Presentation is loading. Please wait.

Presentation is loading. Please wait.

Yingying Dong Boston College 10/16/2008

Similar presentations


Presentation on theme: "Yingying Dong Boston College 10/16/2008"— Presentation transcript:

1 Yingying Dong Boston College 10/16/2008
Why Do the Insured Use More Health Care? The Role of Insurance-Induced Unhealthy Behaviors Yingying Dong Boston College 10/16/2008

2 A Real Life Story From the New York Times (Nov. 25, 2002):
Mr. & Ms. Brooks dropped their health insurance because of increased premiums. Then… “Mr. Brooks, 50, has stopped taking Lipitor to control high cholesterol and has started taking over-the-counter herbal supplements. Ms. Brooks no longer takes Singulair for asthma and has adopted an exercise program intended to regulate her breathing. Ms. Brooks estimates they are saving $150 a month by not using prescription drugs. ‘We changed our diets a lot to help the effectiveness of the supplements, and maybe that’s a good thing,’ she said.” --Broder, John. “Problem of Lost Health Benefits is Reaching into the Middle Class.” Also cited by Dave & Kaestner (2006) 1/29

3 Motivation – (1) Health insurance is associated with an increased use of health care. (Yelin et al, 2001; Meer and Rosen, 2003; Pauly, 2005; Bajari, Hong and Khwaja, 2006; Deb and Trivedi, 2006) Usually assumed: Individuals who potentially have a greater need of heath care are more motivated to get insured (selection effect). Insurance reduces the effective price of health care and hence induces individuals to use more. ★ What’s missing: Insurance encourages unhealthy behaviors, which may cause increased use of health care. Unhealthy behaviors: Drinking, Smoking, Insufficient Exercise, and unhealthy diet 2/29

4 Motivation- (2) Intuitions:
Insurance lowers the offsetting cost for the negative health consequences of unhealthy behaviors. disincentive effect Car insurance reduces precaution and stimulates car accidents; Workplace injury compensations increase injuries. Individuals with health problems substitute medication for behavioral improvement. substitution effect Easily accessible health care may distort the perceived risk of unhealthy behaviors. distorted image 3/29

5 Illustration of Causalities
Health Insurance Use of Health Care Selection Effect (Ex post) Moral Hazard Pure Price Effect Health Related Behaviors True Moral Hazard (1) Causalities traditionally studied (2) A fuller View of Causalities True Moral Hazard: the disincentive effect of health insurance on individuals’ healthy behaviors, which may generate additional medical care demand. Pure Price Effect: the effect that health insurance lowers the effective price of health care and hence induces individuals to use more care ceteris paribus. 4/29

6 Literature Review Most existing literature studies the insurance effect on medical utilization; a small strand examines the insurance effect on health behaviors. They do not look at the structural causal relationships among the three. Literature examining the insurance effect on medical utilization: Reviews: Zweifel and Manning (2000), Buchmueller et al (2005) Studies examining the insurance effect on prevention, including behaviors (discrete outcomes): Kenkel (2000), Courbage and Coulon (2004) A few studies that do consider all three focus on discrete outcomes Khwaja (2002, 2006) and Card et al. (2004) 5/29

7 Research Questions & Approaches
(1) What are the effects of health insurance on individuals’ health behaviors? (2) What is the (total) effect of health insurance on health care utilization? (3) Within the total effect of health insurance on health care utilization, how much is caused by the pure price effect, and how much is caused by individuals’ behavior change when they have insurance? Start from a theoretical model to derive the structural model for the insurance decision, health behaviors, and health care utilization Solve the structural model to obtain the semi-reduced form equations determining behaviors and care utilization as functions of endogenous health insurance Derive the structural parameters of interest: the direct and indirect effects of health insurance on health care utilization Empirical analysis adopts the generalized Tobit specification with transformations on the dependent and lagged dependent variables. 6/29

8 Theoretically, Empirically, Contributions
Set up a two-period dynamic forward-looking model and derive the structural causal relationship among the insurance, behaviors and health care utilization; Model continuous choices instead of discrete choices. Empirically, Distinguish between the extensive margin (changes in the proportion of the unhealthy behavior participants) and the intensive margin (changes in the quantity of unhealthy behaviors given participation) of the insurance effect; Accommodate the distribution characteristics of the data and adopt flexible transformations; Analyze the insurance effects on heavy drinking, and how the insurance-induced drinking affects medical utilization. 7/29

