deb and Trivedi, 2002; Cockx and Brasseur, 2003; Riphahn et al., 2003; Sapelli and Vial, 2003; Winkelmann, 2004; Zimmer and Trivedi, 2006; Barros et al., 2008; Schokkaert et al., 2010; Bíró, 2014). In the Netherlands, there is limited data available on individual health care expenses, which is evidently a better measure for health care consumption. Therefore, usually the only relevant measure is the number of doctor visits in a certain year. Even though this measure does not fully describe health care demand, Van Ophem and Berkhout (2009) believe that using General Practitioner visits as a proxy for health care utilization is a valid procedure in the current Dutch situation. Their reasoning is twofold. Firstly, the General Practitioner serves as a gatekeeper in the Dutch health care system. Patients are required to see a General Practitioner, before getting access to other medical services like Medical Specialists at a hospital. The only exception is in the case of an emergency when individuals can go, or often will be brought, straight to the hospital. Secondly, although visiting a General Practitioner is free for the patient, the results from such a visit usually bring costs. General Practitioners will frequently prescribe drugs or will refer to a Medical Specialist, of which the involved costs will be charged to the patient. Individuals are believed to take this fact into account when visiting a General Practitioner in the first place. Given these arguments, Van Ophem and Berkhout (2009) consider the number of visits to a General Practitioner to be a good proxy for health care demand. Moreover, the decision to visit a General Practitioner is an individual choice which is made by the patient independently. Decisions resulting from this visit, for example using drugs or seeing a Medical Specialist, are made in accordance with the General Practitioner. Hence, these subsequent decisions are not merely individual and the opinion of the physician will commonly be decisive. For this reason, together with the arguments put forward by Van Ophem and Berkhout (2009), Vluggen (2015) has chosen to model the number of visits to a General Practitioner and to not take the number of visits to a Medical Specialist into account. She argues that in this way the effect of deductibles on the independent choice of the individual to visit a General Practitioner can be investigated, which will not be influenced by opinions of other doctors. Section 2.3 elaborates on how the demand for health care will be specified in the current investigation.

2.2.3 Choice for voluntary deductible

Next to investigating the effect of the compulsory deductible on the demand for health care, this paper also aims to estimate the effect of the voluntary deductible on health care demand. A number of papers investigate the choice for voluntary deductibles in health care insurance. Gorter and Schilp (2009) solely investigate the choice for a voluntary deductible, whereas Mueller and Monheit (1988) and Schellhorn (2001) study the relation between the choice for a voluntary deductible and the subsequent demand for health care. In his model, Schellhorn (2001) allows for endogeneity between deductible choice and demand for health care. Both Mueller and Monheit (1988) and Schellhorn (2001) simply consider a constant influence of voluntary deductible on

(18)

16

health care consumption, while in a related study, Sapelli and Vial (2003) allow for a more comprehensive specification where all variables in health care demand are allowed to vary across deductible choice.

As stated before, Van Ophem and Berkhout (2009, 2012) investigate the decision to opt for a voluntary deductible in the Netherlands. They state that a fundamental complication in their paper is the measurement of health care demand across different heights of voluntary deductibles individuals can choose from. It is aimed to estimate the difference in demand for different levels of the deductible, but the question arises what to compare. Since in the Netherlands individuals can choose from six options, i.e. a voluntary deductible of €0, €100, €200, €300, €400 or €500, a fair 120 (5!) combinations can be compared. For that reason, they decide to only model the choice between no voluntary deductible and a positive voluntary deductible. Even though this method seems inefficient as not all available information is used, Schellhorn (2001) shows that the difference between using full information or bivariate information about the voluntary deductible is not substantial. In his study, the estimated effects of most explanatory variables are very similar in both cases. Another argument of Van Ophem and Berkhout (2009, 2012) is that in the Netherlands, a relatively low proportion of their sample opt for any positive voluntary deductible, i.e. about 30 percent. In the sample of the current investigation, the yearly percentage of individuals opting for a positive voluntary deductible is even lower on average, as can be seen from Table 3. Hence, in this study, an identical approach is used regarding the explanatory variable which represents the choice for a voluntary deductible.

