The stability of the relation between individual
health expenditures and health insurance
An Instrumental Variable analysis
Bob de Vries
1
Introduction
A recent publication of Shen (2013) studies the relationship in health care expenditures and whether or not having a health care insurance in the United States of America. A problem that Shen (2013) identifies, is that buying health care insurance depends on ones expected health costs and thus on ones current health status. This means that some-one who feels or actually is unhealthy is more likely to take insurance than somesome-one who feels or is healthy. This is called adverse selection. Furthermore, having insurance or not influences the amount of health care expenditures. This is called moral hazard (Meer & Rosen, 2004). It means that someone who is insured will take more risk, leading to more health issues, and has easier access to health care, which then both results in costs for the insurance company.
The presence of adverse selection and moral hazard, suggests that the insurance status (i.e. having insurance or not) is endogenous, making Ordinary Least Squares (OLS) es-timation a non-suitable eses-timation method. According to Shen (2013) another problem is the decision whether to make use of health care in any form, especially seeing a gen-eral practitioner. As the decision is based on personal incentives and emotions, it is hard to correct for these personal values in the estimation of health care utility and thereby health care expenditures. Some papers (e.g., Vera-Hernandez, 1999; Holly et al., 2002; Wooldridge, 2002) deal with this by making assumptions about the distribution of the error term. However, economic health literature does not provide any justification for these assumptions. Therefore usage of a parametric estimation is unlikely to provide any good estimation outcomes.
The central question of this thesis is, whether there is a stable relation between indi-vidual health expenditures and health insurance status. The last problem of the decision making process in whether to make use of health care is very complicated as one needs a multi-step approach to model the process (Shen, 2013). Therefore, this paper only focuses on the endogeneity of being insured or not which is caused by moral hazard and adverse selection. Instrumental Variable estimation (IV) and in this case Two-Stage Least Squares (2SLS) is used to tackle the problem of endogeneity. This method can be seen as a semi-parametric estimation, as it makes no assumptions on the distribution of the error term.
In this thesis I use data from the Medical Expenditure Panel Survey (MEPS) from 1996 until 2012. This is a survey, meaning it is a time series of cross-section studies. In principle, every cohort contains data from different persons. This survey collects data of different aspects in the health care system, such as sort, frequency, costs and payment of health care. The survey is conducted by the U.S. department of Health and Human Services and contains data on households, employers and medical providers. According to Shen (2013) instruments that are correlated with health care insurance, but not with utilization are difficult to find. I, therefore firstly show which instruments can be used. Then, I perform an IV-analysis on the data of 1996, 2000, 2004, 2008 and 2012. Afterwards I look at the outcomes of the individual regressions to see whether there is a significant change in the relationship between individual health expenditures and health insurance status.
This paper is organized as follows, section 2 describes the economic theory and sec-tion 3 describes the data set and introduces the model, to be estimated by 2SLS. Secsec-tion
2
Theory on endogeneity
There is a certain relation between health expenditures and some explanatory variables. According to Shen (2013), we can divide these in demographic, socioeconomic sta-tus and health related characteristics. We include demographic variables, such as age, gender and race/ethnicity (white/non-white) as these are very likely to have effect on health expenditures and are included in almost all research (e.g., Shen, 2013; Dunn, 2014). Also we include family size as larger families have workload spread out over the family and therefore might have less health issues. Furthermore, I include marital sta-tus (married/other) as this has a positive effect on health expenditures (Cabrera-Alonso, 2003). Region (North East, Midwest, South and West) is included as different regions might have different populations and differences in insurance coverage. As socioeco-nomic characteristics I include, years of education as higher educate people might be more aware of their health and therefore take different decisions. Also income and most importantly insurance status are included. For health related characteristics I include variables like BMI, as fat people are less healthy and therefore are likely to make more use of healthcare (Wee, 2005). Also, the fact whether or not a responder had a flu shot or his or here blood pressure measured in the last year, can have effect on the expenditures, as these people are more aware of their health status. Finally, perceived health status of both mental health and physical health status may have influence on health expenditures as people who feel healthier are less likely to take insurance and make use of healthcare. As discussed earlier, instrumental variable estimation can be used when explanatory variables are correlated with the error term of a regression. In this is case it is caused by moral hazard and adverse selection. In such a case, as ordinary least squares estimation
will not provide consistent estimates, it is not an appropriate method. I therefore make use of the fact that, if a good instrument can be found, the tests are accurate and there-fore the IV estimator is best in terms of variance from all linear estimators. Although using instrumental variable estimation does provide evidence for the type of relation between covariates and the dependent variable, causation still has to be based on eco-nomic theory. Furthermore, an instrumental variable itself should have the following three characteristics. It has to have no effect on its own, which can be tested by an OLS regression of the dependent variable on the instruments and all other covariates. The instruments also have to be correlated with the endogenous variable and have to be ex-ogenous. Otherwise the same problem exists for the instrument as for the endogenous variable. The appropriateness of the instrument can be tested by the Sargan test for over identifying restrictions (Heij, 2014) as I do for the IV-regression of each year.