9 Basic Theoretical Model – (1)
A two-period dynamic forward-looking model; The individual draws utility from composite good consumption (Ct), unhealthy behaviors (Bt), and health (Ht ); Assume rational addiction of unhealthy behaviors. Utility function: Health evolution equation: Budget constraint: Ht = Initial health status/Health stock; Mt = medical care utilization; Bt = (Bt1, Bt2, Bt3)': drinking, smoking and exercise; st = health shock, which may depend on Ht, Bt; t = permanent taste parameter; = taste shifter; Wt = present value of total wealth; It = insurance dummy; dt = the insurer co-payment rate. 8/29

10 Basic Theoretical Model – (2)
The individual invests in health in 1st period (t), and bears the addiction and negative health consequences of unhealthy behaviors in 2nd period (t+1). C: choice set: {It} {Ct, Bt, Mt} {Ct+1, Bt+1} F: information set: {Bt-1,Ht, t,t} {Bt-1,Ht, It, t ,t} {Bt,Ht+1,It,t,t+1, st+1} S: shock set: {st , st+1, t+1} {st, st+1, t+1} {st+1} Expected Utility maximization s.t. At the beginning of 1st period In 1st period In 2nd period 9/29

11 Structural & Semi-reduced Form Equations
Maximizing expected utility by backward induction  Structural equations (FOC’s) for Mt, Bt and the insurance decision It Bt-1, PIt : exclusion restrictions fI() = PIt*: willingness-to-pay for insurance Intuitions Solve (1),(2) for Bt and Mt  semi-reduced form equations Assume U() is quadratic and health production function is linear  linear functions for Bt and Mt and the It index 10/29

12 Direct and Indirect Insurance Effects on Care Utilization - (1)
Eqs. (1), (4) and (5) imply the following decomposition Total effect = direct price effect + indirect effect (I) 13 31 Obtained from Eq. (4) (II) To be backed out 3  1 Obtained from Eq. (5) Obtained from the Eq. (5) Obtained from Eq. (4) 11/29

13 Direct and Indirect Insurance Effects on Care Utilization– (2)
(I) is a vector form of (II) is the solution of a linear equation system Three unknowns 12/29

14 General Data Problems A significant fraction of zeros (68% zero drinking, 83% zero smoking) Discrete change between zero & non-zero consumptionextensive margin of the insurance effect Continuous change in the positive level of consumptionintensive margin of the insurance effect Two part specification: Insurance effect on the probability of non-zero consumption Insurance effect on the level of consumption, given participation  Generalized Tobit model (sample selection model) A skewed distribution of positive observations (nonnormality) Transformation on the dependent and lagged dependent variables IHS (Inverse hyperbolic sine), Box-Cox, log 13/29

15 Empirical Model & Identification
Generalized Tobit (sample selection) model with transformations on the dependent and lagged dependent variables where Yt = Bt or Mt ; Insurance is endogenous joint estimate It and Bt or Mt . PIt is an exclusion restriction (Instrumental Variable): Age≥65 dummy (Card, Dobkin and Maestas, 2004) Self-employment status(Meer and Rosen, 2003; Deb and Trivedi, 2006) 14/29

16 IV Validity-(1) Relevance Exogeneity insurance holding rate:
Age<65 (88.1%), Age≥65 (99.1%) Self-employed(78.7% ), Non Self-employed (92.6%) Exogeneity Measures of medical care utilization and health behaviors would have evolved smoothly with age in the absence of the discrete change in insurance coverage at age 65. Include in all equations a smooth age profile function, tentatively include age dummy in the outcome equations, coefficients are not significant Sargan’s and Basmann’s over-identification tests Hansen’s J test Using panel data check the transition into/out of self-employment on behaviors 15/29

17 IV Validity-(2) Table 2-(a) The impact of transition out of self-employment on health behaviors Treatment 1 Control 1 Diff.-In-Diff. P-value Change in drinking (level) .271(.300) -.087(.197) .358(.359) .319 Change in drinking (prob.) -.028(.021) -.029(.014) .001(.025) .965 Change in smoking (prob.) -.009(.008) -.014(.007) .005(.010) .637 Change in exercising (prob.) -.050(.033) -.027(.018) -.023(.038) .536 Table 2-(b) The impact of transition into self-employment on health behaviors Treatment 1 Control 1 Diff.-In-Diff. P-value Change in drinking (level) -.509(.419) -.207(.041) .301(.421) .476 Change in drinking (prob.) -.022(.028) -.030(.004) .008(.028) .787 Change in smoking (prob.) -.029(.015) -.012(.002) -.017(.015) .267 Change in exercising (prob.) .029(.041) -.014(.005) .043(.041) .300 16/29

18 Sample Description RAND HRS : 3rd, 4th and 5th (Year 96, 98, 00) waves of data Not include: Age>70, On social security disability insurance, Deceased within 2 years since being observed in the sample N=14,289 Dependent variables : Dummies: smoking, exercising (3 or more times per week), drinking, visiting a doctor/hospital (at least once for the past two years) Levels: # of alcoholic drinks consumed per week # of doctor/hospital visits per year 17/29

19 Summary Statistics-(1)
Insured (n=13,016) Uninsured (n=1,273) Mean Std. Dev. Visiting a doctor/hospital .941 .235 .811 .391 # of visits 4.35 7.74 3.02 4.98 Current period: Smoking .167 .373 .262 .440 Exercising .504 .500 .485 Drinking .330 .470 .265 .441 Positive # of alcoholic drinks 7.14 8.50 10.17 12.61 The insured are more likely to visit a doctor or hospital, and they also have more visits on average than the uninsured. Insurance is associated with healthier behaviors; e.g., The insured are less likely to smoke; more likely to drink, but on average drink much less. ★ These may not be causal: confounding factors or selection effects. 18/29

20 Summary Statistics-(2)
Insured (n=13,016) Uninsured (n=1,273) Mean Std. Dev. Last period: Smoking .184 .387 .274 .446 Exercising .523 .499 .516 .500 Drinking .353 .478 .295 .456 Positive # of alcoholic drinks 7.01 8.46 10.58 15.29 Last period diagnosed disease*: Cancer .065 .246 .048 .214 Hypertension .367 .482 .342 .475 Heart disease .119 .324 .072 .259 Lung disease .050 .218 .038 .191 Age 61.65 5.19 59.48 4.97 The insured have healthier behaviors ex ante (last period); The insured are in general older, tend to have chronic diseases; * This list leaves out some chronic diseases due to space limit. 19/29

21 Summary Statistics-(3)
Insured (n=13,016) Uninsured (n=1,273) Mean Std. Dev. Male .406 .491 .341 .474 Hispanic .067 .251 .235 .424 Last period health status: Fair/ Poor .166 .372 .269 .444 Good/Very good .652 .476 .571 .495 Excellent .182 .386 .159 .366 Education: Less than high-school .192 .394 .437 .496 High-school or GED .487 .311 .463 College or above .422 .494 .252 .434 Income($1000) 62.80 86.19 36.27 65.93 Number of children 3.45 2.05 4.01 2.54 The insured also tend to have higher income, higher education… 20/29

22 Estimated Insurance Effects
Table 1 Insurance effects on the probabilities of using health care and health behaviors Dependent variable Coeff. Of Insurance (SE) Marginal Effect Visiting a doctor/hospital (0/1) .037 (.146) .008 Exercising(0/1) -.112 (.127) -.045 Smoking(0/1) .166 (.192) .017 Drinking(0/1) -.187 (.145) -.065 Table 2 Insurance effects on the levels of health care utilization and drinking HIS Box-Cox Log % change (SE) # Model (1) # of visits / year .367(.069)*** 1.67 .365(.069) *** 1.66 .368(.056) *** 1.68 # of alcoholic drinks/week .122(.100) .900 .120(.098) .886 .108(.095) .794 Model (2) .347(.069)*** 1.58 .359(.068) *** 1.63 .348(.057) *** .043(.097) .318 .041(.095) .302 .029(.093) .214 Model (1): Two-part Model ; Model (2): Sample selection Model; ***Significant at .01 level; 21/29

23 Further Investigation on Drinking
Puzzle 1: Insurance may decrease the probability of drinking, although it may increase the amount of drinking by the drinkers. Puzzle 2: The insurance effect on the probability of drinking and that on smoking have opposite signs. Unlike smoking, a low level of drinking is generally considered as healthy. In the data, the insured have a smaller proportion of smokers, but a larger proportion of drinkers; whereas the insured drinkers tend to be light drinkers. Distinguish between light drinking and heavy drinking Disincentive effect of insurance on healthy behaviors: Does insurance holding encourage heavy drinking? 22/29