2.2.4 Determinants for the number of doctor visits

Previous research analyzing what factors influence the number of visits to a General Practitioner is abundant, however, limited investigations consider the number of visits to Medical Specialists. Practically all studies include variables describing an individual’s health status, age, income and job status, however, a considerable wide range of alternative explanatory variables is used in previous studies regarding health care demand.

Schellhorn (2001) investigates the effect of voluntary deductibles in health insurance on the demand for health care in Switzerland for men and women separately. In this study, models for visits to both a General Practitioner and Medical Specialists are estimated and the choice for a voluntary deductible is considered to be endogenous. Therefore, GMM techniques are exploited to account for this endogeneity. Explanatory variables representing health status, age, education, job status, income and ethnicity are used, along with characteristics of an individual’s place of residence, like dummies for small towns, large towns and metropolitan areas. Regarding visits to a General Practitioner, significant positive effects on the number of visits for both men and women are found for age, having a chronic disease, being a drinker and having overweight. A significant negative effect is found for men who live in metropolitan areas. A higher level of education provides significant results for both men and women. For men the effect on the number of visits is positive, while for women a negative effect is found. Differences in significance for men and

(19)

17

women are found for being a foreigner, with a small significant positive effect for women. Income appears to have no clear impact on the number of General Practitioner visits. Regarding visits to a Medical Specialist, significant positive effects on the number of visits for both men and women are found for a higher level of education, having a chronic disease and having underweight. Surprisingly, significant negative effects are found for age and being a heavy smoker. Like with the number of visits to a General Practitioner, differences in significance for men and women are found for being a foreigner, with a significant positive effect for men. Both income and overweight appear to have no clear impact on the number of visits to a Medical Specialist.

Riphahn et al. (2003) aim to model the demand for health care in Germany by applying a bivariate random effects estimator in a count data setting, to enable efficient estimation of this model type with panel data. By means of this bivariate system, the possible correlation between General Practitioner and Medical Specialist visits in a certain time period is taken into account. This aspect should also be considered in the current study, since in the Netherlands the General Practitioner serves as a gatekeeper to further medical care and therefore these variables are likely to correlate. However, Riphahn et al. (2003) do not account for the possibility that individuals may act upon a referral with a delay. They use explanatory variables like age, degree of health satisfaction, whether the individual has a handicap or not and its degree, whether the individual is married, years of schooling, income and job status. In their investigation, the only significant positive effect on both visits to a General Practitioner and Medical Specialist is found for the degree of handicap. Significant negative effects on visits to both doctors are found for age, degree of health satisfaction and schooling. Self-employment has a significant negative effect on the number of visits to a General Practitioner, while the estimated coefficient for visits to Medical Specialists is insignificant.

Vluggen (2015) employs the Negative Binomial count model as well as the Arellano-Bond difference GMM estimation method to estimate the effect of the increase of the compulsory deductible on the number of General Practitioners in the Netherlands. She argues, however, that the Arellano-Bond estimator provides the more reliable results, as this method is able to cope with potential endogeneity of health status indicators and insurance choice. Next to usual variables like age, income and job status, a wide range of health-related variables are used: having a chronic disease, number of days spent in the hospital last year, having undergone an operation last year, self-reported health status, health status compared to last year, contact with certain medical specialists, regularly suffering from certain complaints, being told by a physician to suffer from certain diseases, and the extent to which the individual’s physical health hinders daily, social and work-related activities. Following the results of the Arellano-Bond GMM estimator, significant positive effects are found for, among others, having undergone an operation in the previous year, all variables representing contact with certain specialists, and being diagnosed by a physician to suffer from diabetes, a too high blood sugar level or a high blood pressure. For age, a significant negative effect is found.

Based on the variables used in the previously discussed papers, the availability of data in the LISS survey which is used for this investigation, and common knowledge about determinants for

(20)

18

visits to Medical Specialists, the explanatory variables for the current study are carefully chosen. Section 3.3 elaborates on the exact set of variables used in this study.