As stated before ones insurance status may be affected by adverse selection and ones insurance status may influence ones health care expenditures, which is called moral hazard. To correct for the endogeneity caused by adverse selection and moral hazard, it is needed to find appropriate instruments for the insurance status. A good instrumental variable has a strong correlation with the endogenous variable but not with the error term (Bound et al, 1995). According to Meer and Rosen (2004), the self-employment status fits these characteristics. Perry and Rosen (2001) have shown that there is a negative correlation between self-employment and medical insurance. Meer and Rosen (2004) have shown that the self-employed are not a homogenous group and thus have to be divided in to sub-groups (i.e. sole proprietorships, partnerships and corporations), as different types of organizational form have effect on ones health care insurance status especially. People who are employed in corporations have fully deductible health
ex-penditures, meaning they are the most likely to buy insurance from the three groups. Also, people in corporations can buy insurance at beneficial rates as they can purchase insurance with a bigger group than people employed in sole proprietorships and part-nerships. According to Meer (2004), there is no evidence that self-employment itself is correlated with health care expenditures, which is investigated by an OLS regression. Therefore I use dummies for self-employed, sole proprietorships and corporations as instruments for moral hazard.
To tackle the problem of adverse selection an instrument that has correlation with expected costs, but not with the expenditures, has to be found. In the first place, one could think of using health care status as an instrument. Although, this covariate has a correlation with expected costs, it does not solely has its effect on health expenditures via the expected costs, but also directly. Therefore this covariate cannot be used as an instrument. Instead the type of insurance should be used. As people who expect the need of more health care tend to take a more generous insurance coverage (Dunn, 2014), there is a correlation between the type of insurance coverage and insurance coverage itself. If I arrange the type of coverage from small to big, meaning small covers the least amount of health care and big the most, this correlation is positive. On the other hand the type of insurance coverage itself would not necessarily lead to higher expenditures, as the higher expected costs are compensated with higher risk premiums, and therefore seems a good instrument to deal with the endogeneity of adverse selection. Unfortunately, MEPS does not contain data on the type of insurance, and therefore I do not further take into account the effect of adverse selection.
By using the instrumental variables and covariates a good model to investigate the stability of individual and health care expenditures can be estimated.
3
Data set and model
As this thesis deals with the effect of whether being insured or not on the health care expenditures, I regress the log out of pocket health expenditures on the explanatory variables, although taking in to account the endogeneity caused by adverse selection and moral hazard. I chose, as is done by Dunn (2014), to use the out of pocket prices as some preliminary regressions on the total health expenses gave very strange outcomes. Out of pocket prices are defined as self-expenditures. For uninsured this means all expenses, and for insured, this means all extra expenses that are not covered by insurance.
