The relation between health insurance and medical expenditures : a study of the development of the endogeneity in the early 21st century

(1)

The relation between health insurance and medical expenditures

A study of the development of the endogeneity in the early 21

st

century

Marlou Beringer (10439048)

Bachelor Thesis Econometrics and Operational Research University of Amsterdam

Tutor: dr J.C.M. van Ophem Date: 24-12-2014

(2)

2

Content

1. Introduction 3 2. Background Theory 4 3. Data 8 4. The Model 12 5. Estimation issues 14 6. Results 16 6.1 Comparable model 16 6.2 Optimal model 18 7. Conclusion 20 References 22

Appendix A – Summary Statistics 24

Appendix B – Auxiliary Regressions 28

Appendix C – Estimation Results Comparable Model 35

(3)

3

1. Introduction

For a long time, whether or not to have health insurance was an individual choice in the United States of America. As a result, almost 47 million American citizens were uninsured in 2011 (Fronstin, 2012), which is about 17.6 percent of the U.S. population. This relatively high uninsured rate had serious consequences for the U.S. public health, as uninsured individuals are likely to postpone necessary medical care and completely forego preventive care. Since 31 March 2014, every citizen of the United States is obliged to buy a certain minimum level of health insurance1. Beyond this minimum, every individual may determine the height of their insurance coverage themselves.

Many researchers have analysed the relation between having health insurance and the demand for medical care. But there are many difficulties in studying this causal relationship. On the one hand, the insurance market is presumed to suffer from adverse selection: people who expect to need much health care are more inclined to purchase medical insurance. On the other hand, whether an individual seeks medical care may depend on whether he or she has insurance. People who have high insurance coverage tend to demand significantly more health care than people without or with lower coverage. This phenomenon is known as moral hazard. Moral hazard and adverse selection are closely related and in empirics they can hardly be distinguished. These two phenomena result in what econometricians call endogeneity.

Endogeneity appears to be a major issue in studying the effect of insurance on health care expenditures. This paper will focus on how this endogeneity developed over time in the early 21st century. Using data from the Medical Expenditure Panel Survey, the model will be estimated for the years 2000, 2004, 2008 and 2012. For every year, a comparable model and an optimal model will be estimated. Because one of the variables – private health insurance – is endogenous, OLS will not give consistent estimators. In order to get consistent estimators, one or more instrumental variables have to be identified and the instrumental variables estimation method2 has to be applied. OLS is used to find valid and relevant

1

The law in which this is captured is called the “Patient Protection and Affordable Care Act of 2010”, also known as “Obamacare”.

2

For more information about the instrumental variables estimation method - or more general: the two-stage least squares method - see Heij et al. (2004). The general method is also presented in Section 4.

(4)

4 instruments. After the model is estimated by IV, the Hausman-test3 is used to check for the presence of endogeneity over the years and I analyse the stability of the endogeneity over time.

This paper is organised as follows. First, some theoretical background is given on the subject of this paper. Thereafter, the data used in this paper are described and summarised for the four different years. Then, in Section 4, the model and Hausman-test are introduced. Section 5 describes the empirical estimation issues with respect to the available variables and instruments in the four different years, and in Section 6, the estimation results are given. The last section provides the conclusion and some further research directions.

2. Background Theory

A few decades ago, Manning, Newhouse, Duan, Keeler, Leibowitz, and Marquis (1987) analysed the results of the RAND Health Insurance Experiment4 (HIE), the largest health policy study in the history of the United States which started in 1971 and continued for fifteen years. One of the main questions in their paper is whether fee-for-service insurance plans with different rates of coinsurance5 yield different expenses on medical care. Their experiment showed clear results: the medical expenses per individual on the free insurance plan are 45 percent higher than medical expenses of people on the insurance plan with a 95 percent coinsurance rate. However, their estimates can hardly be compared to estimates of studies with nonexperimental data. As in nonexperimental data the level of insurance is an individual’s own choice, we must allow for the possibility of endogeneity between health insurance and the demand for medical care.

The question might be raised how the interrelated nature of insurance coverage and the demand for medical care is caused. There are two phenomena which explain a large part of this endogeneity. The first one is adverse selection: people who expect to have high costs on medical care prefer more generous insurance plans than healthy individuals. Adverse

3

Also called the Wu–Hausman test or the Durbin–Wu–Hausman test.

4

For more information about the design of the Rand Health Insurance Experiment, see Manning et al. (1987)

5

The coinsurance rate is the percentage of medical costs that insured individuals have to pay themselves. The remaining amount will be paid by the insurance company.

(5)

5 selection is seen as a major problem in insurance markets, because it eventually may destroy the market. According to Cutler and Zeckhauser (1998), it can lead to three categories of inefficiencies. Firstly, individuals are likely to fail to select the optimal health plan, as prices of health plans to customers do not represent the actual marginal costs for the insurer. Secondly, preferable risk spreading decreases because of the great variability in premiums for insurance plans, which correspond to the variability in risk. Thirdly, insurers manipulate their offerings to attract healthy people and expel sick and old individuals, in order to prevent heavy losses. All three inefficiencies are in some way related to each other. In practice, these inefficiencies are able to occur as a result of informational asymmetries in insurance markets. Most studies confirm the existence of adverse selection (Browne, 1992; Cutler and Zeckhauser, 1998; Cutler and Reber, 1998), however, Cardon and Hendel (2001) have failed to find significant evidence of adverse selection and asymmetric information. They think the gap in medical care expenditure between the insured and uninsured can be attributed to the fact that a large part of the U.S. population is insured via the employer, which is substantially cheaper than regular private insurance. Akerlof (1970) is also sceptical about the existence of adverse selection. He states that insurers induce their own “adverse selection” by interconnecting with employers to offer employer-based insurance plans, as employees often require relatively little medical care. Consequently, health insurance is least available for the people who need it the most.

The second phenomenon which explains endogeneity is the fact that individuals who have a high level of insurance coverage tend to demand more treatment than those who have a lower level of coverage. This is known as moral hazard. Individuals with higher insurance coverage may demand health care inefficiently, in the sense that the costs can be underestimated regarding the actual benefits. This causes welfare loss for the insurer, and eventually also for the health seekers: the high demand for medical care raises the expenses for the insurer, which eventually enhances the cost of insurance coverage for individuals. Manning and Marquis (1996) confirm the overutilization of health care due to moral hazard. As a solution, they indicate that a coinsurance rate of approximately 55 percent would be optimal to negate the effects of moral hazard. However, Albert Ma and Riordan (2002) find evidence that the conventional wisdom that moral hazard causes excessive demand is not universally true. They show that demand management by means of co-payments can result

(6)

6 in deficient demand relative to what they call the ideal insurance6. They also find that a combination of demand and supply management via managed care7 accomplishes ideal insurance under certain conditions.

As mentioned before, it is difficult to analyse the causal relationship between insurance and the demand for medical care. According to Shen (2013) there are three main empirical challenges in studying health care decisions. The first challenge is the complex character of the decision-making process. An individual decides if he or she wants to visit a health care provider, and then the individual and the health care provider together decide what treatment the patient will get. A two-part model is needed to deal with this decision-making process. Another challenge is the fact that there are many people who do not have any health expenditures at all. Positive expenditures are only observed from individuals who decide to visit a health care provider, so the variable expenditure is censored. A way to deal with censored data is to use the Tobit model (Tobin, 1958) or the Heckman model (Heckman, 1976, 1979). However, both models make distributional assumptions about the error terms, which are not necessarily satisfied in practice (Duan et al., 1983). The last challenge is the possible endogeneity between health insurance and the demand for health care: an individual’s decision to utilize medical care may depend on the level of insurance coverage – which is known as moral hazard – but because the level of insurance is a choice variable for the individual, individuals who have a greater demand for medical care may have more incentive to buy a higher level of insurance coverage – which is known as adverse selection. This demonstrates that adverse selection and moral hazard are closely related to each other.

Adverse selection and moral hazard together show that health care demand and the level of insurance are interrelated. This causes the variable insurance to be endogenous with respect to the demand for medical care. There are a few ways to deal with this endogeneity. Some publications use instrumental variable estimation (Cameron et al., 1988; Wooldridge, 2010; Dunn, 2013). A special case of instrumental variable estimation is used by Kowalski

6_{Albert Ma and Riordan (2002, p.87) describe the ideal insurance contract in which moral hazard is absent. “In}

this regime, illness and loss are assumed to be completely contractible; payments and treatment decisions can be contingent on the severity of illness”.