24 Further Investigation on Heavy Drinking
Table 3 Insurance effects on heavy drinking Heavy drinking Probability Weekly # of drinks (IHS model) Cutoff percentile # of drinks Sample mean Coeff.(SE) Marginal effect Marginal effect (SE) Change in level 50th >4 12.8 .093(.175) .011 .131(.066) ** 1.67 60th >6 14.9 .190(.165) .013 .155(.071) ** 2.31 70th (mean) >7 16.8 .127(.168) .007 .120(.070) * 2.02 80th >12 21.4 .229(.188) .005 .177(.074) ** 3.79 85th >14 27.2 .307(.211) .003 .232(.100) ** 6.31 90th >15 28.5 .330(.278) .002 .255(.099) ** 6.98 * Significant at .1 level; **Significant at .05 level; ***Significant at .01 level 23/29

25 Increased Number of Visits due to Insurance-Induced Drinking
Increased medical utilization due to insurance-induced drinking: Among the above-the-median drinkers, the increased number of visits caused by insurance-induced drinking is at most 0.3%. Averaging it over the whole population will make it even smaller. If the number of visits caused by the insurance-induced changes in smoking and exercise is of the same magnitude, then the increased number of visits caused by insurance-induced unhealthy behaviors as summarized by drinking, smoking and insufficient exercise is less than 1%. 24/29

26 Conclusions Health insurance encourages individuals’ unhealthy behaviors, in particular, heavy drinking, but this does not induce an immediate perceivable increase in their use of health care. The insurance effects at the extensive margin are in general less significant than those at the intensive margin. Eg. The effects on the probabilities of smoking, drinking/heavy drinking, and visiting a doctor/hospital are small and insignificant; The effects on the quantity of heavy drinking and on the number of visits are more considerable and statistically significant. Within the total effect of insurance on health care utilization, the pure price effect is dominant. Policy implications 25/29

27 Kept Back to Get Ahead? Kindergarten Retention and Academic Performance

28 Kept Back to get ahead? – Kindergarten Retention and Academic Performance
Basic facts from the Early Childhood Longitudinal Study –- Kindergarten Cohort (ECLS-K): Not every school allows children to be held back in kindergarten; Children in retention schools (schools that permit repeating kindergarten) and non-retention schools are significantly different in terms of parents and family characteristics. We observe children being held back in K if and only if 1) they attend a school that allows for repeating K, and 2) receive the treatment of repeating K. The two decisions are jointly determined: common observed and unobserved parental/family characteristics may affect both. 26/29

29 Kept Back to get ahead?— Repeating K Decision
School choice (S): S = I(XS' + s > 0) (1) S = 1 if attending a retention school; S = 0 otherwise; XS = vector of observables; s = unobservable; XS' + s =a child’s propensity of attending a retention school. Decision of repeating kindergarten (D*): D* = I(XD' + D > 0) (2) XD = vector of observables; D = unobservable; XD' + D = a child’s propensity of repeating kindergarten. Observed repeating kindergarten status (D); D = SD* = SI(XD + D > 0) (3) D=1 if repeating kindergarten; D=0 otherwise 27/29

30 Kept Back to get ahead?— Performance Outcome Equation
Suppose for each child, we can observe his test scores when he repeats kindergarten and when he does not. Denote the two potential outcomes as Y1 and Y0: Y1 = XY'1 + Y1 Y0 = XY'0 + Y0 The observed test score is Y = DY1 + (1  D)Y0 = D(XY'1 + Y1) + (1 D)(XY'0 + Y0) For simplicity, assume 1 and 0 are different only in constant terms, but not slopes  Y = XY' + D + DY1 + (1  D)Y (4) 28/29

31 Kept Back to get ahead?– The Full Model
A double hurdle model: S = I(XS + s > 0) (1) D*= I(XD + D > 0) (2) D = SD* (3) Y = XY + D + DY1 + (1  D)Y (4) Parameter of interest: ATT = E(Y1 – Y0 | D=1)=  + E(Y1  Y0 | D = 1) Develop a control function estimator Main results: Repeating K has positive but diminishing effects on academic performance over time; Controlling for unobservable makes a difference, simple matching method underestimates these effects. 29/29

32 ^_^ Thank you !!!


Download ppt "Yingying Dong Boston College 10/16/2008"

Similar presentations


Ads by Google