Printer Friendly

Incentives in obesity and health insurance.

Health insurance is widely regarded as essential for financing the production of good health, hut is insurance always benficial for our health? Ex ante moral hazard may induce individuals with insurance to engage in behaviors that they otherwise would not undertake in the absence of insurance. Using data from the 1993 2002 Behavioral Risk Factor Surveillance System, we attempt to isolate the effects of ex ante moral hazard to determine the potential consequence of having health insurance on measures of body weight, in our analyses, we control for a variety of confounding factors that may influence body weight and address the endogenous nature of health insurance. Our results suggest that having insurance is associated with higher body mass but not the probability of being obese.

**********

Health insurance is widely regarded as essential for financing the production of good health, but is insurance always beneficial for our health? Insurance reduces the monetary cost that individuals pay for health care; however, this reduction also can lead people to change their behaviors. This "'moral hazard" associated with health insurance can manifest itself not only by altering purchasing decisions, but also by changing other health-related behaviors. These two types of behavioral changes are termed "ex post moral hazard" and "ex ante moral hazard," respectively (Ehrlich and Becker 1972).

Ex post moral hazard arises through visiting a provider more frequently upon receiving health insurance. With ex ante moral hazard, the change in behavior occurs prior to physician contact. It results from engaging in riskier behaviors upon receiving health insurance, knowing that the option of visiting a physician is available. It is this ex ante moral hazard in particular that may be bad for one's health. In the absence of insurance, individuals have strong incentives to engage in behaviors that help prevent injury and illness--for example, eating nutritious foods, exercising regularly, and avoiding risky activities. In the presence of insurance, however, the incentives to engage in health-promoting behaviors are lessened as the costs incurred from being sick are lowered.

In the United States, the percentage of health care expenditures paid directly by consumers has been declining fairly consistently since the 1960s. Data from the Centers for Medicare and Medicaid Services (CMS) from 1960 to 2004 show this decline, revealing that the share of total personal health care expenditures paid for by consumers fell from 47% in 1960 to 13% in 2004. Personal health expenditures include payments for hospital, physician and other professional care, nursing home and home health care, durable medical equipment, and prescription drugs. At the same time, health care costs are increasing (now estimated to be rising twice as fast as inflation) and fewer people are being covered by health insurance. Individuals who lose insurance have incentives to engage in preventive, health-promoting activities, while those who have insurance and pay less out of pocket may have the opposite reaction. The research question in this paper is whether insurance status is associated with preventive health behaviors. We use the case of body weights in the United States to answer this question.

Body weight and obesity are good outcomes to study because plausibly weight may be affected by the availability of health insurance and the ex ante moral hazard problem. Most experts agree that body weight can be lowered with proper diet and exercise, making obesity and its associated conditions preventable by a change in behaviors not directly related to the receipt of medical care. Although the "disease" status of obesity is still debated, health insurance for the most part does not cover weight loss treatment and only in isolated cases does it cover gastric bypass surgery, which carries many risks and is recommended only for the morbidly obese, (1) Nevertheless, individuals may alter their behavior after receiving health insurance due to the host of obesity-related diseases, such as diabetes and heart disease, that are covered by health insurance.

One caveat to note is that the relationship between obesity and insurance can be confounded by the ex post moral hazard problem if insurance coverage encourages people to visit the doctor and they receive and follow advice to lose weight (Dave and Kaestner 2009). However, the extent to which physician advice is given and followed is debatable. Some studies have shown such counseling to be effective in promoting weight loss strategies (Kant and Miner 2007; Loureiro and Nayga 2006), while others have found physician counseling to have a minimal effect on the actual behavior of patients (Wee et al. 1999: Conway et al. 1995; Nagasawa et al. 1990; Clark 1991; Eraker, Kirscht, and Becker 1984; Ammerman et al. 1993). Patients may exhibit certain characteristics atypical of nonpatients, and physician counseling is not consistent across different demographic groups that exhibit similar ailments (Abid et al. 2005: Taira et al. 1997; Kreuter et al. 1997). To help guard against the confounding effects of doctor advice, we restrict the sample to those individuals who report no visits to a physician in the past year. While this restriction limits the potential effects of confounding factors, it renders our conclusions less generalizable to the population as a whole.

Obesity is defined by the National Institutes of Health as having a body mass index (BMI) of 30 kg/[m.sup.2] or greater. The percentage of individuals classified as obese has risen dramatically. Estimates using the National Health Examination Survey show that 12.7% of the U.S. population age 18 and older were obese in the early 1960s. The proportion rose slightly to 13.9% in the early 1970s and to 14% in the late 1970s. By the late 1980s and early 1990s, however, 21.6% of the population was classified as obese, and this number grew to an astounding 31.7% by 2004. Obesity carries many risks for a host of disorders, including heart disease, hypertension, stroke, cancer, depression, and blindness (Must et al. 1999: Mokdad et al. 2003; RNIB 2006).

Obesity is a national and global epidemic and has at its roots many potential causes. A variety of economic causes have been explored including reductions in job strenuousness (Philipson 2001: Lakdawalla and Philipson 2002), technological innovation in food processing and preparation (Cutler, Glaeser, and Shapiro 2003), the growing availability of restaurants (Chou, Grossman, and Saffer 2004; Rashad, Grossman, and Chou 2006), urban sprawl (Ewing et al. 2003), and time preference for the present (Komlos, Smith, and Bogin 2004; Smith, Bogin, and Bishai 2005; Zhang and Rashad 2008). Relatively few studies, however, have focused on the possible role of health insurance as a contributing factor to rising rates of obesity. We examine obesity in the context of a model in which the status of health insurance might play a role in determining body weights.

As discussed in more detail later, the relationship between health insurance and obesity status is complicated by structural endogeneity and the potential influence of other confounding factors such as work status and income. For example, individuals with higher incomes are less likely to be obese yet more likely to have health insurance. As a result, health-related behaviors may be positively correlated with health insurance status while not resulting from it. Yet, is it the case that these people would be even thinner if they had no health insurance, since they would not discount the future heavily when they were without insurance? Or would they instead be heavier without health care, as medical services are believed to improve health outcomes?

The net effect of health insurance on body weight is ambiguous. In general, existing theories regarding the production of health in the context of insurance may help guide predictions, but ultimately it is an empirical question. If health insurance has a negative causal influence on good health, then moral hazard may be a true concern. Yet if the opposite holds, this might lend further support for expanded or universal health insurance coverage due to the benefits that health insurance yields.

Using data from the Behavioral Risk Factor Surveillance System (BRFSS) from 1993 to 2002, we aim to uncover the effect that health insurance has on an individual's body weight and obesity status. We employ instrumental variable techniques to address the endogeneity of health insurance status. To account for variables affecting caloric intake and expenditure, which are likely to affect weight, we control for state-level variables such as prices for fast food and food at home, in line with recent work by Chou, Grossman, and Sailer (2004) and Rashad, Grossman, and Chou (2006).