7

Albert Ma and Riordan (2002) think of managed care as giving health care providers incentives to withhold treatment.

(7)

7 (2009), that is censored quantile instrumental variable (CQIV) estimation. This CQIV estimator addresses the limitations in studying health care decisions: it allows the estimates to differ across the skewed expenditure distribution, it allows to handle censoring at zero expenditure without making any distributional assumptions, and it accounts for the endogeneity between price and health care expenditure.

Instrumental variable estimation is the most common method to deal with the endogeneity issue. However, one of the biggest empirical issues is finding instruments that are relevant and exogenous, which means they need to be correlated with the demand for insurance but are not allowed to correlate with the use of health care respectively. Examples of instruments that are used in papers are whether or not a family member is injured (Kowalski, 2009) or the negotiated price between insurers and health care providers (Dunn, 2013). Shen (2013) uses marital status and region as instruments, but these instruments are mainly based on economic theory rather than on the conditions for valid instruments8. In his paper he emphasizes that he recognizes the difficulty in finding valid instruments. Other papers are able to avoid the problem of endogeneity by using experimental data (Rosett and Huang, 1973; Manning et al., 1987; Newhouse and Rand Corporation Insurance Experiment Group, 1993). However, experimental data are rare and often outdated.

This section has shown that endogeneity is an important issue when studying the relation between health insurance and the demand for medical care. As we have seen, much has been written about this endogeneity. Some papers attempt to model the whole interrelated nature of insurance, utilization, and expenditures (Duan et al., 1983; Shen, 2013), while others focus merely on the relation between insurance and expenditures (Rosett and Huang, 1973; Manning et al., 1987; Kowalski, 2009). To my knowledge, none of the existing literature analyses the development of the endogeneity between private insurance and health care expenditures over time. This paper contributes to the literature by studying the latter issue. Section 4 describes the models that will be used to study the endogeneity.

8

For more information about the conditions for valid instruments, see Heij et al. (2004). Besides, Section 4 of this paper provides the formulas associated with these conditions.

(8)

8

3. Data

The data used in this study are from the Medical Expenditure Panel Survey (MEPS), which is a set of large-scale surveys of both households and individuals, their health providers, and employers in the United States. The U.S. Department of Health and Human Services started with this survey in 1996 to collect data on health services, expenditures and insurance coverage of U.S. civilians. MEPS consists of three different components: a Household Component, which collects data from households and their members; an Insurance Component, which collects data from employers on the health insurance plans they offer their employees; and a Medical Provider Component, which collects data from medical providers like hospitals, physicians, psychologists, and pharmacies. The survey is intended to be nationally representative for the population of the United States of America. In this paper, we use the MEPS data of 2000, 2004, 2008 and 2012. We consider the subsample of individuals of 18 years and older. The final samples consist of 14,641 individuals in 2000, 19,172 individuals in 2004, 18,919 individuals in 2008, and 11,517 individuals in 2012.

As mentioned before, health insurance is likely to be endogenous with respect to medical expenditures. In this paper, three categories of health insurances are distinguished, namely private insurance, public insurance and insurance via the employer. The category private insurance consists of individuals who purchased their insurance coverage directly from the insurer. The category public insurance consists of individuals who are covered by any public health insurance, like TRICARE9, Medicare10 or Medicaid11. In the United States, employers often offer health insurance to their employees. Therefore, the third category consists of the individuals who obtained their insurance coverage via their employer. However, there is also a growing number of people who is uninsured. The percentage of uninsured individuals in our study population increased over the years, from 16 percent in 2000 to almost 22 percent in 2012. The endogenous variable that we are considering in this paper is solely private insurance, as this kind of insurance is assumed to be an individual’s own choice, whereas the others are not.

9

TRICARE is the health care program for uniformed service members and their families.

10

Medicare is the federal health care insurance program for people who are 65 and older or disabled.

(9)

9 The definition of health care expenditures, which is our dependent variable, is the total amount paid by an individual for medical services. This includes the direct payments made by the individual, as well as the payments made by the insurer. Payments for over-the-counter drugs are not available in the MEPS data. As the expenditures are deduced from both the MEPS Household Component and the Medical Provider Component, the data are more reliable than those from regular surveys. In our model, we take the natural logarithm of the total amount of health care expenditures, as this makes the model easier to interpret by giving relative changes in medical expenditures. This is a conventional method which is also used in similar papers (Manning et al., 1987; Shen, 2013; Dunn, 2013).

The explanatory variables used in this paper are almost similar to those used by Shen (2013). Besides the insurance variables, several demographics, socioeconomic characteristics, and health-related characteristics are used. The first includes gender, age, race/ethnicity (white, black, Asian, other), marital status (married, other), family size, and region (Northeast, Midwest, South, West). The socioeconomic characteristics are covered by income and years of education. The health-related characteristics include a number of comorbidities, the perceived mental health status, whether they are currently smoking and the body mass index. The comorbidity variables are represented by asthma, arthritis, cancer, diabetes, emphysema, heart disease, high blood pressure, stroke, angina, and high cholesterol. Also a variable for the total number of comorbidities is included to take differences in people’s physical health status into account, which is often done in health studies (Klabunde et al., 2000). The perceived mental health status variable indicates whether the individual evaluates its own mental health as strong or weak.

Some variables may require some more explanation on how they are composed and used in our model. First of all, the medical expenditures. We are dealing with a lot of zero observations regarding the expenditures, varying from 16.8 percent in 2000 to 21.9 percent in 2012. As the natural logarithm of zero is undefined, we added one to all the expenditures in our dataset in order to prevent a major loss of observations from individuals with no health expenses. The same trick is used for income, as we also take the logarithm of income. The second variable which needs to be clarified is the health insurance variable. For all individuals, monthly data on their type of insurance are available. As in our model insurance is a dummy variable, we prefer to have disjoint categories: we do not want overlap in the

(10)

10 different insurance categories caused by people who have had multiple types of insurance in the concerning year, because this would make the model obscure. Considering the fact that private insurance is our endogenous variable, we would like to have as many observations as possible of individuals with private insurance. Individuals are said to have private insurance in the concerning year if they have had private insurance in any of the months. Then, individuals are said to have insurance via the employer if they have had insurance via the employer in any of the months, with the restriction that they have never had private insurance in the concerning year. Individuals are said to have public insurance if they have had public insurance in any of the months, with the restriction that they have never had private insurance or insurance via the employer in the concerning year. An individual is considered to be uninsured if he or she has been uninsured all year.

Furthermore, some health-related variables need some more explanation. The variable heart disease includes coronary heart disease diagnosis, heart attack diagnosis and other heart diseases. The total number of comorbidity variable adds all comorbidities mentioned earlier, with the different heart diseases counted separately. For the body mass index variable (BMI), the individuals are divided in four groups: underweight, normal, overweight and obese. People who have a BMI below 18.5 are considered to be underweighted, a BMI between 18.5 and 25 indicates a normal weight, individuals with a BMI between 25 and 30 are considered to be overweighted and people with a BMI over 30 are considered obese (Centers for Disease Control and Prevention, 2006). Lastly, the mental health variable needs to be clarified. Individuals are asked about their perceived mental health status three times in the concerning year. They can rate their mental health status as 1-Excellent, 2-Very Good, 3-Good, 4-Fair, or 5-Poor. For the mental health variable, we have taken the average of the three responses. If the average is below 4, the individual’s mental health is presumed to be good. If it is 4 or higher, the individual’s mental health is presumed to be weak.