The following variables were included. CHIRO, which is the number of chiro based practitioner visits in a certain year. OFFICEBASED, the number of office based prac-titioner visits. OPTOMETRIST, the number of visits to an optometrist. DENTIST, the number of dentist visits in a certain year. OUTPATIENVISIT, the number of visits to a hospital, without staying overnight. FLUSHOT, a dummy, with value one if person had a flu shot in a certain year. BLOODPRESSURE, 1 if a person had its blood pressure measured. PRESCRIPTION, 1 if the person had prescription drugs. AGE and AGE-squared, the age and the squared of age in a certain year. BMI, the body mass index. MARRIED, a dummy with value 1 if a person was married, zero otherwise. EDU-CATION, the type of grade that someone achieved, ordered from low to high. MALE, dummy with value of 1 if someone is male, zero if female. WHITE, a dummy with value of 1 if someone is white, zero otherwise. RACEOTHER, with a value of 1 if someone was nor black nor white and zero otherwise. MIDWEST, SOUTH and NORTHEAST, dummies for a certain region. WEST is the region that it is compared with. SMOKER, a dummy with value 1 if someone smokes, zero otherwise. FAMSIZE, the size of the family. MENTALHEALTH and PERCEIVEDLHEALTH, dummies with value of 1 if
someone perceived its mental or physical health as good or very good, zero otherwise. In the dataset I included all data from a Full Year Consolidated Data File from a certain year, provided by the MEPS database. Descriptive statistics of the dataset are presented in Table 1; all used variables are depicted in Table 2. I excluded people who were younger than eighteen or older than 68 at the end of the year of estimation. I also excluded people who were not employed in any way, as insurance is linked to employment in the United States. If data was not available for an observation on a certain variable, I assigned the value zero. For these variables, I included a dummy to see if the observations with missing values significantly differed from the others. I did so for BMI, education and family size. BMI, SMOKER and RACEOTHER were not available for some years. In case of BMI and SMOKER, I left it out of the equation. In case of RACEOTHER, WHITE was a dummy with a value of 1 for white, and zero for all other races. If expenditures were zero or negative, I assigned the value of 1 to make a log-transformation possible. The main variable that I wanted to explain is the logarithm of out-of-pocket prices (LOGOUTOFPOCKET), taking in to account the endogeneity of insurance coverage. Taking the logarithm of the out of pocket expenses after replacing all zeros and missing values generates this variable. The insurance coverage is indicated by the variable NOINSURANCE in the regressions. This is a dummy variable with a value of 1 assigned to someone who has no insurance of any kind and zero otherwise. The instruments for NOINSURANCE were PROPRIETORSHIP, PARTNERSHIP and INCORPORATED. These are dummies with value 1 if someone was self-employed in a sole proprietorship, a partnership or a corporation, respectively.
The number of included observations rises from some 10.000 in 1996 to more than 16.000 in 2012 (Table 1). Although the amount of observations in 1996 leads to signif-icant estimations, the variance is a bit smaller in 2012, which implies more accuracy.
We see that the group of private insured drops from 79 percent in 1996 to 66 percent in 2012. Contrary to private insurance, the group of public insured and non-insured becomes larger over time. We see that especially the expenditures between 1000 dollars and 10.000 dollars rises over time and expenditures below 1000 dollars drop. The group with no expenditures rises some ten percent between 1996 and 2012, which implies that people use less health care for little interventions. For Age, the group of 50+ enlarges by seven percent, which is according to the ageing of population in the United States. The male versus female ratio in the sample stays fairly stable over time. The white ver-sus non-white ratio was eight to two in 1996 and became four to six in 2012, which is more representable for the population in the United States. We observe differences of the proportion of a certain region represented in the sample. Especially in the south we observe a rise of more than ten percent from 1996 until 2012. This may be due to sam-ple selection or interstate migration. In income we see some very significant movement. The amount of people with an income below 30.000 dollars becomes smaller and above 50.000 dollars becomes a lot bigger. This may be partly explained by inflation.
Firstly, I did an Ordinary Least Squares (OLS) regression of NOINSURANCE on all explicatory variables and instruments, to see whether there is a significant relation between the instruments and whether or not being insured. Explicatory variables such as age, education, race, gender, family size and individual income were being corrected for, as these factors have important effects on ones health care status and utility of health care insurance as previously discussed (Meer and Rosen, 2004). Also other explicatory variables that are assumed to be exogenous were included, such perceived mental and physical health status, the use of a dentist, optometrist, chiropractitioner and the use of prescription drugs. The used explicatory variables are depicted in Table 2. Secondly I performed an OLS regression of LOGOUTOFPOCKET on the explicatory variables
and the instruments, to see if there is a significant influence of the instruments on the out of pocket health expenditures. If this was the case, a certain instrument could not be used and was used as an exogenous explicatory variable in further regressions. The set of instruments used may therefore differ per year, and contained one, two or all three of the defined instruments. Hereafter, a Breusch-Paegan test on homoscedasticity was per-formed. If there was enough evidence to reject the null hypothesis of homoscedasticity, the model was estimated with White standard errors. After this, and IV-analysis was performed on the logarithm of out of the pocket health expenditures. I studied two sets of equations. Equation one regresses the possible endogenous variables on the instru-ments and all exogenous covariates. Equation two deals with the health expenditures. This is a so-called Two Stage Least Squares approach and is mathematically written as follows.
Step 1: ¯
I= δ + θ1Incorporated + θ2Sole Proprietorship + θ3Partnership + X + φ
With I the insurance status (i.e. 1 if insured and 0 if not), δ the intercept, X the exoge-nous regressors and φ the error term and Incorporated, Sole Proprietorship and Partner-ship, n x 1 dummy vectors.