Literature Review

The literature examining ex ante moral hazard is somewhat limited, with many of the studies examining the effects of health insurance coverage on the receipt of preventive services (Roddy, Wallen, and Meyers 1986; Lillard et al. 1986; Cherkin, Grothaus, and Wagner 1990: Card, Dobkin, and Maestas 2004). A few studies have examined health behaviors directly. For example, using data from the RAND Health Insurance Experiment, Newhouse (1993) looked at differences in BMI, levels of physical activities, smoking, and alcohol consumption among individuals enrolled in cost-sharing insurance plans and free plans. The results showed no difference in these behaviors between the two groups. Kenkel (2000) also found little evidence of a moral hazard effect in his analysis of individual behaviors using the 1990 National Health Interview Survey (NHIS). His analysis suggests that people with private health insurance are more likely to engage in health-promoting behaviors than those without insurance, with the one exception that men with health insurance are more likely to be obese. Kenkel stated that his results may be biased if omitted factors jointly determine insurance status and health practices.

Courbage and de Coulon (2004) examined the ex ante moral hazard question using data from the 2000/2001 wave of the British Household Panel Survey. Their outcomes of interest included smoking and frequency of exercising, defined as walking, swimming, or playing sports. Insurance in the United Kingdom is provided nationally to all residents; however, a secondary market exists where residents buy private insurance to avoid the waiting lists prevalent in the national insurance market. The authors used the purchase of this secondary insurance as their test of ex ante moral hazard. Using probits and an instrumental variables strategy, the authors found that having secondary insurance does not reduce preventive efforts and in fact may increase them. However, given that all residents are covered by the national insurance, these results are not surprising. Their analysis essentially tests the speed of receiving care, not the presence or generosity of insurance.

Card, Dobkin, and Maestas (2004) took a unique approach to examining the relationship between health behaviors and insurance by looking at smoking, exercise, and obesity among the near-elderly and the elderly. Eligibility for Medicare at age 65 was used as an exogenous measure of insurance coverage. Employing data from the 1999-2002 BRFSS surveys, they found that, in general, these health behaviors do not change with Medicare eligibility. They did show, however, that being age 65 or older is associated with a rise in the probability of being overweight or obese among blacks and low-educated minorities. Such a result is consistent with the ex ante moral hazard problem, but is not consistent with the authors' supposition that increased access to medical care will reduce poor health habits because doctors dispense advice on the health consequences of the behaviors. The authors dismissed the positive coefficients as a product of misspecification or sampling error since the results seemed to be driven by a downward dip in obesity just prior to age 65.

Dave and Kaestner (2009) also used eligibility for Medicare at age 65 as an exogenous measure of insurance coverage. They analyzed a sample of high school dropouts who were uninsured just prior to becoming eligible for Medicare. Controlling for the number of doctor visits, they found that exercise decreases, and smoking and drinking increase, as uninsured male high school dropouts become eligible for Medicare. They found no effects on these behaviors among women.

Bhattacharya and Sood (2007) addressed the obesity externality by looking at the current scenario where health insurance is not risk rated for obesity, showing that coverage therefore would shield people from the full costs of an unhealthy lifestyle. They estimated the welfare cost of obesity using the 1998 Medical Expenditure Panel Survey and the 1997 National Health Interview Survey. Excluding the uninsured, who they presume do not face the obesity externality, they estimated the increase in medical expenditures for insured people shifting from their optimal weights, and compared this value with predetermined costs for the uninsured. The authors suggested increasing the coinsurance rate, and also hinted at subsidizing a healthy lifestyle by reducing the welfare loss through technological change that decreases the costs of engaging in a healthy lifestyle. More recently, Bhattacharya and Packalen (2008) used the Medical Expenditure Panel Survey (MEPS) and the NHIS to analyze the ex ante moral hazard that arises in this context through a positive innovation externality.

Our paper adds to this current literature by examining the potential for ex ante moral hazard using recent data for nonelderly adults. While the randomized nature of the RAND study conducted by Newhouse (1993) may be the ideal sample design, those data were collected before the large rise in body weights seen today and the results may no longer be applicable. Kenkel (2000) worried about endogeneity in his study, and we address this. Card, Dobkin, and Maestas (2004) examined the behavior of the near-elderly and elderly: we focus on nonelderly adults. Our results confirm those of Kenkel (2000) and Card, Dobkin, and Maestas (2004) and show that having health insurance is associated with higher body weights, although there is no association with the probability of being obese. Results are more robust for those above the poverty threshold.

Methodology

Zweifel and Manning (2000) describe a model for ex ante moral hazard and discuss the determinants of the optimal amount of preventive effort exerted by an individual. This effort is determined by the probability of illness, the monetary loss from illness, labor supply, wages, health insurance coverage, sick pay, and insurance premiums. The benefit of engaging in prevention efforts is the decreased probability of suffering losses from illness, while its costs are the opportunity costs of engaging in prevention. In this model, prevention is measured in time units and the monetary costs of these efforts are ignored. However, such monetary costs would be included in the opportunity costs of prevention. One important result that comes from this model is the theoretical ambiguity of the effects of health insurance on the prevention effort. The level of insurance coverage affects premiums and these changes alter both the marginal costs and benefits of prevention. The net effect is ambiguous and therefore becomes an empirical question. (2)

The possibility of ex post moral hazard also must be considered in making predictions of the effects of health insurance on obesity status. This may arise if insurance coverage encourages people to visit the doctor, and the treatment they receive (perhaps in the form of advice) encourages weight loss (Dave and Kaestner 2009; Kant and Miner 2007; Loureiro and Nayga 2006). In this case, a negative relationship would arise between insurance coverage and obesity. On the other hand, there is some evidence of the minimal effectiveness of physician counseling on the diet and exercise behaviors of patients (Wee et al. 1999; Clark 1991: Ammerman et al. 1993). As noted earlier, to help ensure that ex post moral hazard does not confound our results, we limit our estimation sample to only those people who have not seen a doctor within the past year of the survey.

The relationship between obesity and health insurance is further complicated in that health status may determine insurance status, and other factors may influence or be influenced by both body weight and health insurance. For example, those who are obese are more likely to have certain illnesses or to seek insurance against their potential future maladies. Alternatively, obese people may have a time preference for the present (or discount the future more heavily than nonobese people) and choose not to have insurance. We use instrumental variables to avoid these confounding effects. The basic estimation equation is of the following form:

[Body weight.sub.i] = [[alpha].sub.0] + [[alpha].sub.1] [HealthIns.sub.i] + [[alpha].sub.2] [X.sub.i] + [[alpha].sub.3] [U.sub.i] + [[epsilon].sub.1] (1)

where i indexes individual observations, Body weight represents one of three measures of weight (discussed later), HealthIns is a dichotomous indicator for the presence of health insurance, and [X.sub.i] represents the vector of other relevant variables such as the probability of illness, the potential monetary loss from illness, labor supply, and wages. As subsequently discussed, we include measures for income and education, but unfortunately some of the variables that are important in the theoretical model are not available in existing data sets. While demographic and socioeconomic variables will help control for some of these unobserved factors, we recognize that many of these factors will remain unobserved in the error term.