(11)

11

Table 1a: Summary Statistics of Study Population of 2012

Variable N % Variable N %

All 11517 100,0

Expenditures Income

No expenditures 2521 21,9 Less than $20,000 5512 47,9 Less than $1,000 3673 31,9 $20,000-$30,000 1716 14,9

$1,000-$2,000 1389 12,1 $30,000-$50,000 2078 18,0

$2,000-$5,000 1714 14,9 Over $50,000 2211 19,2

$5,000-$10,000 1031 9,0 Years of education

Over $10,000 1189 10,3 Less than high school 866 7,5

Insurance High school 5267 45,7

Private 4569 39,7 College or higher 5384 46,7

Via employer 1667 14,5 Comorbidities

Public 2770 24,1 Asthma 992 8,6

Uninsured 2511 21,8 Arthritis 2909 25,3

Gender Cancer 1080 9,4

Male 5318 46,2 Diabetes 1148 10,0

Female 6199 53,8 Emphysema 229 2,0

Age Heart disease 1460 12,7

18-34 3613 31,4 High blood pressure 3865 33,6

35-49 3069 26,6 Stroke 421 3,7

50-64 2866 24,9 Angina 292 2,5

65+ 1969 17,1 High cholesterol 3453 30,0

Race Number of comorbidities

White 8183 71,1 0 4910 42,6

Black 2308 20,0 1 2463 21,4

Asian 732 6,4 2 or more 4144 36,0

Other 294 2,6 Mental health

Marital Status Strong 11093 96,3

Married 5704 49,5 Weak 424 3,7

Other 5813 50,5 Currently smoking

Family size Yes 2030 17,6

1 or 2 5495 47,7 No 9487 82,4

3 or 4 4021 34,9 BMI index

5 or more 2001 17,4 Less than 18.5 192 1,7

Region 18.5-25 3698 32,1

Northeast 1782 15,5 25-30 4007 34,8

Midwest 2296 19,9 Over 30 3620 31,4

South 4498 39,1

West 2941 25,5

Table 1b: Descriptive Statistics of Study Population of 2012 Variable Mean Std. Dev. Min Max

Expenditures 4467,03 14808,77 0 537120 Age 46,06 17,60 18 85 Family size 2,97 1,65 1 12 Income 30298,07 32976,24 0 237974 Years of education 12,72 3,03 0 17 Number of comorbidities 1,44 1,79 0 11 BMI index 28,09 6,40 9,8 82

(12)

12 Not all variables are available in every year of the MEPS data, because MEPS has expended their survey in the course of the years. In 2008 and 2012, all variables mentioned before are available. In 2004, the variables cancer and high cholesterol are missing. Besides these variables, arthritis and body mass index are missing in 2000. Table 1a provides summary statistics and table 1b provides some descriptive statistics of the subsample of 2012. Summary statistics and descriptive statistics of the study population of the other years used in this paper can be found in Appendix A.

4. The Model

To examine the health care expenditures, we first use OLS to estimate a simple linear model. Let there be k explanatory variables and let n be the number of observations. The empirical model we are estimating is given by

E = Xβ + δI + ε, (1)

where E = the n x 1 vector of the logarithm of the expenditures;

X = the n x k matrix of explanatory variables;

β = the k x 1 vector of parameters;

I = the n x 1 vector of the variable private insurance (dummy);

δ = the parameter of private insurance; ε = the n x 1 vector of disturbances.

The first column of X consists of ones, as a constant term is included in the model. If we define b as the k x 1 vector of estimates of β, d as the estimated parameter of private insurance, and e as the n x 1 vector of residuals computed from the data, the estimated model can be written as

(13)

13 As we are dealing with an endogenous variable, OLS will not give consistent estimators. Therefore, we also use the instrumental variables (IV) estimation method to get consistent estimators. The IV estimators of β can be computed by the two-stage least squares method, abbreviated as 2SLS. Let k0 be the number of potential endogenous

regressors in X. In order to use the 2SLS method, we need m ≥ k0 instruments zi, which

satisfy the following conditions (Heij et al., 2004):

Cov[ziεi] = 0, i = 1, … , n, (3)

Cov[zixi] ≠ 0, i = 1, … , n. (4)

The first condition is called the exogeneity condition. The second is known as the relevance condition. In addition to these two conditions, the instruments are required to be not relevant in the explanation of E. When all three conditions are satisfied, the instruments are said to be valid. As private insurance coverage is the only variable that is considered to be endogenous, we need at least one valid instrumental variable. The instruments used in this paper will be discussed in the next section. Let Z be the matrix of instruments. The two-stage least squares estimation method is composed as follows (Heij et al., 2004).

Stage 1: Regress X on Z, with fitted values:

Ẋ = Z(Z’Z)-1Z ‘X. (5)

Stage 2: Regress E on Ẋ, with parameter estimates:

bIV = (Ẋ’ Ẋ)-1 Ẋ’E. (6)

To check for the presence of endogeneity, the test is used. The Hausman-test on exogeneity consists of the following three steps (Hausman, 1978):

Step 1: Regress the potential endogenous variable, private insurance, on all exogenous

variables xj and all instruments zj, and keep the vector of residuals ν:

(14)

14

Step 2: Regress E on all exogenous variables and the residuals vector v. In other words, we

use OLS in the following model:

E = Xω + αν + η, (8)

Step 3: Check if the coefficient of the vector of residuals v is significant by means of the

F-test statistic. The null hypothesis is that private insurance is exogenous. So if the F-test statistic is significant – that is, if it is greater than 10 – private insurance must be treated as endogenous. Note that we can only ascertain the presence of endogeneity. We cannot assign a theoretical “amount of endogeneity” to the result, based on the magnitude of the test statistic.

5. Estimation issues

As stated before, the goal of this paper is to study the endogeneity over time in the early 21st century. Since not all variables are available in every year, it is hard to compare the different models and outcomes of the Hausman-test if we estimate them optimally – that is, with all available variables. Therefore, two models are estimated for every year: one with as many of the explanatory variables mentioned in Section 3 as possible in the concerning year, hereafter denoted as the optimal model, and one with the variables available in 2000, hereafter denoted as the comparable model. We need the comparable model in order to compare the results afterwards, but keep in mind that this model might be misspecified as some variables are omitted. Evidently, both models are equal for the year 2000.

For the comparable models, the same instruments are used for every year in order to be able to compare the results afterwards. The instruments used in all comparable regressions are income and South region. For the optimal models, the best available instruments are selected in each separate year. In 2004, South region, smoking and BMI overweight and obese are used as instruments. South region, income, angina and number of comorbidities are used in 2008. Finally, in 2012, South and Northeast region are used as instruments, as well as smoking.

(15)

15 Recall that the three conditions for valid instrumental variables are that the instruments need to be 1) not relevant to health expenditures, 2) relevant to private insurance, and 3) exogenous. The first condition is satisfied if the instruments do not have significant coefficients in the regular OLS estimation with expenditures as the dependent variable. The second condition – the relevance condition – is satisfied if the instruments do have significant coefficients in OLS estimation with private insurance as the dependent variable, which is the first step of the Hausman test described in the previous section. An even stronger requirement is that the F-test statistic needs to be greater than 10. For the third condition – the exogeneity condition – Sargan’s (1958) test can be used. The null hypothesis of this test is that the instruments are exogenous, so we want an insignificant outcome in order to have exogenous instruments.

Table 2: Results Sargan test

Comparable Model Optimal model

Year χ²-value p-value χ²-value p-value

2000 0,12177 0,7271 0,12177 0,7271

2004 1,65121 0,1988 2,78358 0,4262 2008 2,94158 0,0863 1,49732 0,6829 2012 0,16399 0,6855 0,58297 0,7472

From the tables given in Appendix B, one can derive that the instruments in the optimal model of each year satisfy both the first and second condition at a 5% significance level. However, this is not the case in the comparable model. We can derive that both South region and income - the instruments used in the comparable model - satisfy both conditions at a 5% significance level in every year, except for the year 2004, where the first condition is not satisfied for income. From table 2 we see that – again using a 5% significance level - the Sargan test is satisfied for all years for both the optimal and the comparable model, indicating that the set of instruments used are exogenous in every model.

We can conclude that all instruments are valid in both the comparable and optimal model of every year, except that income does not satisfy the first condition in the comparable model of 2004. However, we cannot evade this problem, as we have to use the same instruments for every year in the comparable model in order to be able to compare the results. South region and income appeared to be the optimal instruments considering the fact that they are valid in all other years.

(16)

16

6. Results

6.1 Comparable model

Before we discuss the results, recall that the coefficients represent percentages, as we have taken the logarithm of the expenditures. Note that the estimated coefficients are approximately equal to the marginal effects. Table 3 provides the instrumental variable estimation results of the comparable model of 2012. The results of the comparable models of the other years can be found in Appendix C. As we can see from the tables, insured individuals have significantly more health expenditures than uninsured people, with all p-values of the coefficients being significant. The increase in level of expenditures due to having private insurance fluctuates between 160% and 288% in the years between 2000 and 2012. For insurance via the employer, this increase fluctuates between 174% and 296% and for public insurance it fluctuates between 187% and 274%.