Step 2:
With ¯Y the out of the pocket health expenditures, α the intercept, X the n x m matrix of m exogenous regressors and ε the error term vector. ¯I is from the regression in step 1. This leads to the IV-estimation of β and γ.
4
Results and Analysis
Firstly, the OLS regression of NOINSURANCE on all explicatory variables and the instruments showed significant relation between the instruments and the insurance status for every year. The correlation of the instruments with the out of pocket prices was determined by an OLS regression of LOGOUTOFPOCKET on all explicatory variables and possible instruments included as exogenous variables. Only non-significant possible instruments were really used as instruments. For 2012, 2004 and 2000 these were Sole Proprietorship and Partnership. For 2008 only Partnership, which is why a Sargan test could not be performed. For 1996, all instruments were significant, meaning the output of the 2SLS regression cannot be analyzed properly. In each regression I used a set of different instruments.
Hereafter, a Breusch-Paegan test on homoscedasticity was performed. All years ex-cept 2012 contained heteroscedasticity, so White errors were used in further regressions for those years. This is depicted as a star (*) in Table 5. A summary of the parameter estimation for having insurance by both estimation methods is depicted in table 5.
Before discussing further results, it is necessary to look at the outcome of all OLS and Two-Stage Least Squared regressions of LOGOUTOFPOCKET on the explicatory variables as depicted in Table 2. If we look at the 2SLS estimations of the out of pocket expenses, we see that only for the years 2004 and 2012 the parameters for NOINSUR-ANCE seem somehow reliable and explainable. We observe very strange results for the years 1996, 2000 and 2008 (Table 4). Here, the Sargan test is not significant, barely significant or not calculable (Table 5). We see that in 2004 and 2012, were the 2SLS estimation seems accurate, the Sargan test is significant, meaning the instruments are valid. However, the Durbin-Wu-Hausman test is not significant in both years, meaning
having insurance is exogenous and OLS is therefore consistent. As Shen (2013) de-scribes that having insurance is endogenous, this may be explained by the fact that the instruments are not as good as Meer and Rosen (2004) state. As the outcomes of the 2SLS regressions do not seem to make sense, I focused on the outcomes of the OLS regression as described in the second step of the Data and Model chapter of this thesis and as depicted in Table 3.
If we observe the parameter estimations for out of pocket expenses, we see that the estimation for not having insurance is more or less the same for each year, ranging from 0,222 to 0,306. This means that someone that does not have insurance has out of pocket payments of between 22% and 30% more on average than someone who is insured. Although the estimations seem consistent over time, they are not according to the economic theory (Shen, 2013), in which people that are not insured have lower expenses. This can be explained by the fact that I did not look at the total expenses, but the out of pocket expenses, in which the payment by the insurance company is not accounted for.
The following are outcomes of the control variables. Visiting the chiropractitioner, dentist and optometrist lead to significant higher health care expenses. Also people who have their cholesterol and blood pressure measured in the last year have significant higher expenditures. Besides the direct effect of the measurement on the expenses, this may be due the fact that these people are less healthy and therefore have more health expenditures. Also having prescription drugs has a positive effect on health expenses. Besides the direct effect of the costs of prescription drugs, it is likely that people using drugs are less healthy and therefore make more use of the health care system, leading to significant higher costs. From on 2004, BMI is available in MEPS. In 2004 and 2012 the estimate for BMI is significant, however the effect is only marginally positive. Marriage
is significant with p-value smaller than 0.05 in 2012 only and has a positive effect on the expenditures of around 11%. The logarithm of income is significant in all years, except 2004 and has a positive effect of around 6%. People with higher incomes are more likely to take the risk of not having insurance, as they can pay themselves when needed, which is why they have higher out of pocket expenditures. In 2004 the estimate is not significant. White people have out of pocket expenditures of between 10% and around 40% more than other people. Males have between 10% and 24% less health care expenses than women. The outcomes of the estimates of region are not unequivocally pointed in a positive or negative direction. This can be caused by the sample selection and by interregional migration, however the MEPS does not contain data on this issue. The family size has a negative influence on the expenditures. One extra family member leads to a decrease of expenses of between 5% and 9% and is highly significant for all years. Perceived health status had a negative effect on the expenses of between 16% and 38%. This can be explained by the fact that people who have a perceived health status of good, very good or excellent are actually healthier than people with worse perceived health. Also, people with a good, very good or excellent perceived health status make less use of health care, as they feel healthy, which lowers the expenses. Finally, we see that being self-employed leads to higher expenses of up to 50% in 2004. This was to be expected, as self-employed people are less likely to have insurance and therefore have higher expenses as not having insurance leads to higher expenses.