Another problem to consider occurs when health insurance status is determined by weight:

[HealthIns.sub.i] = [[beta].sub.0] + [[beta].sub.1] [Body weight.sub.i] + [[beta].sub.2] [X.sub.i] + [[beta].sub.3] [F.sub.i] + [[beta].sub.4] [U.sub.i] + [[epsilon].sub.2], (2)

where the variables arc the same as in equation 1, and [F.sub.i] represents variables that predict health insurance status but not body weight. Given this, a simple estimation of equation 1 will yield a biased estimate of the coefficient on health insurance if there arc common unobservable factors ([U.sub.i]) influencing both weight ([[alpha].sub.3] [not equal to] 0) and health insurance ([[beta].sub.4] [not equal to] 0), which is analogous to an omitted variable bias, or if weight is a determinant of health insurance status ([[beta].sub.1] [not equal to] 0). Our estimation techniques attempt to address all of these sources of endogeneity. We discuss details later.

We empirically estimate equation 1 using a pooled cross-section of individuals over time. Our goal is to obtain a consistent estimate of the effect of health insurance on measures of body weight. Assuming we are able to avoid the problems of endogeneity, a positive coefficient is indicative of the presence of ex ante moral hazard: that is, having health insurance leads to unhealthy behaviors that contribute to larger body weights. A zero or negative coefficient will indicate the absence of any ex ante moral hazard effect.

Data

Our analysis uses 10 years of individual-level data from the Behavioral Risk Factor Surveillance System over the period 1993-2002. This is a repeated cross-section, and it is important to note that the same individuals are not followed over time, and thus changes in weight status and insurance status are not observed. As the largest telephone-based health survey available, the BRFSS has tracked health conditions and risk behaviors for adults in the United States since 1984. The survey is conducted by state health departments in collaboration with the Centers for Disease Control. Not all states were included in the early years of the data: however, 49 states plus the District of Columbia were included by 1993, our first year of analysis. We begin in 1993 and end in 2002 since these are the years for which information is available on all of our variables of interest. These data are publicly available from the Centers for Disease Control.

Information on self-reported body weight and height are available in all years of data. Using this information, we create some measures of weight: the body mass index, or BMI, a dichotomous indicator of being overweight or obese, and a dichotomous indicator of being classified as obese. BMI is defined as weight in kilograms divided by height in squared meters, and it is the measure that the National Institutes of Health use to track obesity over time. The dichotomous indicator of being overweight or obese is equal to 1 for individuals with a body mass index greater than or equal to 25 kg/[m.sup.2], and the dichotomous indicator for obesity is equal to 1 for individuals with a body mass index greater than or equal to 30 kg/[m.sup.2]. We also examine a dichotomous indicator of being overweight only--that is, anyone who is recorded as obese is excluded from the analysis, so the comparison is between being overweight and being normal weight or underweight.

While some measures of obesity, such as biometrical impedance analysis (BIA), may be more superior measures of obesity (Burkhauser and Cawley 2008; Wada and Tekin 2007), they are costly and are not routinely measured in physical examinations. The body mass index is a nationally representative figure that measures weight changes over time fairly accurately. To somewhat mitigate error due to self-reports, we use objective measures of weight and height from the third National Health and Nutrition Examination Survey (NHANES) to construct an adjusted, and more accurate measure of obesity. Because NHANES gathers information on both self-reported and actual weight and height, we adjust BMI in the BRFSS using this information. This is done separately by age, gender, and race, and has been employed previously (Chou, Grossman, and Saffer 2004; Cawley 1999). (3)

Health insurance in the BRFSS is measured by a dichotomous indicator for whether or not the individual has any kind of health care coverage, be it from private or public sources. The question is asked as follows: "Do you have any kind of health care coverage, including health insurance, prepaid plans such as HMOs, or government plans such as Medicare?" Unfortunately, the wording of the question does not allow us to differentiate between insured and uninsured spells, and the question may not necessarily reflect continuous insurance coverage throughout the year. At best we can say it represents a point-in-time estimate. We argue that this is an adequate variable for our analysis, as it includes those individuals who have had some type of insurance and who may engage in behavior reflective of ex ante moral hazard. We also note that our measure is similar to the question from the Current Population Survey (CPS) that is often used in measuring health insurance status (Gruber 2008). (4)

To further validate our insurance measure, we use the Medical Expenditure Panel Survey to compare results from a point-in-time indicator of insurance to an annual coverage indicator. To do this, we use data from 2002, which is the last year of our analysis. From MEPS, we compare a point-in-time estimate (from December 2002) to one indicating that the individual had health insurance in each month in 2002. The percentage of individuals with insurance for the point-in-time and year-round variables are 83.7% and 77.7%, respectively. We use these two measures to estimate equation l using the same techniques presented for the BRFSS. The MEPS regression results using the two different measures of insurance are qualitatively similar. The coefficients on the insurance status indicators are insignificant in almost all models, and in one case, negative and significant. Thus, the MEPS results point to no evidence of ex ante moral hazard. These results are available from the authors upon request.

Other individual characteristics in the BRFSS include the following variables: age and age squared; gender: race or ethnic category as represented by indicators for white (the omitted reference category), black, Hispanic, and other race; level of education as represented by dichotomous indicators for less than high school (the omitted reference category), some high school, high school degree, and college degree: family income and income squared; marital status; and the number of children under 18 in the household. (5) We limit our sample to individuals between the ages of 25 and 55. We exclude those under age 25 because the time preferences of these individuals may make their incentives and outcomes very different from older individuals. We exclude those older than 55 to avoid potential changes in behaviors brought on by the anticipated receipt of Medicare. In a supplemental analysis, we stratify the sample by poverty status, using yearly age- and family size-specific thresholds from the Bureau of Labor Statistics, to avoid potential changes in behaviors brought on by the anticipated receipt of Medicaid and due to the potential concern that those in poverty are already underinsured.

In line with Chou, Grossman, and Saffer (2004) and Rashad, Grossman, and Chou (2006), we also include in all models some state-level variables that have been shown to be important determinants of obesity status and body weights. These are state-level food, soft drink, and cigarette prices. These prices are obtained from the American Chamber of Commerce Researchers Association (ACCRA) and are given for various cities across the United States every quarter. The ACCRA food-at-home price is made up of a weighted average of 13 food prices, in which the weights are the reported average expenditure shares of these food items by consumers according to ACCRA. These 13 foods are: steak, beef, sausage, chicken, tuna, milk, eggs, margarine, cheese, potatoes, bananas, lettuce, and bread. The ACCRA fast-food price is formed by taking the average prices of a hamburger (McDonald's), a pizza (Pizza Hut), and fried chicken (KFC). (6) The price of a 2-liter bottle of Coca Cola is included as a proxy for soft drink prices. Cigarette prices are included due to the metabolic and appetite-suppressing effects that smoking may have. A cost of living index is also reported for each city. Before averaging prices in each state by quarter, we divide each price by the city's cost of living to account for regional variation in prices. The four quarters are then averaged, yielding a price for each state in each year. All annual prices are divided by the consumer price index, generating real prices in 1982 to 1984 dollars.