2.3 Approach of this study

This study investigates whether the yearly increase of the compulsory deductible and the choice for a voluntary deductible have any effect on the way patients make use of health care services. It builds forth on the paper of Vluggen (2015), who examined the effect of the increase of the compulsory deductible on the yearly number of visits to a General Practitioner. As stated before, her fundamental argument to model the number of General Practitioner visits and not the number of Medical Specialist visits is the fact that visiting a General Practitioner is an independent choice of the individual, while the decision to see a Medical Specialist has to be made in accordance with the General Practitioner. However, since visits to a Medical Specialist are included in the compulsory deductible while visits to a General Practitioner are not, it might be interesting to incorporate Medical Specialist visits. Therefore, this research aims to investigate the effect of the yearly increase of the compulsory deductible on the number of visits to Medical Specialists at a hospital.

Investigating the effect of the compulsory deductible on the demand for health care based on individual data is not very obvious. Panel data techniques are designed for data which vary over both time and individuals. The height of the compulsory deductible, however, varies over time but is equal for each individual. Therefore, a complication arises when estimating the effect of the compulsory deductible on health care demand, since this effect cannot be isolated from the effect of other factors that are equal for all individuals but vary over time, such as the business cycle. For this reason, it might be more interesting to explore the relation between the voluntary deductible and the number of visits to Medical Specialists, since the height of the voluntary deductible does vary across individuals. Hence, in addition to investigating the effect of the compulsory deductible on health care demand, this study examines how the choice for a voluntary deductible effects individual demand for medical care.

Since the dependent variable is non-negative and integer, it might seem obvious to make use of count models. Various specifications of count models are exploited by, among others, Schellhorn (2001), Riphahn et al. (2003), Van Ophem and Berkhout (2009, 2012), Schokkaert et al. (2010), Stroosnier and Van Ophem (2014), and Vluggen (2015). However, the presence of endogeneity due to simultaneity between the dependent variable and explanatory variables which represent an individual’s health status and insurance choice should not be ignored. Among others, Schellhorn (2001) and Vluggen (2015) make use of GMM estimation methods, which account for endogeneity and predeterminedness of regressors. Vluggen (2015) exploits the Arellano-Bond difference GMM estimator, which is able to deal with the complication of unobserved time-invariant effects in panel data, since the individual-specific effects are removed by taking first differences. For these reasons, this study focuses on GMM methods to estimate the effect of deductibles on health care demand.

(21)

19

Finally, this paper contributes to existing literature by employing a dynamic panel data model to estimate the effect of deductibles on health care demand. The first lag of both the dependent variable and some relevant explanatory variables are included in the model. Furthermore, interaction terms are introduced which are the multiple of a relevant regressor and the height of the compulsory deductible2. A detailed description on the construction of the interaction terms and the interpretation of their estimated coefficients can be found in Section 4. This section also elaborates on the model estimated in this study in more detail, as well as on the endogeneity and predeterminedness of included explanatory variables.

2 To the author’s knowledge, interaction terms have never been used in investigations of the effect of deductibles on health care demand.

(22)

20

3. Data

This section focuses on the data used in this study. The data come from the LISS (Longitudinal Internet Studies for the Social sciences) panel3. The LISS panel is collected by CentERdata, an institute for data collection and investigation established at Tilburg University, The Netherlands. The panel is randomly drawn from the population register by Statistics Netherlands (Centraal Bureau voor de Statistiek) and is based on a true probability sample of households and individuals in the Netherlands. This means that every household has an equal probability of being chosen and therefore the sample is representative for the composition of the Dutch population. Participants that could not otherwise participate due to the fact that they do not have Internet access, are provided with a computer and Internet connection. Panel members complete monthly online questionnaires and are paid for each completed survey. One member of the household provides the household data and this information is updated at regular time intervals.

The LISS panel has been in operation since October 2007 and consists of approximately 4,500 households, comprising 7,000 individuals. A study on the representativeness of the initial LISS panel compared to the Dutch population shows that in most categories the differences are quite small, except for the distribution over age groups and number of household members. In the first few years elderly people and single person households were underrepresented in the LISS panel (De Vos, 2010). To improve the representativeness of the panel, between June and December 2009 the recruitment of a refreshment sample was carried out. In the preceding years the gap between elderly and other age groups, and the gap between single person and multiple person households decreased. A second refreshment sample was carried out between October 2011 and May 2012 and a third was carried out between November 2013 and June 2014. Like the original sample, these samples were drawn randomly from the population register by Statistics Netherlands. Another major difference in the composition of the LISS panel compared to the Dutch population, is that in the LISS panel certain ethnicity groups are underrepresented. This dissimilarity arises in every year of the panel and could be caused by the fact that households in which no adult is capable of understanding the Dutch language are excluded from the reference population (CentERdata, 2017).

A major part of the interview time available in the LISS panel is reserved for the LISS Core Study. The LISS Core Study is a yearly repeated longitudinal study, which is designed to keep track of changes in the life course and living conditions of the LISS panel members. It contains studies on different topics such as Health, Religion and Ethnicity, Work and Schooling and Economic Situation. Next to the LISS Core Study, researchers have the possibility to use the panel for data collection for different research purposes. In the past years, the LISS panel has been used by a considerable variety of disciplines from the social sciences, but also from the medical and biomedical sciences.

(23)

21

3.1 Construction of the dataset

In this investigation use is made of data originating from the following surveys: • Health Core Study

• Background variables LISS panel

The Health Core Study contains information about individuals’ medical status, how they perceive their own health, what type of health insurance was taken and how often use was made of services of a General Practitioner and various Medical Specialists. The set of Background variables contains information on general characteristics of LISS panel members like gender, age, job status and household composition. For this study all nine available waves of the Health Core Study are used, which corresponds to data between 2007 and 2016. Table 4 gives an overview of the waves in terms of the month of data collection and the number of households participating in each wave. Note that data of 2014 is not available in the Health Core Study and that in 2015 the month of data collection differed from the other years. Possible complications of this missing year are discussed in Section 5.2.

The final dataset used for this investigation has been constructed as follows. First, the initial datasets of the different waves of the Health Core Study have been edited in Stata to obtain a dataset with relevant variables for this study. All variables that are used in this investigation have been renamed and the remaining irrelevant variables have been removed from the sample. In this way, the names and number of variables are equal in each wave. The same has been done for the relevant variables in the set of background variables of the concerning years. Next, the waves of each category have been merged to obtain a separate dataset for each year and thereafter all years have been combined to create the required panel dataset for this study.

Each year between 5,000 and 6,700 participants filled out the Health questionnaire, however, these participants were not necessarily the same in each year. This is due to the fact that participants deceased or stopped participating for other reasons and that certain households started participating during one of the refreshment samples. Consequently, an unbalanced panel dataset has been created by merging the different years. To investigate yearly changes in behavior regarding medical consumption caused by the yearly increase in the compulsory deductible it is more convenient to deal with a balanced panel, which means that data is available for every individual or household in every year of the study. The merged panel dataset initially contained 12,159 unique households spread over the different years. To obtain a balanced panel, participants who did not fill out the Health questionnaire in all nine consecutive waves have been left out of the sample. After this, 2,001 individuals remained. Since the compulsory deductible does not apply to individuals under the age of 18, their health care demand will not be affected by the yearly increase in the compulsory deductible. For that reason, minors have also been left out of the sample, which resulted in a sample of 1,964 remaining individuals.