5
Conclusions
As the 2SLS outcomes were strange, I used the OLS estimations. First of all, we have seen a significant negative influence of not being insured on out of the pocket health care expenses. The estimated effect is around 30%, with a drop in 2008 of around 8%. Therefore, it can be stated that the relation between individual health expenditures and health insurance is quite stable over time. As we do not use the total amount of health expenditures, which includes the health care insurance itself and the payments by the insurance company, the relation is negative. This means that someone who is insured has to pay less extra than someone who is not insured, which is in line with intuition.
Furthermore, I wanted to correct for ones willingness to take risk, as it may influence the decision to take health care insurance. Unfortunately, MEPS does not contain data on someone’s willingness to take risk. I suggest for future research to add questions to the panel that can model some risk behavior. To tackle the endogeneity of the adverse selection, I had to find proper instruments. As Dunn (2014) suggested, I had to use the type and amount of insurance coverage per policy. The MEPS database, however, does not contain any data on this subject, which makes this impossible. Finally, the MEPS database is made up out of cross sections. In this way it is not possible to follow an individual over time. To really get to know upon which factors an individual makes decisions whether to take insurance or not, a longitudinal database has to be created.
References
[1] Cabrera-Alonso, J., Long, M.J., Bangalore, V., Lescoe-Long, M., A., Marital Status and Health Care Expenditures among the Elderly in a Managed Care Organization, Health Care Management, 2003, vol. 3, pp 249-255
[2] Dunn, A., Health Insurance and the Demand for Medical Care: Instrumental Vari-able Estimates Using Health Insurer Claims Data. Bureau of Economic Analysis Working Papers, 2014
[3] Heij, C., Boer, P. de, Franses, P.H., Kloek, T. and Dijk, H.K. van, 2004, Econometric Methods with Applications in Business and Economics, Oxford University Press, pp 396-418.
[4] Holly, A., Hospital Services Utilization in Switzerland: The Role of Supplemen-tary Insurance, Institute of Health Economics and Management, University of Lau-sanne, 2002.
[5] Meer, J., Rosen, H.S., Insurance and the Utilization of Medical Services. Social Science Medicine, vol. 58, Issue 9, May 2004, pp 1623 - 1632.
[6] Perry, Craig, W., Rosen, H.S., Insurance and the Utilization of Medical Services Among the Self Employed. NBER Working Paper 8490. August 2001.
[7] Shen, C., Determinants of Health Care Decisions: Insurance, Utilization, and Expenditures. Review of Economics and Statistics, 2013, vol. 95, pp. 142-153 [8] Vera-Hernandez, A.M., Duplicate Coverage and Demand for Healthcare. The case
[9] Wooldridge, J.M., Econometric Analysis of Cross Section and Panel Data, Cam-bridge: MIT Press, 2002.