All models also include state and year indicator variables. The state indicators help to capture any unobserved time-invariant state effects that may influence obesity and may be correlated with health insurance status. Time dummies are included to capture secular trends in obesity.

Estimation

We use a variety of techniques to address the problems of endogeneity of health insurance in the body weight equation and the confounding effects of ex post moral hazard. Ultimately, we rely on instrumental variable techniques to draw conclusions, but restrictions on the sample help minimize the influence of confounding factors. That being said, restrictions on the sample limit the generalizability of our results.

The first restriction we place on the sample is that we limit it to employed individuals. This restriction is useful because it helps limit the amount of unobserved heterogeneity that may be correlated with the body weight measures and insurance status. The provision of health insurance is often tied to the labor market, and those who are not employed may have very different characteristics and incentives than employed individuals. Also, we need this restriction since the instruments we use the percentage of each state's workforce employed in firms of different sizes--works theoretically only for individuals who are employed. (More details on the instruments follow.) Note that many employed individuals, through the receipt of employer-provided

health insurance, already may pay indirectly for a large portion of their health insurance, rendering our estimates of ex ante moral hazard conservative. This is the case where individuals in the labor market receive lower wages in exchange for generous fringe benefits such as health insurance.

The second restriction limits the sample to those individuals who are classified as "healthy" and who have not visited the doctor in the past year. (7) Healthy individuals are defined as those who report that their general health is very good or excellent, and who do not report diabetes, high cholesterol, or any heart problems. (8) The healthy sample is considered because this is a group for which reverse causality, or structural endogeneity, is less likely to be an issue since healthy people are unlikely to purchase insurance for health reasons. In addition, this helps rule out preexisting condition clauses that might prevent an overweight or sick person from purchasing insurance. Limiting the sample to those who have not visited the doctor in the past year is important to ensure that our estimated coefficients measure the ex ante rather than the ex post moral hazard. In other words, we hope to eliminate the possibility that insurance coverage lowers body weight. This would occur if insurance encourages doctor visits that lead to treatment and advice regarding weight loss.

In all tables, ordinary least squares (OLS) and probit models provide baseline estimates. These are compared with models that directly account for the endogeneity of health insurance status.

When BMI is the dependent variable, OLS is used for the baseline model, followed by a two-stage least squares model. The instruments used are the percentage of each state's workforce employed in firms of sizes of 100 to 499 employees and 500+ employees. These annual workforce data come from the U.S. Small Business Administration. We believe that firm size is a useful instrument on a theoretical basis because health insurance is strongly tied to employment in the United States, and firm size is a known predictor of whether health insurance is offered to employees, with individuals in large firms more likely to have health insurance (Fronstin 2006). At first glance, the instruments appear to be valid. The coefficients in the first stage are positive, as predicted, and the F-statistic on their joint significance of 12.04 is significant and larger than the Bound, Jaeger, and Baker (1995) suggested value of 10. (First-stage results are available from the authors upon request.) The instruments also pass the overidentification test indicating that the instruments are uncorrelated with the error term and are properly excluded from the second-stage equation. However, the first stage partial R-squared is extremely low, indicating that the instruments are very weak. This is also evident by the fact that the two-stage least squares (TSLS) coefficient on health insurance is extremely large relative to the OLS coefficient and it becomes statistically insignificant. The Hausman test does not reject the consistency of OLS. These results make the TSLS estimates untrustworthy.

Lewbel (2007) presents an instrumental variables (IV) technique that is useful when valid external instruments are weak or unavailable. This procedure relies, in part, on the presence of heteroskedasticity in the error term of the first-stage equation. A Breusch-Pagan (1979) test confirms that this heteroskedasticity is present in our model. The Lewbel IV procedure is one of TSLS that uses (Z-[bar.Z])[[??].sub.2] as the identifying instruments. Here, Z is a vector of independent variables that may include any available excluded instruments, although such instruments are not a requirement and identification can be achieved without them. In addition, Z may include all independent variables or just a subset of them. [bar.Z] is a vector of means of the Z variables, and [[??].sub.2] is the residual from the first-stage regression (health insurance on the independent variables). (9) Lewbel shows that this instrument can identify the parameter of interest when Cov(Z,[[epsilon].sup.2.sub.2]) [not equal to] 0 and Cov (Z,[[epsilon].sub.1][[epsilon].sub.2])=0. The model can be estimated by TSLS or generalized method of moments (GMM), and the usual tests for the validity of the instruments can be applied. (10) We tried both estimation procedures and the results are nearly identical, so the TSLS are shown. Sabia (2007) successfully uses this procedure to identify the effects of body weight on academic performance among adolescents and finds the Lewbel IV results to be more plausible than the TSLS results that rely on instruments of questionable validity.

When the dichotomous indicators of weight are considered (overweight/obese, obese only, overweight only), probit estimates provide baseline and bivariate probits are used to account for the endogeneity of health insurance. Identification can be achieved in the bivariate probit without external instruments, although we caution that this only works well when the distribution assumption is correct (Monfardini and Radice 2007). Models were tested with and without the firm size instruments, but the results are insensitive to their inclusion. We only show the models with the instruments, but given their weakness we caution that the results will be biased if the assumption of joint normality is wrong.

Results

Table l shows sample means for the full sample and separate measures for those with and without health insurance. All three measures of body weight show statistically significant differences in values for those with and without health insurance; those with health insurance have a larger BMI and a higher probability of being classified as overweight. In contrast, the probability of being obese is slightly lower for those with insurance than those without. Of course these summary statistics do not account for any confounding factors. It is not surprising that the table of means also shows that people with health insurance are more educated, are older, tend to be married and have more children, and have higher incomes than those without health insurance.

Table 2 shows the results for BMI. The first column is the baseline OLS model. The second column uses a TSLS with the percentage of the states' workforces in firms of different sizes as instruments. Column 3 presents results from the Lewbel IV, with no external instruments, and finally, column 4 shows the Lewbel 1V with the external instruments. The coefficient on having a health plan is positive in all models, and is statistically significant in the OLS and Lewbel IV models. As discussed previously, the TSLS models are not trustworthy because of the weak instruments. (11) However, the Lewbel IV models appear to perform well. The instruments have strong first-stage F-statistics, pass the overidentification test, and the Hausman test rejects the consistency of the OLS coefficient. The magnitude of the Lewbel IV coefficient is not sensitive to the inclusion of the external instruments (which is not surprising given their low predictive power), and indicate that a switch from no health insurance to having health insurance is associated with an increase in the BMI of .25 kg/[m.sup.2]. To put this into context, consider an average male who is 5'10" tall and weighs 185 lbs. His BMI is 26.2. An increase in one unit of BMI translates into a weight gain of 7.1 pounds for this man, so a .25 unit increase

is a weight gain of about 1.8 pounds. (12) When the sample is stratified by poverty status, coefficients reveal that the statistically significant results are being driven by those above the poverty threshold and not by those officially classified as poor. (Stratified results are available from the authors upon request.)