After merging the datasets and leaving out the observations mentioned before, there were still some missing values on variables that are presumably relevant when investigating their effect on

(24)

22

the number of visits to Medical Specialists. As it is undesirable to reduce the sample even further, the majority of these missing values has been generated manually. This has been done using different procedures. One method involves using the value of the preceding or consecutive year of the concerning variables, by inserting the value of one of these surrounding years in the open spot of the missing observation. This procedure has been conducted for all dummy variables that contained missing values, that is, for the variables representing the height of the voluntary deductible, having complementary insurance, household position, having a paid job, having a chronic disease, being a smoker, drinking at least five days per week on average, highest level of education, being told by a physician to suffer from certain diseases, and suffering from certain complaints. For the variables net income and weight, missing values have been replaced by the average of the preceding and following year when both are known and by the value of either the preceding or following year otherwise. For missing observations concerning age, a similar procedure has been conducted. Here the value of the preceding or following year plus or minus one was used, or plus or minus one and a half if the observation of 2015 was missing. In some cases the missing value actually represents a zero. In these cases the missing observations have been replaced by zeros. This has been done for the variables representing the number of visits to a General Practitioner and Medical Specialists, the number of days in the hospital and undergoing an operation. An overview of the number of values that have been manually generated can be found in Table 5.

After the procedure of generating missing values manually there were still some missing values within the income variable. This is due to the fact that for some individuals, the net income is unknown for all nine consecutive waves. Therefore there is no reference point for their income, so no educated guess can be made concerning their income. The concerning individuals have also been left out of the sample. This left 1,898 remaining participants in the sample.

Finally, the values of some variables of interest, i.e. the number of visits to a General Practitioner or a Medical Specialist reported by the participants, are subject to some outliers. For example, the data on the number of visits to a General Practitioner contains one reported value of 400 and two values of 800. This many visits are highly unlikely, as this would mean that the individual visited the General Practitioner more than once or twice a day on average. In this investigation it is presumed that more than 100 visits to a General Practitioner per year is very unlikely. Hence, the individuals who reported higher numbers have been removed from the sample, which leaves 1,893 remaining individuals in the sample. Regarding the number of visits to Medical Specialists at a hospital, only one outlier of 210 visits can be identified. However, the individual who reported this value was 84 years old in the concerning year which makes this seemingly high number much more reasonable. Therefore, no outliers have been removed concerning the number of visits to a Medical Specialist, which means that the final sample used in this investigation consists of 1,893 participants.

Leaving out the households that did not participate in all waves, minors, individuals for whom the income was unknown for all years, and outliers led to an extreme loss in the number of observations. However, the final sample of 1,893 participants seems to be sufficiently

(25)

23

representative for conducting this research, since the composition of the final sample is fairly similar to the composition of the initial LISS sample. Table 6 provides a short overview of some sample statistics of the initial LISS sample and final sample of 1,893 participants. For the sake of comparison some statistics of the Dutch population are provided. Comparing the initial and final sample, it can be observed that in the final sample the lowest age group is underrepresented. This can be justified by the fact that all minors have been removed from the sample since the compulsory deductible does not apply to them. Moreover the distribution across income classes has slightly changed, with more individuals in the lower classes and less in the higher classes. Previous literature suggests that income has no significant effect on health care demand (Schellhorn, 2001; Van Ophem and Berkhout, 2009). Therefore, the distribution across income classes is considered to be rather irrelevant. Finally, the average number of visits to a General Practitioner and Medical Specialists at a hospital stayed roughly the same. This is convenient, since these variables serve as a proxy for health care demand.

3.2 Dependent variable

As stated in the literature review, this study investigates whether the yearly increase of the compulsory deductible and the choice for a voluntary deductible have any effect on the way patients make use of health care services. In previous research, the number of visits to a General Practitioner has generally been used as a proxy for health care demand. In this study, however, the dependent variable of concern is the number of visits to Medical Specialists at a hospital. Some descriptive statistics of the dependent variable are presented in Table 7. For the sake of clarity and as a matter of comparison, descriptive statistics for the number of visits to a General Practitioner are provided in Table 8.

The values of the number of visits to a Medical Specialist might be subject to reporting errors. Some peaks can be observed in the reported number of visits at the counts of 10, 15, 20, 50, or sometimes even 100. These potential errors are possibly caused by the fact that individuals did not keep perfect track of the exact number of visits and hence reported rounded numbers. These kind of reporting errors could affect the standard errors in the estimation. However, the number of potentially rounded values is a relatively small fraction of the total sample, so this aspect is considered to be negligible.

3.3 Explanatory variables

The LISS Health survey includes a great many variables which could serve as potential explanatory variables of the model in the current investigation, many of which are closely related to each other. Since including more explanatory variables does not necessarily result in a better model fit, especially when the included variables are interrelated, a well-considered choice has to be made of which variables to include as regressors in the model. As stated in Section 2.3, this

(26)

24

study builds forth on the paper of Vluggen (2015). The explanatory variables used in the current investigation are therefore based on the variables used in the study of Vluggen, as well as on variables used in other relevant papers discussed in Section 2.2, such as Schellhorn (2001) and Riphahn et al. (2003). A limiting factor is that not all variables used in the latter studies are available in the dataset of LISS. Based on the availability in the LISS dataset, the mentioned previous studies, and common knowledge about determinants for visits to Medical Specialists, the following explanatory variables are in some form used in the model of this investigation:

• GPvisits: the number of visits to a General Practitioner in the past year; • voluntary_deductible: dummy for having voluntary deductible;

• complementary_insurance: dummy for having complementary insurance; • householdhead: dummy for being household head;

• bmi: Body Mass Index – weight (in kilograms) divided by the square of length (in meters); • alcohol: dummy for drinking at least five days per week on average in the past year; • chronicdisease: dummy for having a chronic disease;

• operation: dummy for undergoing an operation in the past year; • educationdum(●)4: dummies for the highest level of education; • healthstatusdum(●): dummies for self-reported health status;

• healthcomparisondum(●): dummies for health status compared to last year;

• disease_(●): dummies for being told by a physician in the past year to suffer from certain diseases;

• complaints_(●): dummies for regularly suffering from certain complaints.

In addition to the aforementioned variables, some other variables were used in initial estimations of the model. These variables involve gender, age, having a paid job, net income, being a smoker and the number of days in the hospital in the past year. Since these variables appeared to have relatively high p-values, it was decided to eliminate them from the model. However, this does not directly imply that these variables do not have any effect on the number of visits to Medical Specialists. In panel data models, dummies for individual-specific and time-specific effects are typically included. These variables absorb everything that is specific for an individual or a certain time period. Therefore, variables such as gender, age, and income presumably do have effect on the number of visits to Medical Specialists, but in this model these effects are captured by the dummies for individual-specific and time-specific effects.

A complete overview of the initial set of variables exploited in this investigation, including the different outcomes of the categorical variables, is provided in Appendix A. Appendix B elaborates on how the final set of explanatory variables used in this study is established. Quantitative information on both the explanatory variables included in the model and the eliminated variables is presented in Table 9.

4 Explanatory variables including a dot, (●), represent a group of dummy or categorical variables starting with the preceding characters.

Referenties

GERELATEERDE DOCUMENTEN

The result section of this paper with its regression model and the additional analyses provided evidence for a strong significant positive relation between the level

RQ: How does the adoption of the hospital physician as a new generalist profession influence the vertical differentiation and horizontal integration of the medical department and

Despite the larger sample size, the voluntary deductible has no significant effect on an individual probability to be categorized as an excessive smoker, drinker or

(2016) and the empirical finding of Gorter and Schilp (2012), I hypothesize that risk preference, measured by an individual’s financial risk attitude, is related to

Moreover, as there exist several methods to match individuals with the aid of propensity scores, some of these methods are reviewed to make sure the best method for this research

Hierdie gedig is in die vorm van ’n dialoog tussen die twee skaapwagters Meliboeus en Tityrus.. Meliboeus het sy plaas verloor weens die grondhervorming na afloop van

Wijngaards VU Agent Based Modelling of Dynamics: Biological and Organisational Applications 2002-12 Albrecht Schmidt Uva Processing XML in Database Systems 2002-13 Hongjing Wu TUE