[10] Wee, C.C., Phillips, R.S., Legedza, A.T.R., David, R.B., Soukup, J.R., Colditz, G.A., Hamel, M.B., Health Care Expenditures Associated With Overweight and Obesity Among US Adults: Importance of Age and Race, American Journal of Pub-lic Health, 2005, vol. 1, pp 159-165
Table 1: Descriptive statistics
Year 1996 2000 2004 2008 2012
N 10136 (in %) 11466 (in %) 14874 (in %) 14574 (in %) 16495 (in %) Insurance Private 7985 78,78 9007 78,55 10650 71,60 10266 70,44 10907 66,12 Public 430 4,24 484 4,22 1033 6,95 1073 7,36 1392 8,44 Uninsured 1721 16,98 1975 17,22 3191 21,45 3235 22,20 4196 25,44 Expenditures 0 2394 23,62 2977 25,96 4186 28,14 4303 29,53 5809 35,22 <1000 9450 93,23 10501 91,58 13041 87,68 12750 87,48 14756 89,46 1000-2000 478 4,72 643 5,61 1116 7,50 1115 7,65 1012 6,14 2000-5000 175 1,73 279 2,43 600 4,03 561 3,85 580 3,52 5000-10000 26 0,26 40,00 0,35 97 0,65 120 0,82 113 0,69 10000+ 7 0,07 3 0,03 20 0,13 28 0,19 34 0,21 Age <40 5607 55,32 6073 52,97 8259 55,53 7471 51,26 8513 51,61 40-49 2673 26,37 3018 26,32 3751 25,22 3561 24,43 3796 23,01 50+ 1856 18,31 2375 20,71 2864 19,26 3542 24,30 4186 25,38 Gender Man 5286 52,15 5930 51,72 7673 51,59 7439 51,04 8456 51,26 Vrouw 4850 47,85 5936 51,77 7201 48,41 7135 48,96 8039 48,74 Race White 8416 83,03 9394 81,93 11902 80,02 10698 73,40 6689 40,55 Non-White 1720 16,97 2072 18,07 2972 19,98 3876 26,60 9806 59,45 Region Northeast 1982 19,55 1758 15,33 2177 14,64 2223 15,25 2582 15,65 Midwest 2293 22,62 2459 21,45 3011 20,24 2939 20,17 3171 19,22 South 2560 25,26 4368 38,10 5904 39,69 5402 37,07 6130 37,16 West 2355 23,23 2881 25,13 3782 25,43 4010 27,51 4612 27,96 Income <20000 4498 44,38 4332 37,78 5741 38,60 4953 33,99 5538 33,57 20000-30000 2156 21,27 2273 19,82 2822 18,97 2778 19,06 3090 18,73 21
Table 2: Variables VARIABLE DESCRIPTION
DEPENDENT:
LOGOUTOFPOCKET logarithm of outofpocket payments EXPLANATORY:
AGE Age of a certain person AGE_2 Age-squared
BLOODPRESSURE dummy: 1 if a person had its blood pressure measured CHIRO number of chiro based practitioner visits in a certain year CHOLESTEROL dummy: 1 if person had its cholersterol measured
DENTIST number of dentist visits in a certain year EDUCATION years of education
FAMSIZE total number of family members in same household FLUSHOT dummy: 1 if person had a flu shot in a certain year INCORPORATED dummy: 1 if a person is incorporated
LOGINCOME logarithm of a person’s income MALE dummy: 1 if a person is male
MARRIED dummy: 1 if the person was married
MENTALHEALTH dummy: 1 if person perceives mental health as good or very good MIDWEST dummy: 1 if a person lives in the region Midwest
NOINSURANCE dummy: 1 if a person has no insurance
NORTHEAST dummy: 1 if a person lives in the region Northeast OFFICEBASED number of office based practitioner visits
OPTOMETRIST number of visits to an optometrist
OUTPATIENTVISIT number of visits to a hospital, without staying overnight PARTNERSHIP dummy: 1 if a person is a partner in a company
PERCEIVEDHEALTH dummy: 1 if person perceives physical health as good or very good PRESCRIPTION dummy: 1 if person had prescription drugs
PROPRIETORSHIP dummy: 1 if a person has a sole proprietorship
RACEOTHER dummy: 1 if a person has another race than white or black SMOKER dummy: 1 if a person smokes
Table 3: Ordinary Least Squares
YEAR (2012) (2008) (2004) (2000) (1996)
DEPENDENT VARIABLE LOGOUTOFPOCKET LOGOUTOFPOCKET LOGOUTOFPOCKET LOGOUTOFPOCKET LOGOUTOFPOCKET CHIRO 0.7661 (0.000) 0.6315 (0.000) 0.4887 (0.000) 0.6189 (0.000) 0.5928 (0.000) OFFICEBASED 1.8380 (0.000) 1.8478 (0.000) 1.6294 (0.000) 1.5664 (0.000) 1.7031 (0.000) BLOODPRESSURE 0.2643 (0.000) 0.2541 (0.000) 0.1828 (0.000) 0.1926 (0.000) 0.3095 (0.000) CHOLESTEROL 0.6244 (0.000) 0.6628 (0.000) 0.1874 (0.000) 0.1271 (0.000) 0.0071 (0.852) FLUSHOT 0.0903 (0.475) -0.0369 (0.717) 0.0482 (0.245) 0.0382 (0.328) 0.0335 (0.435) OPTOMETRIST 0.6351 (0.000) 0.6661 (0.000) 0.6494 (0.000) 0.5567 (0.000) 0.5346 (0.000) DENTIST 1.2570 (0.000) 1.1082 (0.000) 1.0739 (0.000) 1.1616 (0.000) 1.1268 (0.000) OUTPATIENTVISIT 0.6417 (0.000) 0.4661 (0.000) 0.4018 (0.000) 0.3712 (0.000) 0.4264 (0.000) PRESCRIPTION 2.0576 (0.000) 2.0482 (0.000) 2.1918 (0.000) 2.0705 (0.000) 1.7046 (0.000) AGE -0.0281 (0.001) -0.0159 (0.074) -0.0380 (0.000) -0.0358 (0.000) -0.0291 (0.004) AGE_2 0.0005 (0.000) 0.0004 (0.000) 0.0007 (0.000) 0.0006 (0.000) 0.0006 (0.000) BMI 0.0007 (0.773) -0.005 (0.065) 0.0029 (0.160) D_BMI 0.0182 (0.842) -0.0029 (0.973) -0.1205 (0.138) MARRIED 0.1139 (0.000) 0.0624 (0.066) 0.0951 (0.003) 0.061 (0.088) -0.0493 (0.189) LOGINCOME 0.0629 (0.000) 0.0631 (0.000) 0.0614 (0.000) 0.0550 (0.002) 0.0042 (0.766) EDUCATION 0.0330 (0.000) 0.0410 (0.000) 0.0491 (0.000) 0.0430 (0.000) 0.0452 (0.000) D_EDUC 0.4517 (0.000) 0.2712 (0.082) 0.1953 (0.133) 0.5052 (0.006) 0.3230 (0.530) MALE -0.1044 (0.000) -0.1193 (0.000) -0.1972 (0.000) -0.2358 (0.000) -0.1628 (0.000) WHITE 0.3735 (0.000) 0.1727 (0.000) 0.1010 (0.017) -0.1029 (0.599) 0.3955 (0.000) RACEOTHER -0.1197 (0.194) 0.2080 (0.000) 0.2189 (0.000) MIDWEST -0.0188 (0.651) -0.0788 (0.073) 0.0118 (0.778) 0.1136 (0.013) 0.1223 (0.011) SOUTH 0.0600 (0.084) 0.0997 (0.007) 0.1454 (0.000) 0.2462 (0.000) 0.2653 (0.000) NORTHEAST -0.2500 (0.000) -0.2764 (0.000) -0.1309 (0.005) 0.0553 (0.271) 0.1194 (0.020) SMOKER -0.0845 (0.028) -0.1161 (0.003) -0.0156 (0.677) -0.0243 (0.528) FAMSIZE -0.0857 (0.000) -0.0618 (0.000) -0.0670 (0.000) -0.0624 (0.000) -0.0567 (0.000) D_FAMSIZE -0.7823 (0.654) 0.5470 (0.522) -0.3874 (0.108) 3.5048 (0.000) -0.2317 (0.105) MENTALHEALTH -0.0144 (0.839) -0.0586 (0.455) -0.2162 (0.006) -0.1127 (0.205) 0.2549 (0.003) PERCEIVEDHEALTH -0.1626 (0.002) -0.3588 (0.000) -0.3119 (0.000) -0.3755 (0.000) -0.2839 (0.000) NOINSURANCE 0.2869 (0.000) 0.2215 (0.000) 0.2896 (0.000) 0.2807 (0.000) 0.3063 (0.000) INCORPORATED 0.2589 (0.003) 0.2650 (0.001) 0.5185 (0.000) 0.2555 (0.003) 0.3328 (0.000) PARTNERSHIP 0.2846 (0.101) 0.2551 (0.125) 0.0108 (0.945) 0.3923 (0.010) 0.3430 (0.004) PROPRIETORSHIP 0.0635 (0.242) 0.1689 (0.005) 0.0565 (0.316) 0.2061 (0.001) 0.2094 (0.001) _cons 0.1144 (0.575) 0.1836 (0.406) 0.4987 (0.015) 0.7255 (0.002) 0.2526 (0.279) N 16495 14574 14874 11466 10136 23
Table 4: Two-Stage Least Squares
YEAR (2012) (2008) (2004) (2000) (1996)
DEPENDENT VARIABLE LOGOUTOFPOCKET LOGOUTOFPOCKET LOGOUTOFPOCKET LOGOUTOFPOCKET LOGOUTOFPOCKET NOINSURANCE 0.6637 (0.008) 1.7874 (0.105) 0.5754 (0.048) 1.6462 (0.000) 1.9359 (0.000) CHIRO 0.7428 (0.000) 0.5647 (0.000) 0.4761 (0.000) 0.6323 (0.000) 0.5456 (0.000) OFFICEBASED 1.8867 (0.000) 2.0021 (0.000) 1.6597 (0.000) 1.6580 (0.000) 1.8106 (0.000) BLOODPRESSURE 0.3156 (0.000) 0.3789 (0.000) 0.2028 (0.000) 0.2829 (0.000) 0.4085 (0.000) CHOLESTEROL 0.6185 (0.000) 0.6629 (0.000) 0.2014 (0.000) 0.1732 (0.000) 0.05418 (0.189) FLUSHOT 0.1077 (0.397) 0.0114 (0.920) 0.0524 (0.209) 0.0776 (0.081) 0.0333 (0.459) OPTOMETRIST 0.6390 (0.000) 0.6683 (0.000) 0.6484 (0.000) 0.5201 (0.000) 0.5684 (0.000) DENTIST 1.2861 (0.000) 1.2222 (0.000) 1.0926 (0.000) 1.2321 (0.000) 1.2317 (0.000) OUTPATIENTVISIT 0.6567 (0.000) 0.5080 (0.000) 0.4080 (0.000) 0.3912 (0.000) 0.4539 (0.000) PRESCRIPTION 2.0725 (0.000) 2.1274 (0.000) 2.2051 (0.000) 2.1260 (0.000) 1.7580 (0.000) AGE -0.0336 (0.000) -0.0239 (0.031) -0.0397 (0.000) -0.0427 (0.000) -0.0268 (0.012) AGE_2 0.0005 (0.000) 0.0005 (0.000) 0.0007 (0.000) 0.0007 (0.000) 0.0006 (0.000) BMI 0.0011 (0.649) -0.0051 (0.056) 0.0028 (0.170) D_BMI 0.0088 (0.923) -0.0762 (0.477) -0.1386 (0.095) MARRIED 0.1401 (0.000) 0.1682 (0.041) 0.1149 (0.002) 0.1804 (0.000) 0.1140 (0.043) LOGINCOME 0.0760 (0.000) 0.1135 (0.003) 0.0717 (0.000) 0.1156 (0.000) 0.0498 (0.011) EDUCATION 0.0401 (0.000) 0.0743 (0.002) 0.0531 (0.000) 0.0711 (0.000) 0.0811 (0.000) D_EDUC 0.5560 (0.000) 0.5448 (0.033) 0.2336 (0.086) 0.8071 (0.000) 0.6191 (0.254) MALE -0.1105 (0.000) -0.1563 (0.000) -0.2033 (0.000) -0.2618 (0.000) -0.1954 (0.000) WHITE 0.4042 (0.000) -0.0157 (0.909) 0.0697 (0.186) -0.2232 (0.212) 0.4057 (0.000) RACEOTHER -0.0822 (0.391) 0.3880 (0.003) 0.2579 (0.000) MIDWEST -0.0155 (0.710) -0.0895 (0.054) 0.0135 (0.748) 0.1880 (0.000) 0.1864 (0.001) SOUTH 0.0415 (0.254) -0.0089 (0.916) 0.1323 (0.000) 0.2188 (0.000) 0.2607 (0.000) NORTHEAST -0.2320 (0.000) -0.2642 (0.000) -0.1221 (0.009) 0.1093 (0.041) 0.1709 (0.002) SMOKER -0.1007 (0.013) -0.1917 (0.004) -0.0256 (0.506) -0.0679 (0.101) FAMSIZE -0.0870 (0.000) -0.0633 (0.000) -0.0654 (0.000) -0.0711 (0.000) -0.0687 (0.000) D_FAMSIZE -0.8051 (0.645) 0.4178 (0.630) -0.3324 (0.169) 3.8662 (0.021) -0.3704 (0.020) MENTALHEALTH -0.01367 (0.848) -0.0444 (0.587) -0.2049 (0.009) -0.1239 (0.187) 0.3117 (0.001) PERCEIVEDHEALTH -0.1411 (0.011) -0.3026 (0.000) -0.3003 (0.000) -0.2733 (0.000) -0.1693 (0.018) INCORPORATED 0.2179 (0.014) 0.1078 (0.440) 0.4972 (0.000) 0.1642 (0.063) PROPRIETORSHIP -0.1361 (0.540) _cons -0.2248 (0.465) -1.1785 (0.233) 0.2434 (0.462) -0.6313 (0.136) -1.4366 (0.004) N 16495 14574 14874 11466 10136 24
Table 5: Summary statistics of the effect of having no insurance on out of pocket expenditures Year 1996 2000 2004 2008 2012 OLS Parameter 0,306 0,281 0,290 0,222 0,287 Variance 0,049* 0,046* 0,040* 0,041* 0,036 2SLS Parameter 1,936 1,646 0,575 1,787 0,664 Variance 0,425* 0,358* 0,292* 1,102* 0,252 Sargan 0,004 0,093 0,823 0,185 Durbin-Wu-Hausman 0,000 0,001 0,334 0,129 0,138 * White Standard Errors