Results for the remaining explanatory variables are as predicted for the most part. Those individuals who are college-educated, younger, female, single, and with fewer children on average have lower BMIs. These results are generally consistent across specifications.

To test the threshold effects, we next turn to an analysis of whether having health insurance is associated with the probabilities of being classified as: 1) overweight or obese, 2) overweight only, or 3) obese. Table 3 shows the results. The presence of health insurance is positively related to the probabilities of being overweight/obese and of being overweight. The effects appear to be concentrated on the threshold between normal weight and overweight since the models for obesity show no statistically significant effect of health insurance on the probability of being obese. All these conclusions hold whether the probit or bivariate probit is considered. Using the bivariate probit results, having insurance is associated with an increase of 11.1 percentage points in the probability of being overweight. The implication here is that there does appear to be an ex ante moral hazard effect, where individuals who have health insurance have less incentive to engage in preventive behaviors. However, the effect is small in terms of additions to BMI, and it is concentrated only along the normal overweight boundary. There is no apparent effect of health insurance on the probability of being obese. Results for the remaining explanatory variables are qualitatively similar to those for the model in which BMI is the dependent variable.

Discussion

Few would argue that health insurance is undesirable. The benefits of insurance to the health and welfare of individuals are highly valued, and are sometimes viewed as one of the basic human rights. But of course health insurance is not without costs, and the moral hazard problems associated with insurance add to these costs. This paper examines one particular manifestation of the moral hazard problem, the ex ante moral hazard as it pertains to body weight. Our hypothesis is that in the presence of insurance, people have less incentive to guard against illness and change their health-related behaviors (i.e., poor diet and less exercise) accordingly. Using a large data set of individuals, we estimate the relationship between health insurance status and body weight accounting for the possible endogeneity of health insurance coverage.

Our results suggest a small, but measureable ex ante moral hazard problem, particularly for employed individuals above the poverty threshold. Having health insurance is associated with an increase in BMI of .25 kg/[m.sup.2], or approximately 1.8 additional pounds on an average male. The presence of health insurance is also associated with an increase in the probability of being classified as overweight: however, there is no statistically significant effect on the probability of being classified as obese. We caution that the results from our study are not directly comparable to that of other studies, nor are they generalizable to the adult population because of the restrictions we place on the sample. We limit our sample to those who are employed, in good health, and have no reported doctor visits in the past year. This is done to minimize the propensity for reverse causality from body weight to health insurance status and to mitigate the potential for ex post moral hazard to confound our results. While not directly comparable, our results are in line with those obtained by Newhouse (1993) and Kenkel (2000), who both find little evidence of an adverse effect of health insurance on health behaviors.

In conclusion, our results demonstrate that health insurance can lead certain individuals to change health-related behaviors and to gain weight: however, the magnitude is small and the effect is concentrated only along the boundary of what is considered to be overweight. Obesity is not affected by the presence of health insurance. In other words, Americans are not getting fat because of their health insurance.

References

Abid, A., D. Galuska, L. K. Khan, C. Gillespie, E. S. Ford, and M. K. Serdula. 2005. Are Healthcare Professionals Advising Obese Patients to Lose Weight? A Trend Analysis. Medscape General Medicine 7(4): 10.

Ammerman, A. S., R. F. DeVellis, T. S. Carey, et al. 1993. Physician-Based Diet Counseling for Cholesterol Reduction: Current Practices, Determinants, and Strategies for Improvement. Preventive Medicine 22(1):96-109.

Bhattacharya, J., and M. Packalen. 2008. The Other Ex-Ante Moral Hazard in Health. NBER Working Paper No. 13863. Cambridge, Mass.: National Bureau of Economic Research.

Bhattacharya, J., and N. Sood. 2007. Health Insurance and the Obesity Externality. Advances in Health Economies and Health Services Research 17:279-318.

Bound, J., D. A. Jaeger, and R. M. Baker. 1995. Problems With Instrumental Variables Estimation When the Correlation between the Instruments and the Endogeneous Explanatory Variable is Weak. Journal of the American Statistical Association 90(430):443-450.

Breusch, T., and A. Pagan. 1979. A Simple Test for Heteroscedasticity and Random Coefficient Variation. Econometrica 47(5): 1287-1294.

Burkhauser, R., and J. Cawley. 2008. Beyond BMI: The Value of More Accurate Measures of Fatness and Obesity in Social Science Research. Journal of Health Economics 27(2):519-529.

Card, D., C. Dobkin, and N. Maestas. 2004. The Impact of Nearly Universal Insurance Coverage on Health Care Utilization and Health: Evidence from Medicare. NBER Working Paper No. 10365. Cambridge, Mass.: National Bureau of Economic Research.

Cawley, J. 1999. Addiction, Calories, and Body Weight. Ph.D. dissertation, University of Chicago.

Cherkin, D. C., L. Grothaus, and E. H. Wagner. 1990. The Effect of Office Visit Copayments on Preventive Care Services in an HMO. Inquiry 27(1):24-38.

Chou, S., M. Grossman, and H. SalTer. 2004. An Economic Analysis of Adult Obesity: Results from the Behavioral Risk Factor Surveillance System. Journal of Health Economics 23(3): 565-587.

Clark, L. T. 1991. Improving Compliance and Increasing Control of Hypertension: Needs of Special Hypertensive Populations. American Heart Journal 121 (2):664 669.

Conway, T., T. C. Hu, E. Mason, and C. Mueller. 1995. Are Primary Care Residents Adequately Prepared to Care for Women of Reproductive Age? Family Planning Perspectives 27(2):66-70.

Courbage, C., and A. de Coulon. 2004. Prevention and Private Health Insurance in the U.K. Geneva Papers on Risk and Insurance 29(4): 719-727.

Cutler, D., E. Glaeser, and J. Shapiro. 2003. Why Have Americans Become More Obese? Journal of Economic Perspectives 17(3):93-118.

Dave, D., and R. Kaestner. 2009. Health Insurance and Ex Ante Moral Hazard: Evidence from Medicare. International Journal of Health Care Finance and Economics. Published online March 11, 2009. DOI 10.1007/s10754-0099056-4.

Ehrlich, I., and G. Becker. 1972. Market Insurance, Self-Insurance and Self-Protection. Journal of Political Economy 80(4):623-648.

Eraker, S. A., J. P. Kirscht, and M. H. Becker. 1984. Understanding and Improving Patient Compliance. Annals of Internal Medicine 100(2):258-268.

Ewing, R., T. Schmid, R. Killingsworth, A. Zlot, and S. Raudenbush. 2003. Relationship between Urban Sprawl and Physical Activity, Obesity, and Morbidity. American Journal of Health Promotion 18(1):47-57.