Table 3: IV Estimation Results Comparable Model 2012: ln(expenditures)

Coefficient SE (robust) z-value P>z ME (percentage points)

Intercept 1,59 0,24 6,56 0 - Private insurance 2,82 0,34 8,19 0 281,96 Employer insurance 2,96 0,24 12,41 0 296,39 Public insurance 2,58 0,23 11,28 0 257,89 Male -1,09 0,05 -20,87 0 -109,00 Age 0,02 0,00 10,80 0 2,40 Race-White 0,07 0,17 0,44 0,661 7,25 Race-Black -0,60 0,17 -3,46 0,001 -59,59 Race-Asian -0,60 0,20 -3,07 0,002 -60,25 Married 0,38 0,06 6,17 0 38,34 Family size -0,24 0,02 -11,52 0 -23,52 Region-Northeast 0,11 0,07 1,43 0,154 10,61 Region-Midwest 0,33 0,07 4,84 0 32,90 Years of Education 0,09 0,01 6,42 0 8,85 Asthma 0,68 0,11 6,00 0 67,55 Diabetes 1,08 0,11 10,08 0 107,86 Emphysema 0,44 0,16 2,81 0,005 44,01 Heart disease 0,44 0,13 3,35 0,001 44,26

High blood pressure 0,84 0,10 8,46 0 83,85

Stroke 0,35 0,13 2,72 0,006 35,38

Angina -0,05 0,18 -0,26 0,792 -4,86

Number comorbidities 0,18 0,08 2,34 0,019 17,93

Mental weak 1,27 0,13 9,90 0 127,47

(17)

17 Over the years in the early 21st century, males spent on average about 110 percent less money on health care than females. Marital status has a highly significant impact on medical expenditures. Being married increases the expenditures by about 37 percent on average. Another important finding is that the marginal effect of number of years of education on medical expenditures is about 10 percent on average. The comorbidities all have a positive effect on the amount of health care expenditures, all having significant p-values, except for angina in 2000, 2008 and 2012. The mental health status has even more impact on the medical expenditures than the physical disorders. Individuals who perceive their mental health as weak, spend on average about 115 percent more on medical care than individuals who perceive themselves as mentally strong. A noteworthy coefficient is that of smoking. People who smoke tend to have lower expenditures on health care than people who do smoke, from around 4 percent in 2004 and 2012 to around 15 percent in 2000 and 2008. However, the coefficients of smoking in 2004 and 2012 are not significant, so these estimates fail to be credible.

Results of the Hausman-test on exogeneity are presented in table 5. Based on the Hausman-test, we can only ascertain the presence of endogeneity. The results of the Hausman-test therefore do not represent a numerical value in terms of the amount of endogeneity. In most of the years – that is in 2000, 2004 and 2012 – significant endogeneity is found between private insurance and health care expenditures. Accordingly, the endogeneity in the early 21st century seems to be stable. However, according to the Hausman-test, there was no significant endogeneity in 2008. This might be a result of a possible misspecification of the comparable model. In the next paragraph the results of the optimal models are discussed.

Table 5: Results Hausman-test

Comparable Model Optimal model

Year F-value p-value F-value p-value

2000 9,93228 0.0016 9,93228 0.0016

2004 13,9214 0.0002 5,40602 0,0121 2008 0,07894 0.7787 0,62928 0,4276 2012 5,39725 0.0202 9,52596 0.0020

(18)

18 6.2 Optimal model

Table 4 provides the instrumental variable estimation results of the optimal model of 2012. The results of the optimal models of 2004 and 2008 can be found in Appendix D. Estimation results of the optimal model of 2000 will not be discussed in this paragraph, as this model is the same as the comparable model discussed previously. Again, insured people spend significantly more on medical care than uninsured individuals. But in this model, the results are even more excessive. Between 2004 and 2012, individuals with private insurance spent between 199% and 453% more on medical care than the uninsured. For insurance via the employer, the increase in level of expenditures fluctuates between 228% and 397% and for public insurance it fluctuates between 211% and 353%. All p-values of the coefficients of the insurance variables are significant again.

Table 4: IV Estimation Results Optimal Model 2012: ln(expenditures)

Coefficient SE (robust) z-value P>z ME (percentage points)

Intercept 1,92 0,26 7,39 0 - Private insurance 4,53 0,88 5,14 0 453,47 Employer insurance 3,97 0,55 7,22 0 396,56 Public insurance 3,53 0,54 6,56 0 353,31 Male -1,00 0,06 -18,01 0 -100,24 Age 0,01 0,00 2,28 0,023 0,65 Race-White 0,13 0,17 0,78 0,435 13,31 Race-Black -0,54 0,18 -3,01 0,003 -53,57 Race-Asian -0,53 0,21 -2,56 0,01 -53,21 Married 0,31 0,08 4,00 0 30,88 Family size -0,20 0,02 -8,85 0 -19,78 Region-Midwest 0,18 0,08 2,33 0,02 18,21 ln(Income) -0,04 0,02 -1,91 0,056 -4,47 Years of Education 0,04 0,02 1,68 0,094 4,18 Asthma 0,54 0,12 4,51 0 54,13 Arthritis 0,71 0,10 6,96 0 70,96 Cancer 0,42 0,11 3,68 0 42,02 Diabetes 0,98 0,11 8,69 0 97,83 Emphysema 0,29 0,16 1,81 0,07 28,52 Heart disease 0,28 0,14 2,05 0,041 28,15

High blood pressure 0,55 0,11 5,20 0 55,15

Stroke 0,31 0,14 2,21 0,027 30,56 Angina 0,12 0,20 0,60 0,546 11,84 High cholesterol 0,41 0,11 3,80 0 40,94 Number comorbidities 0,15 0,08 1,94 0,052 15,46 Mental weak 1,14 0,13 8,65 0 113,77 BMI-Underweight 0,34 0,21 1,60 0,109 33,82 BMI-Overweight -0,01 0,07 -0,19 0,848 -1,27 BMI-Obese 0,00 0,07 0,07 0,948 0,46

(19)

19 Note that most of the coefficients are approximately the same order of magnitude as those in the comparable model and that the estimated coefficients are again approximately equal to the marginal effects. In the years between 2004 and 2012, males had on average about 101 percent less medical expenses than females, which is 9 percent less than in the comparable model. Being married increases the level of expenditures by about 33 percent on average, which is 4 percent less than in the comparable model. In the year 2012, the coefficient of years of education is not significant. From the other years we can derive that the marginal effect of number of years of education on medical expenditures is around 10 percent on average, just like in the comparable model. The comorbidities all have a positive effect on the amount of health expenditures again, all having significant p-values, except for emphysema in 2012 and angina in 2004 and 2012. Within the group of comorbidities, the most extreme increase in level of expenditures is caused by diabetes, with an average increase of 100 percent. The mental health status also has a high impact on health care expenditures. Mentally weak individuals spend on average around 104 percent more on health care than individuals who perceive their mental health as strong, which is 11 percent less than in the comparable model. As the variable smoking is used as an instrument in the optimal models of 2004 and 2012, this variable is only adopted in the model of 2008. With the p-value being significant, smokers spent 14 percent less on health care than non-smokers in 2008. Another interesting finding is the fact that all BMI variables fail to be significant in every year. Therefore, medical expenses cannot be explained by BMI.

Before we discuss the results of the Hausman-test, recall that in the year 2000 the optimal model equals the comparable model. Table 5 shows that the results of Hausman-test on exogeneity in the optimal model follow approximately the same pattern as the results of the comparable model. In 2000, 2004 and 2012, significant endogeneity between private

insurance and medical expenditures was ascertained. But just like in the comparable model, no endogeneity was found in the year 2008. Even though in most of the years there was significant proof of endogeneity between private insurance and health care expenditures, we can conclude that we cannot speak of entirely stable endogeneity in the early 21st century.

(20)

20

7. Conclusion

This paper studies how the endogeneity between private health insurance and medical expenditures developed over time in the period from 2000 to 2012. We study this endogeneity using the Hausman-test, analysing two models for every year: a comparable model and an optimal model. The merit of the comparable model is that we can compare the results afterwards. However, the comparable model may be misspecified considering the fact that unlike the optimal model, it does not contain all available variables of the concerning year. Although in most years the comparable model and optimal model are very much alike, both models yield quite different estimates. On the contrary, the results of the Hausman-test are very similar for both models in all different years, even though different instruments are used.

In summary, we have seen that in most of the years in the early 21st century that we have studied, there was significant proof of endogeneity between health insurance and expenditures. In both the comparable and optimal model, endogeneity was ascertained in 2000, 2004, and 2012. However, both models fail to find significant evidence for endogeneity in the year 2008.

Some marginal effects are worth noting. In the comparable model, the increase in level of expenditures due to having any insurance fluctuates between 160% and 296% in the years between 2000 and 2012. In the optimal model, this increase even fluctuates between 199% and 453%. Married individuals spend on average 35 percent more on medical care than non-married individuals. The number of years of education an individual has had also plays a significant role in the level of expenditures. The marginal effect of an extra year of education is on average 10 percent. Furthermore, besides angina, all comorbidities have a positive effect on an individual’s medical expenditures, with all p-values being significant.

There are some future research directions that I want to point out. First, it might be interesting to analyse the endogeneity between insurance and medical expenditures in the years between 2000 and 2012 that have not been analysed in this paper. In that way, a much more detailed representation of the development of the endogeneity over time can be given. Second, it would be useful to take a critical look at how the individuals are divided over the different insurance groups to prevent overlap in the different categories. Maybe

(21)

21 the results of the Hausman-test will be different if the groups are composed differently. A possibility is to place an individual in the insurance category in which it has been for the longest period in the concerning year. Third, a limitation of MEPS is that it does not provide information about different types of insurers, for example, in terms of deductibles and copayments. It would be useful to distinguish differences in generosity of insurance plans, as this influences the total amount of health care expenditures.

(22)

22

References

Akerlof, G. A. (1970). The market for" lemons": Quality uncertainty and the market mechanism. The

quarterly journal of economics, 488-500.

Albert Ma, C. T., & Riordan, M. H. (2002). Health insurance, moral hazard, and managed care. Journal

of Economics & Management Strategy, 11(1), 81-107.

Browne, M. J. (1992). Evidence of adverse selection in the individual health insurance market. Journal

of Risk and Insurance, 13-33.

Cameron, A. C., Trivedi, P. K., Milne, F., & Piggott, J. (1988). A microeconometric model of the demand for health care and health insurance in Australia. The Review of Economic Studies, 55(1), 85-106.

Cardon, J. H., & Hendel, I. (2001). Asymmetric information in health insurance: evidence from the National Medical Expenditure Survey. RAND Journal of Economics, 408-427.

Centers for Disease Control and Prevention (CDC). (2006). Behavioral Risk Factor Surveillance System Survey. BRFSS), Atlanta, Georgia: US Department of Health and Human Services, Centers

for Disease Control and Prevention.

Cutler, D. M., & Reber, S. (1998). Paying for health insurance: the tradeoff between competition and

adverse selection (No. w5796). National Bureau of Economic Research.

Cutler, D. M., & Zeckhauser, R. J. (1998). Adverse Selection in Health Insurance. In Frontiers in

Health Policy Research, 1(1) (pp. 1-32). MIT.

Duan, N., Manning, W. G., Morris, C. N., & Newhouse, J. P. (1983). A comparison of alternative models for the demand for medical care. Journal of business & economic statistics, 1(2), 115-126. Dunn, A. (2013). Health Insurance and the Demand for Medical Care: Instrumental Variable Estimates using Health Insurer Claims. In Health & Healthcare in America: From Economics to Policy. Ashecon. Fronstin, P. (2012). Sources of health insurance and characteristics of the uninsured: analysis of the March 2012 Current Population Survey. EBRI Issue Brief, 376, 1-35.

Hausman, J. A. (1978). Specification tests in econometrics. Econometrica: Journal of the Econometric

Society, 1251-1271.

Heckman, J. J. (1976). The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models. In Annals of Economic and Social

Measurement, 5(4), 475-492.

Heckman, J. J. (1979). Sample selection bias as a specification error. Econometrica: Journal of the

(23)

23

Heij, C., De Boer, P., Franses, P. H., Kloek, T., & Van Dijk, H. K. (2004).Econometric methods with

applications in business and economics. Oxford University Press.

Klabunde, C. N., Potosky, A. L., Legler, J. M., & Warren, J. L. (2000). Development of a comorbidity index using physician claims data. Journal of clinical epidemiology, 53(12), 1258-1267.

Kowalski, A. E. (2009). Censored quantile instrumental variable estimates of the price elasticity of

expenditure on medical care (No. w15085). National Bureau of Economic Research.

Manning, W. G., & Marquis, M. S. (1996). Health insurance: the tradeoff between risk pooling and moral hazard. Journal of health economics, 15(5), 609-639.

Manning, W. G., Newhouse, J. P., Duan, N., Keeler, E. B., Leibowitz, A., & Marquis, M. S. (1987). Health insurance and the demand for medical care: evidence from a randomized experiment. The

American economic review, 251-277.

Newhouse, J. P. & Rand Corporation. Insurance Experiment Group (Eds.). (1993). Free for all?:

lessons from the RAND health insurance experiment. Harvard University Press.

Rosett, R. N., & Huang, L. F. (1973). The effect of health insurance on the demand for medical care. The Journal of Political Economy, 281-305.

Sargan, J. D. (1958). The estimation of economic relationships using instrumental variables. Econometrica: Journal of the Econometric Society, 393-415.

Shen, C. (2013). Determinants of Health Care Decisions: Insurance, Utilization, and Expenditures. The

Review of Economics and Statistics, 95(1), 142-153.

Tobin, J. (1958). Estimation of relationships for limited dependent variables.Econometrica: journal of

the Econometric Society, 24-36.

(24)

24

Appendix A – Summary Statistics

Table A1.1: Summary Statistics of Study Population of 2000

Variable N % All 14641 Expenditures No expenditures 2459 16,8 Less than $1,000 6092 41,6 $1,000-$2,000 2183 14,9 $2,000-$5,000 2176 14,9 $5,000-$10,000 948 6,5 Over $10,000 783 5,3 Insurance Private 7372 50,4 Via employer 2671 18,2 Public 2254 15,4 Uninsured 2344 16,0 Gender Male 6741 46,0 Female 7900 54,0 Age 18-34 4284 29,3 35-49 4782 32,7 50-64 3092 21,1 65+ 2483 17,0 Race White 12140 82,9 Black 2011 13,7 Asian 369 2,5 Other 121 0,8 Marital Status Married 8704 59,4 Other 5937 40,6 Family size 1 or 2 6980 47,7 3 or 4 5282 36,1 5 or more 2379 16,2 Region Northeast 2320 15,8 Midwest 3134 21,4 South 5649 38,6 West 3538 24,2 Income Less than $20,000 7260 49,6 $20,000-$30,000 2458 16,8 $30,000-$50,000 2922 20,0 Over $50,000 2001 13,7 Years of education

Less than high school 1549 10,6

High school 6951 47,5 College or higher 6141 41,9 Comorbidities Asthma 1261 8,6 Diabetes 1049 7,2 Emphysema 201 1,4 Heart disease 1358 9,3

High blood pressure 3456 23,6

Stroke 346 2,4 Angina 366 2,5 Number of comorbidities 0 9375 64,0 1 3380 23,1 2 or more 1886 12,9 Mental health Strong 14254 97,4 Weak 387 2,6 Currently smoking Yes 3254 22,2 No 11387 77,8

(25)

25

Variable N % All 19172 Expenditures No expenditures 3373 17,6 Less than $1,000 6325 33,0 $1,000-$2,000 2453 12,8 $2,000-$5,000 3446 18,0 $5,000-$10,000 1894 9,9 Over $10,000 1681 8,8 Insurance Private 8468 44,2 Via employer 3051 15,9 Public 3949 20,6 Uninsured 3704 19,3 Gender Male 8615 44,9 Female 10557 55,1 Age 18-34 6030 31,5 35-49 5818 30,3 50-64 4245 22,1 65+ 3079 16,1 Race White 15114 78,8 Black 2709 14,1 Asian 810 4,2 Other 539 2,8 Marital Status Married 10603 55,3 Other 8569 44,7 Family size 1 or 2 9292 48,5 3 or 4 6651 34,7 5 or more 3229 16,8 Region Northeast 2818 14,7 Midwest 3840 20,0 South 7829 40,8 West 4685 24,4 Income Less than $20,000 9840 51,3 $20,000-$30,000 3036 15,8 $30,000-$50,000 3472 18,1 Over $50,000 2824 14,7 Years of education

High school 9231 48,1 College or higher 7924 41,3 Comorbidities Asthma 1780 9,3 Arthritis 4137 21,6 Diabetes 1574 8,2 Emphysema 304 1,6 Heart disease 1819 9,5

Stroke 547 2,9 Angina 465 2,4 Number of comorbidities 0 10611 55,3 1 4367 22,8 2 or more 4194 21,9 Mental health Strong 18466 96,3 Weak 706 3,7 Currently smoking Yes 4096 21,4 No 15076 78,6 BMI index Less than 18.5 371 1,9 18.5-25 6588 34,4 25-30 6719 35,0 Over 30 5494 28,7

(26)

26

Variable N % All 18919 100,0 Expenditures No expenditures 3506 18,5 Less than $1,000 6071 32,1 $1,000-$2,000 2400 12,7 $2,000-$5,000 3210 17,0 $5,000-$10,000 1854 9,8 Over $10,000 1878 9,9 Insurance Private 8138 43,0 Via employer 2885 15,2 Public 4009 21,2 Uninsured 3887 20,5 Gender Male 8668 45,8 Female 10251 54,2 Age 18-34 5966 31,5 35-49 5534 29,3 50-64 4532 24,0 65+ 2887 15,3 Race White 13699 72,4 Black 3462 18,3 Asian 1204 6,4 Other 554 2,9 Marital Status Married 10240 54,1 Other 8679 45,9 Family size 1 or 2 8753 46,3 3 or 4 6846 36,2 5 or more 3320 17,5 Region Northeast 2917 15,4 Midwest 3741 19,8 South 7308 38,6 West 4953 26,2 Income Less than $20,000 8855 46,8 $20,000-$30,000 3068 16,2 $30,000-$50,000 3601 19,0 Over $50,000 3395 17,9 Years of education

High school 8689 45,9 College or higher 8540 45,1 Comorbidities Asthma 1682 8,9 Arthritis 4452 23,5 Cancer 1600 8,5 Diabetes 1979 10,5 Emphysema 409 2,2 Heart disease 2405 12,7

Stroke 688 3,6 Angina 555 2,9 High cholesterol 5733 30,3 Number of comorbidities 0 8112 42,9 1 4040 21,4 2 or more 6767 35,8 Mental health Strong 18317 96,8 Weak 602 3,2 Currently smoking Yes 3678 19,4 No 15241 80,6 BMI index Less than 18.5 333 1,8 18.5-25 6176 32,6 25-30 6618 35,0 Over 30 5792 30,6

(27)

27 Table A1.2: Descriptive Statistics of Study Population of 2000

Variable Mean Std. Dev. Min Max

Expenditures 2502,60 6792,50 0 275730 Age 45,84 17,38 18 90 Family size 2,97 1,59 1 14 Income 26389,55 25570,46 0 313984 Years of education 12,36 3,22 0 17 Number of comorbidities 0,58 0,98 0 8

Table A2.2: Descriptive Statistics of Study Population of 2004

Expenditures 3711,74 10022,60 0 440524 Age 45,41 17,37 18 85 Family size 2,96 1,64 1 13 Income 26695,78 28162,80 0 437861 Years of education 12,29 3,25 0 17 Number of comorbidities 0,85 1,25 0 10 BMI index 27,60 6,01 12,8 81,4

Table A3.2: Descriptive Statistics of Study Population of 2008

Expenditures 3967,62 9688,78 0 264510 Age 45,31 17,32 18 85 Family size 3,04 1,69 1 14 Income 29797,01 31476,24 0 221110 Years of education 12,58 3,10 0 17 Number of comorbidities 1,42 1,79 0 12 BMI index 27,96 6,32 9,4 82,1

(28)

28

Appendix B – Auxiliary Regressions

Table B1.1: OLS Estimation Results Comparable Model 2000: ln(expenditures)

Coefficient SE (robust) t-value P>t

Intercept 2,65 0,28 9,56 0 Private insurance 1,44 0,07 20,13 0 Employer insurance 1,63 0,08 20,34 0 Public insurance 1,76 0,08 21,19 0 Male -1,15 0,04 -26,97 0 Age 0,02 0,00 14,65 0 Race-White -0,36 0,23 -1,55 0,121 Race-Black -0,95 0,24 -3,99 0 Race-Asian -0,94 0,27 -3,49 0 Married 0,37 0,05 7,74 0 Family size -0,17 0,02 -10,48 0 Region-Northeast 0,22 0,07 3,34 0,001 Region-Midwest 0,28 0,06 4,61 0 Region-South 0,01 0,05 0,21 0,835 ln(Income) 0,00 0,01 0,50 0,617 Years of Education 0,11 0,01 15,04 0 Asthma 0,44 0,11 4,05 0 Diabetes 0,84 0,11 7,77 0 Emphysema 0,52 0,15 3,49 0 Heart disease 0,59 0,13 4,61 0

High blood pressure 0,76 0,10 7,51 0

Stroke 0,28 0,14 2,08 0,037

Angina 0,08 0,16 0,48 0,634

Number comorbidities 0,27 0,09 3,05 0,002

Mental weak 1,21 0,11 11,26 0

Smoking -0,17 0,05 -3,16 0,002

Table B1.2: OLS Estimation Results Comparable Model 2000: Private insurance

Intercept 0,09 0,04 2,43 0,015 Employer insurance -0,73 0,01 -136,85 0 Public insurance -0,71 0,01 -114,28 0 Male 0,00 0,01 0,15 0,879 Age 0,00 0,00 12,89 0 Race-White -0,01 0,03 -0,18 0,859 Race-Black 0,01 0,03 0,41 0,684 Race-Asian -0,01 0,04 -0,30 0,765 Married 0,06 0,01 8,90 0 Family size -0,02 0,00 -7,21 0 Region-Northeast 0,04 0,01 5,21 0 Region-Midwest 0,06 0,01 8,19 0 Region-South 0,01 0,01 2,03 0,032 ln(Income) 0,03 0,00 24,82 0 Years of Education 0,02 0,00 21,47 0 Asthma 0,02 0,01 1,71 0,088 Diabetes 0,01 0,01 0,73 0,467 Emphysema 0,03 0,02 1,51 0,132 Heart disease 0,01 0,02 0,53 0,593 High blood pressure 0,02 0,01 1,87 0,062

Stroke -0,01 0,02 -0,43 0,669

Angina 0,00 0,02 -0,02 0,982

Number comorbidities 0,00 0,01 -0,21 0,836 Mental weak -0,02 0,02 -1,15 0,249

(29)

29 Table B2.1: OLS Estimation Results Comparable Model 2004: ln(expenditures)

Intercept 2,04 0,17 12,03 0 Private insurance 1,93 0,06 31,16 0 Employer insurance 2,13 0,07 30,58 0 Public insurance 2,14 0,07 31,54 0 Male -1,10 0,04 -28,61 0 Age 0,02 0,00 16,87 0 Race-White -0,03 0,11 -0,27 0,79 Race-Black -0,48 0,12 -4,03 0 Race-Asian -0,52 0,14 -3,68 0 Married 0,40 0,04 9,22 0 Family size -0,22 0,01 -15,14 0 Region-Northeast 0,39 0,06 6,59 0 Region-Midwest 0,42 0,05 7,74 0 Region-South 0,04 0,05 0,84 0,399 ln(Income) 0,03 0,01 3,85 0 Years of Education 0,11 0,01 16,75 0 Asthma 0,74 0,09 8,15 0 Diabetes 1,09 0,09 12,27 0 Emphysema 0,45 0,14 3,24 0,001 Heart disease 0,81 0,10 7,85 0

Stroke 0,37 0,11 3,23 0,001

Angina 0,31 0,13 2,29 0,022

Number comorbidities 0,00 0,07 -0,02 0,988

Mental weak 1,02 0,09 10,81 0

Smoking -0,07 0,05 -1,48 0,14

Intercept 0,03 0,02 1,24 0,216 Employer insurance -0,69 0,00 -139,30 0 Public insurance -0,63 0,01 -121,11 0 Male -0,01 0,01 -1,15 0,252 Age 0,00 0,00 16,20 0 Race-White -0,01 0,02 -0,75 0,455 Race-Black 0,01 0,02 0,52 0,604 Race-Asian 0,01 0,02 0,52 0,6 Married 0,06 0,01 10,78 0 Family size -0,01 0,00 -5,54 0 Region-Northeast 0,05 0,01 7,48 0 Region-Midwest 0,04 0,01 5,39 0 Region-South -0,02 0,01 -3,50 0 ln(Income) 0,03 0,00 32,90 0 Years of Education 0,02 0,00 26,18 0 Asthma 0,03 0,01 2,63 0,009 Diabetes 0,03 0,01 2,17 0,03 Emphysema 0,02 0,02 1,27 0,203 Heart disease 0,02 0,02 1,30 0,193 High blood pressure 0,03 0,01 2,67 0,008

Stroke -0,01 0,02 -0,56 0,574

Angina -0,01 0,02 -0,37 0,713

Number comorbidities -0,01 0,01 -0,69 0,493 Mental weak -0,01 0,01 -0,77 0,441

(30)

30 Table B2.3: OLS Estimation Results Optimal Model 2004: ln(expenditures)

Intercept 2,05 0,17 12,05 0 Private insurance 1,92 0,06 31,08 0 Employer insurance 2,11 0,07 30,37 0 Public insurance 2,10 0,07 31,00 0 Male -1,03 0,04 -26,74 0 Age 0,02 0,00 12,60 0 Race-White -0,01 0,11 -0,09 0,928 Race-Black -0,45 0,12 -3,84 0 Race-Asian -0,44 0,14 -3,10 0,002 Married 0,40 0,04 9,42 0 Family size -0,21 0,01 -14,47 0 Region-Northeast 0,37 0,06 6,27 0 Region-Midwest 0,39 0,05 7,21 0 Region-South 0,02 0,05 0,31 0,754 ln(Income) 0,03 0,01 4,04 0 Years of Education 0,11 0,01 17,21 0 Asthma 0,65 0,09 7,11 0 Arthritis 0,71 0,08 8,51 0 Diabetes 1,06 0,09 11,75 0 Emphysema 0,39 0,14 2,76 0,006 Heart disease 0,73 0,10 7,08 0

Stroke 0,33 0,11 2,86 0,004 Angina 0,23 0,13 1,70 0,09 Number comorbidities 0,01 0,07 0,12 0,901 Mental weak 0,95 0,09 10,07 0 Smoking -0,08 0,05 -1,68 0,093 BMI-Underweight 0,16 0,14 1,14 0,253 BMI-Overweight 0,00 0,04 -0,04 0,967 BMI-Obese 0,06 0,05 1,31 0,189

Table B2.4: OLS Estimation Results Optimal Model 2004: Private insurance

Intercept 0,02 0,02 0,86 0,392 Employer insurance -0,69 0,00 -138,30 0 Public insurance -0,63 0,01 -121,12 0 Male -0,01 0,01 -1,06 0,288 Age 0,00 0,00 14,92 0 Race-White -0,01 0,02 -0,68 0,498 Race-Black 0,01 0,02 0,53 0,599 Race-Asian 0,02 0,02 0,84 0,401 Married 0,06 0,01 10,64 0 Family size -0,01 0,00 -5,49 0 Region-Northeast 0,05 0,01 7,44 0 Region-Midwest 0,04 0,01 5,28 0 Region-South -0,02 0,01 -3,57 0 ln(Income) 0,03 0,00 32,81 0 Years of Education 0,02 0,00 26,34 0 Asthma 0,03 0,01 2,37 0,018 Arthritis 0,02 0,01 1,84 0,066 Diabetes 0,03 0,01 1,92 0,055 Emphysema 0,02 0,02 1,25 0,212 Heart disease 0,02 0,02 1,20 0,229 High blood pressure 0,03 0,01 2,27 0,023

Stroke -0,01 0,02 -0,58 0,565 Angina -0,01 0,02 -0,46 0,646 Number comorbidities -0,01 0,01 -0,64 0,52 Mental weak -0,01 0,01 -0,90 0,367 Smoking -0,04 0,01 -5,95 0 BMI-Underweight 0,01 0,02 0,47 0,64 BMI-Overweight 0,01 0,01 2,10 0,036 BMI-Obese 0,02 0,01 2,52 0,012

(31)

Intercept 1,65 0,18 9,12 0 Private insurance 1,83 0,06 29,01 0 Employer insurance 2,21 0,07 31,02 0 Public insurance 2,09 0,07 30,52 0 Male -1,04 0,04 -26,31 0 Age 0,02 0,00 16,85 0 Race-White 0,10 0,12 0,80 0,422 Race-Black -0,46 0,13 -3,52 0 Race-Asian -0,51 0,14 -3,57 0 Married 0,39 0,04 8,68 0 Family size -0,18 0,01 -12,87 0 Region-Northeast 0,20 0,06 3,27 0,001 Region-Midwest 0,41 0,06 7,18 0 Region-South 0,08 0,05 1,63 0,103 ln(Income) 0,00 0,01 0,62 0,535 Years of Education 0,13 0,01 18,98 0 Asthma 0,74 0,08 8,71 0 Diabetes 1,14 0,08 14,09 0 Emphysema 0,40 0,12 3,38 0,001 Heart disease 0,65 0,10 6,36 0

Stroke 0,33 0,11 3,10 0,002

Angina 0,04 0,13 0,28 0,783

Mental weak 1,11 0,10 11,46 0

Smoking -0,14 0,05 -2,75 0,006

Intercept 0,02 0,02 0,71 0,477 Employer insurance -0,66 0,01 -130,60 0 Public insurance -0,62 0,01 -117,67 0 Male -0,01 0,01 -1,04 0,298 Age 0,00 0,00 11,63 0 Race-White -0,01 0,01 -0,72 0,472 Race-Black 0,02 0,02 1,38 0,168 Race-Asian 0,01 0,02 0,52 0,602 Married 0,06 0,01 9,88 0 Family size -0,02 0,00 -8,30 0 Region-Northeast 0,02 0,01 2,74 0,006 Region-Midwest 0,04 0,01 4,86 0 Region-South -0,04 0,01 -5,36 0 ln(Income) 0,03 0,00 31,80 0 Years of Education 0,03 0,00 30,02 0 Asthma -0,01 0,01 -0,54 0,589 Diabetes -0,01 0,01 -0,78 0,433 Emphysema 0,00 0,02 -0,24 0,809 Heart disease -0,02 0,01 -1,37 0,17 High blood pressure 0,00 0,01 -0,02 0,98

Stroke -0,03 0,01 -2,08 0,037

Angina -0,04 0,02 -2,51 0,012

Number comorbidities 0,02 0,01 2,75 0,006 Mental weak -0,01 0,01 -0,64 0,524

(32)

Intercept 2,06 0,18 11,38 0 Private insurance 1,79 0,06 28,68 0 Employer insurance 2,15 0,07 30,44 0 Public insurance 1,99 0,07 29,31 0 Male -1,00 0,04 -25,23 0 Age 0,01 0,00 7,54 0 Race-White 0,07 0,12 0,56 0,578 Race-Black -0,41 0,13 -3,23 0,001 Race-Asian -0,48 0,14 -3,42 0,001 Married 0,39 0,04 8,85 0 Family size -0,17 0,01 -11,96 0 Region-Northeast 0,19 0,06 3,17 0,002 Region-Midwest 0,39 0,06 6,83 0 Region-South 0,05 0,05 1,05 0,295 ln(Income) 0,01 0,01 1,02 0,308 Years of Education 0,12 0,01 17,94 0 Asthma 0,71 0,08 8,50 0 Arthritis 0,66 0,07 8,82 0 Cancer 0,67 0,08 8,33 0 Diabetes 1,00 0,08 12,47 0 Emphysema 0,34 0,12 2,93 0,003 Heart disease 0,59 0,10 5,86 0

Stroke 0,31 0,11 2,88 0,004 Angina -0,02 0,13 -0,15 0,885 High cholesterol 0,68 0,07 9,23 0 Number comorbidities 0,00 0,06 0,07 0,946 Mental weak 1,00 0,10 10,49 0 Smoking -0,16 0,05 -3,04 0,002 BMI-Underweight 0,00 0,15 -0,02 0,983 BMI-Overweight -0,09 0,05 -1,83 0,067 BMI-Obese -0,07 0,05 -1,43 0,152

Intercept 0,02 0,02 0,99 0,32 Employer insurance -0,66 0,01 -130,18 0 Public insurance -0,62 0,01 -117,69 0 Male -0,01 0,01 -1,03 0,304 Age 0,00 0,00 9,11 0 Race-White -0,01 0,01 -0,77 0,441 Race-Black 0,02 0,02 1,43 0,153 Race-Asian 0,01 0,02 0,63 0,526 Married 0,06 0,01 9,63 0 Family size -0,02 0,00 -8,13 0 Region-Northeast 0,02 0,01 2,76 0,006 Region-Midwest 0,04 0,01 4,82 0 Region-South -0,04 0,01 -5,42 0 ln(Income) 0,03 0,00 31,70 0 Years of Education 0,03 0,00 29,63 0 Asthma -0,01 0,01 -0,52 0,606 Arthritis -0,02 0,01 -1,53 0,126 Cancer 0,00 0,01 -0,31 0,76 Diabetes -0,02 0,01 -1,34 0,181 Emphysema 0,00 0,02 -0,15 0,883 Heart disease -0,02 0,01 -1,40 0,162 High blood pressure -0,01 0,01 -0,66 0,509

Stroke -0,03 0,01 -2,00 0,045 Angina -0,04 0,02 -2,55 0,011 High cholesterol 0,01 0,01 1,28 0,199 Number comorbidities 0,02 0,01 2,44 0,015 Mental weak -0,01 0,01 -0,80 0,423 Smoking -0,05 0,01 -6,81 0 BMI-Underweight -0,01 0,02 -0,76 0,445 BMI-Overweight 0,00 0,01 -0,13 0,899 BMI-Obese 0,01 0,01 1,07 0,283

(33)

Intercept 1,60 0,25 6,47 0 Private insurance 2,01 0,08 25,03 0 Employer insurance 2,46 0,09 26,18 0 Public insurance 2,08 0,09 24,00 0 Male -1,10 0,05 -21,12 0 Age 0,03 0,00 13,41 0 Race-White 0,06 0,16 0,39 0,697 Race-Black -0,56 0,17 -3,27 0,001 Race-Asian -0,58 0,20 -2,97 0,003 Married 0,43 0,06 7,33 0 Family size -0,25 0,02 -12,47 0 Region-Northeast 0,11 0,08 1,31 0,189 Region-Midwest 0,34 0,07 4,56 0 Region-South -0,07 0,07 -1,04 0,298 ln(Income) 0,02 0,01 1,72 0,054 Years of Education 0,11 0,01 11,73 0 Asthma 0,70 0,11 6,24 0 Diabetes 1,08 0,11 10,17 0 Emphysema 0,43 0,16 2,73 0,006 Heart disease 0,46 0,13 3,48 0,001 High blood pressure 0,87 0,10 8,94 0

Stroke 0,33 0,13 2,54 0,011

Angina -0,06 0,18 -0,32 0,747

Mental weak 1,27 0,13 9,85 0

Smoking -0,06 0,07 -0,81 0,417

Coefficient SE (robust) t-value P>t Intercept -0,01 0,03 -0,28 0,782 Employer insurance -0,62 0,01 -91,35 0 Public insurance -0,61 0,01 -91,78 0 Male -0,01 0,01 -1,82 0,069 Age 0,00 0,00 10,74 0 Race-White -0,01 0,02 -0,65 0,515 Race-Black 0,03 0,02 1,37 0,17 Race-Asian 0,03 0,02 1,31 0,189 Married 0,06 0,01 7,50 0 Family size -0,01 0,00 -4,85 0 Region-Northeast 0,02 0,01 2,33 0,02 Region-Midwest 0,03 0,01 3,59 0 Region-South -0,05 0,01 -6,20 0 ln(Income) 0,02 0,00 24,88 0 Years of Education 0,03 0,00 22,13 0 Asthma 0,03 0,01 2,00 0,045 Diabetes 0,01 0,01 0,36 0,716 Emphysema -0,01 0,02 -0,54 0,592 Heart disease 0,02 0,02 1,15 0,251 High blood pressure 0,04 0,01 3,30 0,001

Stroke -0,03 0,02 -1,59 0,112

Angina -0,01 0,02 -0,54 0,592

Number comorbidities 0,00 0,01 -0,05 0,961 Mental weak -0,01 0,01 -0,57 0,566

(34)

Intercept 1,91 0,25 7,71 0 Private insurance 1,93 0,08 24,27 0 Employer insurance 2,34 0,09 25,09 0 Public insurance 1,94 0,09 22,53 0 Male -1,03 0,05 -19,71 0 Age 0,01 0,00 5,55 0 Race-White 0,10 0,16 0,59 0,554 Race-Black -0,46 0,17 -2,70 0,007 Race-Asian -0,43 0,19 -2,24 0,025 Married 0,45 0,06 7,80 0 Family size -0,23 0,02 -11,61 0 Region-Northeast 0,11 0,08 1,37 0,171 Region-Midwest 0,30 0,07 4,02 0 Region-South -0,11 0,07 -1,59 0,112 ln(Income) 0,02 0,01 2,12 0,034 Years of Education 0,11 0,01 11,75 0 Asthma 0,61 0,11 5,44 0 Arthritis 0,77 0,10 7,95 0 Cancer 0,51 0,11 4,83 0 Diabetes 0,97 0,11 9,09 0 Emphysema 0,25 0,16 1,61 0,108 Heart disease 0,32 0,13 2,45 0,014 High blood pressure 0,63 0,10 6,42 0

Stroke 0,23 0,13 1,74 0,081 Angina -0,16 0,18 -0,87 0,382 High cholesterol 0,52 0,10 5,35 0 Number comorbidities 0,15 0,08 1,90 0,058 Mental weak 1,10 0,13 8,63 0 Smoking -0,09 0,07 -1,20 0,23 BMI-Underweight 0,41 0,20 2,04 0,042 BMI-Overweight 0,00 0,06 -0,07 0,943 BMI-Obese 0,03 0,07 0,50 0,619

Coefficient SE (robust) t-value P>t Intercept 0,00 0,03 -0,03 0,977 Employer insurance -0,62 0,01 -91,53 0 Public insurance -0,61 0,01 -92,46 0 Male -0,01 0,01 -1,58 0,114 Age 0,00 0,00 7,69 0 Race-White -0,01 0,02 -0,60 0,551 Race-Black 0,03 0,02 1,54 0,124 Race-Asian 0,04 0,02 1,55 0,122 Married 0,06 0,01 7,45 0 Family size -0,01 0,00 -4,58 0 Region-Northeast 0,02 0,01 2,36 0,018 Region-Midwest 0,03 0,01 3,41 0,001 Region-South -0,06 0,01 -6,32 0 ln(Income) 0,02 0,00 24,77 0 Years of Education 0,03 0,00 22,04 0 Asthma 0,03 0,01 1,81 0,07 Arthritis 0,02 0,01 1,76 0,079 Cancer 0,04 0,01 2,51 0,012 Diabetes 0,00 0,01 -0,13 0,9 Emphysema -0,02 0,02 -0,79 0,432 Heart disease 0,02 0,02 0,89 0,373 High blood pressure 0,03 0,01 2,32 0,02

Stroke -0,03 0,02 -1,71 0,087 Angina -0,02 0,02 -0,64 0,524 High cholesterol 0,04 0,01 3,24 0,001 Number comorbidities 0,00 0,01 -0,25 0,804 Mental weak -0,01 0,01 -1,01 0,313 Smoking -0,02 0,01 -2,33 0,02 BMI-Underweight 0,03 0,02 1,20 0,23 BMI-Overweight 0,00 0,01 0,41 0,684 BMI-Obese 0,01 0,01 1,33 0,184