Fronstin, P. 2006. Sources of Health Insurance and Characteristics of the Uninsured: Analysis of the March 2006 Current Population Survey. EBRI Issue Brief No. 298. Washington, D.C.: Employee Benefits Research Institute.

Gruber, J. 2008. Covering the Uninsured in the United States. Journal of Economic Literature 46(3):571 606.

Kant, A. K., and P. Miner. 2007. Physician Advice about Being Overweight: Association With Self-Reported Weight Loss, Dietary, and Physical Activity Behaviors of U.S. Adolescents in the National Health and Nutrition Examination Survey, 1999 2002. Pediatrics 119(1):142 147.

Kenkel, D. S. 2000. Prevention. In Handbook of Health Economies. Vol. 1B, A. J. Culyer and J. P. Newhouse, eds. Amsterdam: Elsevier.

Komlos, J., P. Smith, and B. Bogin. 2004. Obesity and the Rate of Time Preference: Is There A Connection? Journal of Biosocial Science 36(2):209-219.

Kreuter, M. W., D. P. Scharff, L. K. Brennan, and S. N. Lukwago. 1997. Physician Recommendations for Diet and Physical Activity: Which Patients Get Advised to Change? Preventive Medicines 26(6):825-833.

Lakdawalla, D., and T. Philipson. 2002. The Growth of Obesity and Technological Change: A Theoretical and Empirical Examination. NBER Working Paper No. 8946, Cambridge, Mass.: National Bureau of Economic Research.

Lewbel, A. 2007. Using Heteroskedasticity to Identify and Estimate Mismeasured and Endogenous Regressor Models. Boston College Working Paper.

Lewis, K., M. Ellwood, and J. L. Czajka. 1998. Counting the Uninsured: A Review of the Literature. Urban Institute Occasional Paper No. 8. Washington, D.C.: The Urban Institute.

Lillard, L. A., W. G. Manning, C. E. Peterson, N. Lurie, G. A. Goldberg, and C. E. Phelps. 1986. Preventive Medical Care: Standards, Usage, and Efficacy. RAND publication R-2366-HCFA.

Loureiro, M. L., and R. M. Nayga. 2006. Obesity, Weight Loss, and Physician's Advice. Social Science and Medicine 62(10):2458-2468.

Monfardini, C., and R. Radice. 2007. Testing Exogeneity in the Bivariate Probit Model: A Monte Carlo Study. Oxford Bulletin of Economics amt Statistics 70(2):271-282.

Mokdad, A. H., E. S. Ford, B. A. Bowman, W. H. Dietz, F. Vinicor, V. S. Bales, and J. S. Marks. 2003. Prevalence of Obesity, Diabetes, and Obesity-Related Health Risk Factors, 2001. Journal of the American Medical Association 289(1):76 79.

Must, A., J. Spadano, E. H. Coakley, A. E. Field, G. Colditz, and W. H. Dietz. 1999. The Disease Burden Associated With Overweight and Obesity. Journal o/the American Medical Association 282(16):1523-1529.

Nagasawa, M., M. C. Smith, J. H. Barnes, and J. E. Fincham. 1990. Meta-Analysis of Correlates of Diabetes Patients' Compliance With Prescribed Medications. Diabetes Education 16(3): 192-200.

Newhouse, J. P. 1993. Free for All? Lessons from the RAND Health Insurance Experiment. Cambridge, Mass.: Harvard University Press.

Philipson, T. 2001. The World-Wide Growth in Obesity: An Economic Research Agenda. Health Economics 10(1):1-7.

Rashad, I., M. Grossman, and S. Chou. 2006. The Super Size of America: An Economic Estimation of Body Mass Index and Obesity in Adults. Eastern Economic Journal 32(1):133-148.

Royal National Institute of the Blind. 2006. Obese People Twice As Likely to Go Blind. Press release, September 6.

Roddy, P. C., J. Wallen, and S. M. Meyers. 1986. Cost Sharing and Use of Health Services: The United Mine Workers of America Health Plan. Medical Care 24(9):873-876.

Sabia, J. J. 2007. The Effect of Body Weight on Adolescent Academic Performance. Southern Economic Journal 73(4):871-900.

Smith, P. K., B. Bogin, and D. Bishai. 2005. Are Time Preference and Body Mass Index Associated? Evidence from the National Longitudinal Survey of Youth. Economics and Human Biology 3(2):259-270.

Taira, D. A., D. G. Safran, T. B. Seto, W. H. Rogers, and A. R. Tarlov. 1997. The Relationship between Patient Income and Physician Discussion of Health Risk Behaviors. Journal of the American Medical Association 278(17): 1412-1417.

Wada, R., and E. Tekin. 2007. Body Composition and Wages. NBER Working Paper No. 13595. Cambridge, Mass.: National Bureau of Economic Research.

Wee, C. C., E. P. McCarthy, R. B. Davis, and R. S. Phillips. 1999. Physician Counseling about Exercise. Journal of the American Medical Association 282(16): 1583-1588.

Zhang, L., and I. Rashad. 2008. Time Preference and Obesity: The Health Consequences of Discounting the Future. Journal of Biosocial Science 40(1):97-113.

Zweifel, P., and W. G. Manning. 2000. Moral Hazard and Consumer Incentives in Health Care. In Handbook of Health Economics, A. J. Culyer and J. P. Newhouse, eds. Amsterdam: Elsevier.

Notes

The authors thank Akin Monheit, Christopher Ruhm. Susan Averett, Caroline Carl&, an anonymous referee, and participants at the 2006 Southern Economic Association, 2007 Eastern Economic Association, and 2008 American Society of Health Economists meetings for helpful comments. The authors alone are responsible for errors.

(1) In November 2005, the Centers for Medicare and Medicaid Services proposed national Medicare coverage for bariatric surgery procedures. See the U.S. Department of Health and Human Services website at http://www.cms.hhs.gov/apps/media/press/release.asp? Counter- 1733 for more details.

(2) See Zweifel and Manning (2000) for details.

(3) We find that the correlation between BMI and adjusted BMI is .99. Regression results using BMI and adjusted BMI are also very similar.

(4) It is unclear whether the CPS measures a point-in-time estimate or a year-round one. As Gruber (2008) notes, "A problem with the CPS estimate is that it is a strange hybrid of a point-in-time estimate and a backwards look at the previous year." Lewis, Ellwood, and Czajka (1998) note that "the CPS probably contains a mixed bag of reporting that is, some respondents report health insurance status during the previous year, some report it as of the interview date, and some fail to report it altogether which, in the end, yields estimates that are somewhere between a point-in-time and an uninsured-throughout-the-year estimate."

(5) We recognize that some of these variables may be endogenous as well. Models were tested that excluded the potentially endogenous variables and the conclusions remain the same.

(6) More detail on these variables can be found in Chou, Grossman, and Saffer (2004).

(7) Restricting the sample to employed people reduces the sample size from 868,136 to 721,246. Further restricting the sample to healthy individuals further reduces it to 109,788.

(8) We realize that this is not a perfect stratification, as respondents may not be fully aware of their health status if they have not seen a doctor in the year prior to being interviewed.

(9) This model assumes that [[beta].sub.1] = 0 in equation 2. We believe that restricting the sample to healthy individuals with no doctor visits in the past year justifies this assumption.

(10) More detail on this method can be found in Lewbel (2007).

(11) An IV model was tested using limited information maximum likelihood (LIML), which may perform better than TSLS with weak instruments. The results of the two estimation procedures were almost identical.

(12) Given that some of the previous literature has found differential results by gender, we re-estimated all models for males and females separately. The results are very similar in sign, significance, and magnitude to the results for both genders combined. These tables are available upon request.

Inas Rashad Kelly, Ph.D., is an assistant professor in the Department of Economics at Queens College of the City University of New York. and a faculty research fellow in the Health Economics Program of the National Bureau of Economic Research. Sara Markowitz, Ph.D., is an associate professor in the Department of Economics at Emory University. and a research associate in the Health Economics Program of the National Bureau of Economic Research. Address correspondence to Proof Kelly at Department of Economics. Queens College of the City University of New York. 300A Powdermaker Hall. 65-30 Kissena Boulevard. Flushing. N Y 11367. Email: huts. Kelly@qc.cuny.edu
Table 1. Weighted sample means and proportions

                                                    All respondents
Variable           Description                        (n=109,788)

BMI                Body mass index, measured as      26.13 (4.42)
                   weight in kilograms divided by
                   height in squared meters

Overweight/obese   Dichotomous variable that              .56
                   equals 1 if BMI is equal to or
                   greater than 25 kg/[m.sup.2]

Overweight (a)     Dichotomous variable that              .47
                   equals 1 if BMI is equal to or
                   greater than 25 kg/ [m.sup.2]
                   and less than 30 kg/[m.sup.2]

Obese              Dichotomous variable that              .16
                   equals 1 if BMI is equal to or
                   greater than 30 kg/[m.sup.2]

Health insurance   Dichotomous variable that              .84
                   equals 1 if respondent has
                   some form of health insurance
                   coverage

High school        Dichotomous variable that              .28
                   equals 1 if respondent
                   completed exactly 12 years of
                   formal schooling

Some college       Dichotomous variable that              .29
                   equals 1 if respondent
                   completed at least 13 years
                   but fewer than 16 years of
                   formal schooling

College            Dichotomous variable that              .39
                   equals 1 if respondent
                   graduated from college

Age                Age of respondent                 38.43 (7.90)

Black              Dichotomous variable that              .04
                   equals 1 if respondent is
                   black but not Hispanic

Hispanic           Dichotomous variable that              .05
                   equals 1 if respondent is
                   Hispanic

Other race         Dichotomous variable that              .03
                   equals 1 if respondent is not
                   white, black, or Hispanic

Male               Dichotomous variable that              .64
                   equals 1 if respondent is male

Number of          Number of children in the          1.07 (l.24)
children           household under age 18

Real family        Real household income in          35.65 (26.97)
income             thousands of 1982-84 dollars

Married            Dichotomous variable that              .60
                   equals 1 if respondent is
                   married

Divorced           Dichotomous variable that              .18
                   equals 1 if respondent is
                   divorced or separated

Widowed            Dichotomous variable that              .01
                   equals 1 if respondent is
                   widowed

Food at home       Real state ACCRA food at home      1.03 (.05)
price              price divided by (the cost of
                   living*the CPI) in 1982-84
                   dollars

Fast food price    Real state ACCRA fast food         2.72 (.18)
                   price divided by (the cost of
                   living*the CPI) in 1982-84
                   dollars

Soda price         Real state ACCRA Coke price         .71 (.09)
                   divided by (the cost of
                   living*the CPI) in 1982-84
                   dollars

Cigarette price    Real state ACCRA cigarette        12.97 (2.91)
                   price divided by (the cost of
                   living*the CPI) in 1982-84
                   dollars

                   Without health    With health
                     insurance        insurance
Variable             (n=17,066)      (n=92,722)

BMI                 26.03 (4.79)    26.15 (4.34)

Overweight/obese        .53              .56

Overweight (a)          .43              .48

Obese                   .17              .16

Health insurance         --              --

High school             .38              .26

Some college            .30              .28

College                 .21              .42

Age                 37.33 (8.13)    38.63 (7.85)

Black                   .06              .04

Hispanic                .08              .04

Other race              .04              .03

Male                    .61              .65

Number of           1.00 (l.30)      1.09 (l.23)
children

Real family        19.87 (17.49)    38.56 (27.40)
income

Married                 .40              .64

Divorced                .28              .16

Widowed                 .02              .01

Food at home         1.03 (.05)      1.03 (.05)
price

Fast food price      2.73 (.18)      2.72 (.18)

Soda price           .71 (.09)        .71 (.09)

Cigarette price     12.98 (2.89)    12.97 (2.91)

Note: Standard deviations are in parentheses. Difference between
those with health insurance and those without health  insurance is
statistically significant at the 10% level for all variables except
the food at home price and cigarette price.

(a) Sample omits individuals classified as obese. N=92,086.

Table 2. Effects of health insurance on BMI

                                 (1)                 (2)

                                 OLS                 IV

Health insurance           .147 *** (3.78)        2.270 (.85)
High school                   -.090 (1.32)        -.320 (1.08)
Some college                  -.113 (1.64)        -.392 (1.10)
College                   -.695 *** (9.97)    -1.045 ** (2.34)
Age                        .071 *** (4.29)      .057 ** (2.26)
Age squared                  -.0002 (1.15)       -.0001 (.23)
Black                     1.246 *** (18.47)   1.265 *** (17.53)
Hispanic                   .688 *** (10.76)    .722 *** (9.30)
Other race                -.619 *** (8.27)    -.582 *** (6.57)
Male                      1.309 *** (47.36)   1.326 *** (37.99)
Number of children         .087 *** (7.07)     .099 *** (4.97)
Income                        -.004 (1.46)        -.041 (.88)
Income squared               .00001 (.25)         .0003 (.81)
Married                    .142 *** (3.60)         .012 (.07)
Divorced                  -.390 *** (8.66)    -.403 *** (8.31)
Widowed                        .021 (.16)         -.022 (.15)
Food at home price            -.042 (.07)          .104 (.17)
Fast food price               -.065 (.34)         -.004 (.02)
Soda price                     .050 (.13)         -.069 (.16)
Cigarette price              .029 * (1.73)         .022 (1.17)
Observations                     109788              109788
F-test on instruments                             12.04 [.000]
Overidentification test                            .534 [.465]
Hausman test                                        .58 [.447]

                                 (3)                  (4)
                            Lewbel IV, no     Lewbel IV, with
                             instruments        instruments

Health insurance           .247 *** (4.04)     .249 *** (4.06)
High school                   -.101 (1.47)        -.101 (1.47)
Some college                -.126 * (1.82)      -.127 * (1.82)
College                   -.712 *** (10.15)   -.712 *** (10.15)
Age                        .071 *** (4.25)     .071 *** (4.25)
Age squared                  -.0002 (1.11)       -.0002 (1.11)
Black                     1.247 *** (18.48)   1.247 *** (18.48)
Hispanic                   .690 *** (10.79)    .690 *** (10.79)
Other race                -.617 *** (8.25)    -.617 *** (8.25)
Male                      1.310 *** (47.38)   1.310 *** (47.38)
Number of children         .087 *** (7.11)     .087 *** (7.11)
Income                     -.006 ** (2.03)     -.006 ** (2.04)
Income squared               .00002 (.82)        .00002 (82)
Married                    .136 *** (3.43)     .136 *** (3.43)
Divorced                  -.390 *** (8.67)    -.390 *** (8.67)
Widowed                        .019 (.14)          .019 (.14)
Food at home price            -.035 (.06)         -.035 (.06)
Fast food price               -.062 (.32)         -.062 (.32)
Soda price                     .045 (.11)          .045 (.11)
Cigarette price              .028 * (1.71)       .028 * (1.71)
Observations                    109788              109788
F-test on instruments        951.92 [.000]       906.24 [.000]
Overidentification test      94.845 [.0943]      97.773 [.1128]
Hausman test                   4.49 [.034]         4.58 [.032]

Note: Absolute value of r-statistics in parentheses, p-values in
brackets, and intercept not shown. Models also include state
indicators, year indicators, and missing observation indicators
for price variables. Instruments are the percentage of the
workforce in firms of sires 100-499 and 500+ workers.

* Significant at 10%;  ** significant at 5%; *** significant at 1%.

Table 3. Marginal effects of health insurance on probabilities of
being overweight and obese

                              (1)                 (2)

                                         Overweight/obese

                             Probit             Biprobit

Health insurance        .019 *** (4.13)     .093 *** (4.07)
High school               .015 * (1.92)         .008 (.90)
Some college                .006 (.77)         -.003 (.39)
College                -.055 *** (6.75)    -.067 *** (7.53)
Age                     .008 *** (3.85)     .007 *** (3.56)
Age squared              -.00002 (.98)       -.00002 (.73)
Black                   .100 *** (12.90)    .101 *** (12.96)
Hispanic                .093 *** (12.63)    .094 *** (12.76)
Other race             -.083 *** (9.42)    -.082 *** (9.25)
Male                    .210 *** (65.02)    .210 *** (65.11)
Number of children      .008 *** (5.62)     .009 *** (5.89)
Income                  .001 *** (4.20)       .00001 (.02)
Income squared       -.00001 *** (4.14)     -.000001 (.25)
Married                 .043 *** (9.25)     .038 *** (7.90)
Divorced                -.012 ** (2.24)     -.012 ** (2.33)
Widowed                     .016 (1.02)         .014 (.92)
Food at home price         -.026 (.38)         -.021 (.30)
Fast food price            -.032 (1.40)        -.030 (1.30)
Soda price                  .010 (.22)          .006 (.13)
Cigarette price             .002 (.95)          .002 (.83)
Observations                 109,788            109,788
Rho                                              -.106
[Chi.sup.2] test
  of rho=0                                    10.643 [.001]

                            (3)                 (4)

                                       Overweight

                           Probit             Biprobit

Health insurance        .020 *** (4.01)    .111 *** (4.52)
High school             .032 *** (3.51)     .022 ** (2.32)
Some college             .021 ** (2.32)         .009 (.93)
College                -.029 *** (3.15)    -.044 *** (4.42)
Age                     .007 *** (3.29)    .007 *** (3.01)
Age squared              -.00003 (1.03)      -.00002 (.78)
Black                   .070 *** (7.75)    .070 *** (7.80)
Hispanic                .094 *** (11.24)    .095 *** (11.41)
Other race             -.077 *** (8.18)    -.075 *** (8.00)
Male                    .243 *** (69.02)    .243 *** (69.01)
Number of children      .006 *** (3.59)    .006 *** (3.90)
Income                  .003 *** (7.73)      .001 * (1.82)
Income squared       -.00002 *** (6.99)   -.00001 * (1.76)
Married                 .052 *** (10.17)    .046 *** (8.56)
Divorced                    .007 (1.25)        .007 (1.13)
Widowed                     .018 (1.02)          .015 (89)
Food at home price         -.050 (.65)        -.046 (.61)
Fast food price         -.054 ** (2.14)    -.050 ** (1.99)
Soda price                 -.003 (.05)        -.008 (.16)
Cigarette price            .0003 (.14)       .00002 (.01)
Observations                 92,086             92,086
Rho                                               -.132
[Chi.sup.2] test
  of rho=0                                   13.857 [.000]

                            (5)                 (6)

                                      Obese

                          Probit             Biprobit

Health insurance           .004 (1.30)          .016 (.84)
High school           -.015 *** (2.80)     -.016 *** (2.84)
Some college          -.016 *** (2.85)     -.017 *** (2.86)
College               -.052 *** (9.56)     -.054 *** (8.67)
Age                     .003 ** (1.98)        .003 * (1.91)
Age squared            -.000001 (.04)       .0000003 (.02)
Black                  .077 *** (12.86)     .077 *** (12.88)
Hispanic               .035 *** (6.25)      .035 *** (6.28)
Other race            -.033 *** (5.09)     -.032 *** (5.05)
Male                   .011 *** (4.60)      .011 *** (4.63)
Number of children     .005 *** (4.71)      .005 *** (4.75)
Income                -.001 *** (5.87)     -.002 *** (3.74)
Income squared       .00001 *** (4.35)    .00001 *** (3.06)
Married                    .005 (1.45)           .004(l.15)
Divorced              -.029 *** (7.90)     -.029 *** (7.92)
Widowed                     .002 (.15)          .001 (.12)
Food at home price          .039 (.78)          .040 (.80)
Fast food price             .024 (1.40)         .024 (1.42)
Soda price                  .031 (.90)          .030 (.88)
Cigarette price          .003 ** (1.99)       .003 * (1.96)
Observations                109,788              109,788
Rho                                                -.028
[Chi.sup.2] test
  of rho=0                                      .397 [.528]

Note: Absolute value of t-statistics in parentheses, p-values in
brackets, and intercept not shown. Marginal effects reported. Models
also include state indicators, year indicators, and missing
observation indicators for price variables. Instruments are the
percentage of the workforce in firms of sizes 100-499 and 500+
workers.

* Significant at 10%; ** significant at 5%; *** significant at 1%.
COPYRIGHT 2009 Excellus Health Plan, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2009 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Kelly, Inas Rashad; Markowitz, Sara
Publication:Inquiry
Geographic Code:1USA
Date:Dec 22, 2009
Words:9772
Previous Article:Health care spending and service use among high-cost Medicaid beneficiaries, 2002-2004.
Next Article:Spillover effects of state mandated benefit laws: the case of outpatient breast cancer surgery.
Topics:

Terms of use | Privacy policy | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters