Diabetes Treatments

People with Diabetes to Protect Themselves, Wear Medical Identification diabetes, asthma or a heart condition, as well as …

A Micro Analysis of The Effect of Insurance Mandates on the Behavior of Diabetics: Education vs Moral Hazard

Jonathan Klick American Enterprise Institute jklick@aeiorg Thomas Stratmann George Mason University tstratma@gmuedu

September 11, 2003

Abstract: In the face of rising diabetes rates, many states passed laws requiring health insurance plans to cover medical treatments for the disease As with most insurance, this coverage has the potential to induce moral hazard since the reduction in the cost associated with diabetes lowers the incentive for individuals to eat healthfully and exercise However, there are two potentially countervailing forces First, insurers have the incentive to encourage individuals at risk of developing diabetes to engage in preventive behavior thereby reducing the expected cost of future treatment required by the mandates Second, the mandates require coverage of selfmanagement education for diabetics which has the potential to improve the exercise and diet habits of diabetics Thus, the net effect of these mandates on behavior is ambiguous Using individual level body mass index BMI data as an indicator of healthfulness, we show that diabetics subject to
these mandates exhibit better health characteristics relative to individuals in states that do not have mandates Also, within states with mandates, diabetics covered by the mandates have lower BMIs than individuals not covered by the mandates JEL Classification: I12; I18; J32; J38 Keywords: Obesity; Diabetes; Moral Hazard; Insurance; Mandates

1 Introduction Diabetes is a growing concern in the United States The Centers for Disease Control estimates that more than 17 million people have diabetes and the incidence of the disease has been growing throughout the past decade CDC 2003 Among the complications induced by the disease are blindness, kidney disease, amputations, cardiovascular disease, and a host of other life-threatening problems, placing diabetes as the fifth leading cause of death in the United States The American Diabetes Association estimates that the total cost of diabetes in 2002 in terms of direct medical care and indirect productivity losses amounted to 132 billion in the US ADA 2003 Additionally, analysts estimate that there are another 12 million Americans with a condition known as pre-diabetes Benjamin, Valdez, Geiss, Rolka, and Narayan 2003 Prediabetes is a
condition covering individuals who are at a high risk for developing Type-2 diabetes1 The upward trend of obesity witnessed over the past two decades suggests that the incidence of diabetes and pre-diabetes will continue to grow Mokdad, Ford, Bowman, Dietz, Vinicor, Bales, and Marks 2003 In this context, the legislatures of 46 states have passed laws mandating that health insurance providers cover supplies, services, medications, and equipment for treating diabetes as

The phase between normal blood sugar levels and levels denoting Type-2 diabetes is classified as impaired glucose tolerance IGT or impaired fasting glucose IFG With IGT, the blood sugar level is elevated in the range of 140 to 199 milligrams per deciliter after a two-hour oral glucose tolerance test but does not meet the standard for a Type-2 diabetes diagnosis With IFG, the fasting blood sugar level is elevated in the range of 110 to 125 milligrams per deciliter after an overnight fast but does not reach the Type-2 diabetes threshold 1

1

part of their basic coverage without charging higher premiums for the coverage2 Given the high cost of diabetes treatments, advocates such as the American Diabetes Association
view these mandates are necessary for ensuring that diabetics receive adequate health care As with most insurance coverage, these mandates have the potential to induce moral hazard problems That is, because Type-2 diabetes can largely be avoided through fastidious diet and exercise regimens, individuals facing the costs associated with diabetes have strong incentives to engage in healthful behavior When the cost of medical treatments declines because of state mandates, the relative cost of behavioral prevention increases, inducing individuals to engage in worse diet and exercise practices On the margin, this moral hazard increases the obesity incidence and eventually the diabetes incidence However, insurers have a countervailing interest in limiting the risk exposure placed upon them by diabetes mandates Because the cost of diabetes treatments is high and insurers are limited by the statutes from pricing this increased risk into their premiums, insurance plans might have the incentive to mitigate their risk by encouraging individuals to avoid the behavior that leads to diabetes Through spending on diabetes education and wellness incentives targeted at individuals at risk for
developing diabetes, insurers might counteract the moral hazard created by the diabetes mandates Further, for individuals with diabetes, the mandates include coverage for self management and education programs Mandated coverage for testing supplies has the potential to give diabetics improved awareness of their condition, inducing them to be more vigilant in

Information on which states have diabetes mandates and the year of passage is provided in Table 1 2

2

their behavior The education provisions of mandates might improve access for diabetics to dieticians and diabetes educators Thus, the net effect of these mandates on individual health is ambiguous In this paper, we examine the health effects of diabetes mandates by focusing on individuals body mass indexes BMI for the period 1996-2001, comparing BMIs of individuals in states with mandates to BMIs for individuals in non-mandate states Further, within mandate states, we compare individuals affected by the mandates to individuals for whom the mandates do not apply We find no evidence of moral hazard related to diabetes mandates We also find no evidence that mandates induce insurers to be pro-active in terms of encouraging the
non-diabetic population to engage in preventive behavior Although non-diabetic individuals affected by mandates exhibit lower BMIs relative to unaffected individuals, it appears as though mandate adoption is endogenous to the health characteristics of the affected population For diabetics, however, the data indicate that mandates do lead to more healthful behavior, with affected diabetics exhibiting significantly lower BMIs than unaffected diabetics This result is not an artifact of simultaneity In section 2 of the paper, we discuss the existing literature on the economics of obesity and diabetes Section 3 provides the theoretical context for the expected effect of diabetes mandates on behavior Section 4 discusses our data and research design Results are presented in section 5, followed by concluding remarks

2 The Economics of Obesity and Diabetes

3

Perhaps owing to the recent trends in body weight, the topic of obesity has gained much attention in the economics literature lately Philipson and Posner 1999 argue that the increase in obesity witnessed in the US is a function of technological progress That is, as technology has lowered the price of food and has reduced the amount
of on-the-job exercise that typically takes place in modern American occupations, individuals consume relatively more calories compared to the calories the expend than they did in the past This net increase in caloric intake more than offsets the effects of increased dieting and recreational exercise In an extension of the basic Philipson and Posner framework, Lakdawalla and Philipson 2002 test the major implications of the technological model of obesity They find strong evidence that lower food prices, resulting from improvements in agricultural technology, do lead to a statistically significant increase in body weights Further, they provide some evidence that declining occupational physical activity is also an important contributor to the increase in body weights In a similar vein, Chou, Grossman, and Saffer 2002 use state data on the number of restaurants in an individuals home state, as well as information regarding the price of meals in various restaurants, to explain a large proportion of the variation in individuals BMIs Although, as the authors admit, this approach potentially suffers from a simultaneity bias, their results suggest that individuals facing markets with
relatively many restaurants and low food prices exhibit higher BMIs and obesity incidence Cutler, Glaeser, and Shapiro 2003 also adopt the technological explanation for the rise in obesity, but they focus on the distribution of the increases in body weights They identify that the biggest technologically-based increase in calorie consumption is exhibited in the heavy tail

4

of the weight distribution That is, the increases in weight have been most pronounced for relatively heavy individuals To explain this, they invoke a self-control model in which overweight individuals have difficulties limiting their consumption when food prices decrease They argue that price decreases are actually welfare reducing for this segment of the population Although obesity is linked with a host of physical problems, its connection with diabetes is especially strong In fact, Type-2 diabetes is almost completely limited to the overweight and obese This implies that the economic models of obesity also indirectly apply to diabetes Diabetes does present some interesting questions that are distinct from the general issue of obesity Specifically, while exercise and healthful diets can lower the likelihood of
both obesity and diabetes, there are also medical substitutes for these behavioral treatments in the case of diabetes Kahn 1999 highlights how both behavioral modifications and medical treatments have significantly improved the quality of life for diabetics One particular concern for Kahn is the possibility that diabetic individuals substitute medical treatments for behavioral modifications That is, do medicated diabetics become less fastidious in various behaviors which increase their chances of developing complications from diabetes, such as smoking and eating behaviors? While Kahn finds no evidence of this substitution in his analysis, he notes that clinical diabeticians express concern that improved access to medications for diabetes might lull individuals into a false sense of security, causing them to ignore behavioral prescriptions Similar offsetting behavior has been documented in many other contexts in the economics literature3 In the case of diabetes, the possibility of offsetting behavior raises questions about the ultimate aggregate effect of increasing access to medical treatments for

3

See, for example, Peltzman 1975 and Viscusi 1984 5

diabetes Specifically, since
complications from diabetes represent the costs of poor health habits, the prospect of developing diabetes induces individuals, on the margin, to engage in more healthful behavior Laws requiring insurers to cover medical treatments for diabetes effectively subsidize less healthful behavior, potentially leading more individuals to develop pre-diabetes and diabetes than would be the case in the absence of these laws

3 Offsetting Behavior and the Insurers Response For simplicity we model an individuals food consumption decision assuming that the only cost of eating is an increase in the likelihood that the individual will develop diabetes That is, an individual chooses to eat f identical units of food which imposes no financial cost on him, but consumption does increase the probability p he will develop diabetes4 Diabetes brings a utility cost of D which is independent of the consumption decision and is a decreasing function of available medical treatments m5 The individual faces the following maximization problem:
Max U f - p f D m
f

yielding the following first order condition:

U p - D m 0 f f

Allowing for choices among foods with varying levels of healthfulness would not
change the primary result of this model Allowing for varying losses from diabetes, either instead of or in addition to allowing p to be a function of f, does not change the model qualitatively 6
5

4

which implies that the individual eats up to the point where his marginal increase in utility from food consumption equals the marginal increase in the probability that he will develop diabetes multiplied by the utility cost of diabetes If the individual experiences a positive exogenous shock in access to medical treatments, the effect on his choice of f is represented by:

-p D f m - 2 2 U 2 - p 2 f f

The effect of an increase in medical treatments is to increase the individuals food intake, as long as utility is concave in food consumption and the probability function has a constant or increasing slope with respect to food consumption This implies that the prospect of improved diabetes treatment access will induce a worsening in the diet and, by extension, the exercise habits of individuals who fear the prospect of developing diabetes Mandates requiring that medical treatments for diabetes are included in the basic insurance coverage effectively increase access to
those treatments Thus, we might expect that mandates produce deleterious health effects However, given the high cost of providing medical treatments for diabetes, insurers have the incentive to intervene ex ante to provide countervailing incentives and support to their customers who are at risk for developing diabetes6 Because mandates restrict insurers from pricing the diabetes risk into their premiums, insurers might engage in active preventive
6

Peele, Lave, and Songer 2002 estimate that healthcare expenditures by insurers were three times higher for diabetics compared to all consumers in the examined health plans 7

management to mitigate the risk posed by diabetes mandates Active management has the potential to reap large cost savings with respect to diabetes since behavioral modifications significantly reduce diabetes incidence In one study, Hu, Manson, Stampfer, Colditz, Liu, Solomon, and Willett 2001 find that more than 90 percent of cases of Type-2 diabetes could be prevented by the adoption of healthier lifestyles For diabetics specifically, another important element of the insurance mandates involves the coverage of self-management supplies and education programs
Improving access to devices that monitor an individuals blood sugar level has the potential to make diabetics more aware of their condition, improving their compliance with the diet and exercise directives issued by doctors Further, covering the cost of education programs could make a doctor more likely to suggest that a patient visit a professional dietician or diabetes educator Even if doctors regularly suggest education programs, insurance coverage might make it more likely that patients will follow through on the suggestion Guglielmo 2001 However, with respect to self-management and education, if these options are effective in improving the behavior of diabetics, arguably, insurers would be likely to cover them even in the absence of a mandate As indicated above, complications from diabetes, which would generally be covered by an insurer even if it excluded direct diabetes treatments, tend to be very expensive, making prevention and mitigation potentially good investments Thus, it could be the case that mandating coverage for self-management supplies and education is superfluous

4 Research Design The adoption of diabetes mandates provides us with the opportunity to examine
the

8

incentive effects of increased treatment access on the behavior of individuals Generally, isolating the causal effect of treatment availability is difficult, since improved health technology represents a shock in availability to everyone, leaving analysts without a control group against which to measure the marginal effect of improved access If one focuses not on technology but rather on price changes, as is the case in expanded insurance coverage, there is the potential that election of insurance and personal health behaviors are jointly determined With the adoption of mandates, however, the exogenous increase in access to diabetes treatments that applies to individuals in the adopting state also automatically provides us with three useful control groups Most obviously, individuals in non-adopting states do not experience this access change Secondly, restricting attention to individuals within a given adopting state, uninsured individuals receive no change to access when mandates are passed Lastly, among insured individuals in an adopting state, state mandates only apply to some individuals Specifically, the Employee Retirement Income Security Act ERISA preempts state
regulation of insurance plans offered by firms that self-insure Thus, we are left with a well-defined treatment group, individuals insured by non-ERISA plans in mandate adopting states, and control groups are provided by individuals in mandate adopting states whose insurance plans are governed by ERISA, uninsured individuals in mandate adopting states, and individuals in states that do not adopt diabetes mandates We use individual-level data from the Behavioral Risk Factor Surveillance Survey System for the years 1996-2000 to analyze the effects of diabetes mandates Our measure of health is the body mass index BMI7 The BMI is a normalized weight metric used to classify
7

Weight in pounds BMI Height in inches 2 703

9

an individuals weight status Individuals with BMIs 25 and above are considered overweight, while a BMI of 30 or greater are considered obese For our identification strategy, it is necessary to identify an individuals state of residence, his or her insurance status, and whether his or her insurance plan is governed by ERISA The BRFSS contains information on the respondents state of residence and whether or not the individual presently has some kind of health
insurance coverage is a core element of the survey To obtain a measure for the ERISA status, we use the BRFSS item typcovr1 which indicates how the individual obtains his health insurance Because a large majority of employerprovided health plans are governed by ERISA, we code individuals replying that they receive coverage through their employer or through someone elses employer as not being affected by a diabetes mandate Also, individuals replying that they receive their insurance through Medicare or Medicaid are also coded as being unaffected by mandates Those individuals replying that they or someone else buys the coverage on their own are coded as being affected by any existing diabetes mandate8 We estimate the regression:

BMI i effective mandatei ineffective mandatei i r t s i
where BMIi represents individual is BMI calculated from his survey responses regarding height and weight The effective mandate variable takes the value of one if the individuals state of residence has a mandate in effect during the survey year and if the individual not subject to ERISA preemption 0 otherwise The ineffective mandate variable takes the value of one if the

Individuals can
also respond that they receive their insurance through their spouses employer We count those individuals as falling under the ERISA pre-emption 10

8

individuals state has a mandate in effect during the survey year and if the individual either is uninsured or is subject to ERISA preemption 0 otherwise The implied control group is all individuals living in non-mandate states The vector O has individual-level covariates, D represents a time-invariant race effect corresponding to is reported race, J represents the effect of year t which is common to all individuals surveyed in the same year as i, and L represents a time-invariant state effect that is common for all individuals living in the same state as i For our covariates, we include the individuals age and age squared, recognizing that individuals tend to gain weight as they age but then reach an age where weight actually declines Following Philipson and Posner 1999, we also include income and income squared, expecting that individuals below a certain income level gain weight as their incomes rise, and then the effect of increasing income reverses after some income level is reached We include the individuals education level since
education serves as a proxy for an individuals subjective discount rate Fuchs 1982 We expect that individuals with low discount rates will invest in both education and health We also control for whether or not an individual is unemployed since unemployed individuals are likely to be less active than their employed counterparts, conditional on income levels We also control for the individuals insurance status, recognizing that the choice to buy insurance might correlate with health preferences Other measures of health preferences that we include are whether or not the individual drinks alcohol and whether the individual smokes cigarettes Lastly, we control for a number of other lifestyle attributes such as whether the individual is married, separated or divorced, the number of children the individual has, sex of the individual, and whether the individual is pregnant at the time of the survey Descriptive statistics

11

are presented in Table 2 If the moral hazard effect of the diabetes mandates dominates, we should observe a positive coefficient on the effective mandate indicator and the coefficient for the ineffective mandate should not be distinguishable from zero However, if the
insurer successfully initiates a proactive strategy to induce individuals to better their health to prevent diabetes, we should observe a zero or negative coefficient on the effective mandate variable, again observing a coefficient that is not statistically significant for the ineffective mandate variable If we restrict our attention to diabetics only, we can infer the effect of increasing access to education and self-management supplies, assuming that mandates do improve access A negative coefficient on effective mandate would imply that mandates improve access and access improves health behavior A coefficient that is not statistically different from zero would imply either that improved access does not improve behavior or that mandates do not improve access because insurers already have sufficient incentives to cover self-management supplies and education

5 Results In our first set of regressions presented in Table 3, we include all individuals in the BRFSS data We find that individuals affected by mandates exhibit BMIs that are 06 points lower than individuals in non-mandate states, and nearly 07 points lower than unaffected individuals in mandate states This result is
statistically significant at the 1 level The effect of mandates on exempted individuals is not statistically different from zero To address the problems raised by Moulton 1990, we also estimate these regressions clustering standard errors

12

by state The estimates are robust to this refinement Presumably, insurers are not able to affect their customers behavior instantly, so we develop a trend variable that allows the effect of an insurers preventive strategies to increase The effective trend variable takes the value of 1 for affected individuals during the first year of the mandate, 2 for the second year, and so on The ineffective trend variable takes the value of 1 during the first year of a mandate 2 during the second, etc for mandate state individuals without insurance or who are subject to ERISA preemption If the success of an insurers preventive strategy grows over time, we should observe a negative coefficient for the effective trend variable and no effect for the ineffective trend variable Our results suggest that mandates decrease an effected individuals BMI by 008 points per year the mandate is in effect and the coefficient is statistically significant However, mandates
appear to increase the weight of individuals in mandate states who are not even affected by the mandates relative to individuals in non-mandate states Our ineffective trend variable generates a coefficient of 001 points, and the coefficient is statistically significant To examine whether the documented relationship is causal, we developed placebo policy changes Our placebo for effective mandate variable takes the value of 1 for all insured individuals not subject to ERISA preemption for the two years before their state of residence actually adopts a mandate and 0 otherwise For example, since Arizona adopted a mandate in 1998, our placebo policy variable takes the value of 1 for all insured Arizona residents not subject to ERISA exemptions for the years 1996 and 1997 and 0 otherwise The placebo for ineffective mandate does the same thing for individuals in mandate states who are uninsured or are subject to ERISA preemption

13

If the actual policy change is driving the observed BMI decrease, we should find no systematic effect of these placebos However, as reported in Table 3, the coefficient for placebo for effective mandate is negative and statistically significant This finding
suggests that the observed negative relationship between mandate adoption and BMIs is an artifact of simultaneity Thus, the data do not support a causal inference of mandates Because the effect of mandates might differ between diabetics and non-diabetics, since diabetes prevention is not an issue with respect to diabetics and the mandate provisions regarding self-management supplies and education only apply to diabetics, pooling the two groups could be problematic However, as shown in Table 4 which presents the same regressions estimated with the non-diabetic sample, there is no change in the estimated coefficients While the results in Tables 3 and 4 provide evidence that mandates neither generate a moral hazard effect nor do they induce insurers to undertake preventive strategies with respect to the general population, it is interesting to focus on diabetics individually separately from the non-diabetic population While mandates do not appear to generate offsetting behavior in the population as a whole, they might do so for diabetics As medical treatments become effectively cheaper, diabetics might rationally substitute away from behavioral modifications in the management of their
disease However, mandates might improve a diabetics access to selfmanagement supplies and education Table 5 presents results for the regressions estimated using data only from individuals indicating that they have been diagnosed with diabetes Among diabetics, the effective mandate and effective trend variables generate negative coefficients that are statistically significant at the

14

10 percent level The ineffective mandate and ineffective trend variables generate coefficients that are not statistically different from zero Once we allow for the standard errors to be clustered by state, the effective mandate and effective trend variables generate negative coefficients that are statistically significant at the 2 percent and 01 percent levels respectively These regressions imply that mandates lower the BMI of an affected diabetic by 11 points, which represents a decrease of nearly 4 percent Our placebo tests in the diabetic population do not diminish the causal inference to be drawn from these estimates9 None of the placebo variables is statistically different from zero This implies that mandates are effective in improving the health of diabetics, presumably because they increase
access to self-management and educational resources

6 Why Do the Mandates Have Any Effect? As discussed above, if improved access to self-management supplies and education improves the behavior of diabetics, insurers seemingly have significant incentives to cover them Complications from diabetes represent large costs to insurers, making preventive care an attractive investment Why then would mandates make any difference at all? Surely insurers do not need to be required to do something that appears to be in their financial interests, but, if that is the case, we should not find any effect of mandates on the BMI of affected diabetics

We also performed an instrumental variables analysis to bolster our confidence in the causal inference Using political variables such as percent of seats in each states legislative houses controlled by Democrats and policy variables such as the adoption of other insurance mandates as instruments, even though the instruments proved to be strong, the estimates were not qualitatively different Hausman tests did not allow us to reject the hypothesis that the OLS estimates presented here are consistent 15

9

One possibility is that the existence of
mandates clarifies for doctors and patients the coverage available for these preventive treatments That is, even when an insurer already covers education and self-management supplies, the customer might be unaware of this coverage,10 or a physician might be hesitant to recommend these preventive treatments if he is unsure whether or not the cost will be covered for the patient Guglielmo 2001 presents the views of some physicians suggesting that this is the case He also presents some evidence that insurers are supportive of diabetes mandates, which would be consistent with the possibility that mandates improve awareness of what is already covered An additional possibility is that insurers do not fully internalize the benefits of preventive care since many complications of Type-II diabetes only manifest in later years when an individual is likely to be covered under Medicare as opposed to the private insurers plan If that is the case, even if preventive care is cost-justified over a patients lifetime, an insurer might rationally choose not to cover current preventive treatments since it might be unlikely to reap the future benefits ie, decreased future health expenditures
Effectively, perhaps, the insurers wager that complications will arise only after the individual is covered by Medicare and is no longer a customer of the insurer The data examined here provide no insight into which of these possibilities motivates the beneficial effect of diabetes mandates that we identify

7 Conclusion

Parente, Salkever, and DaVanzo 2003 provide convincing evidence that efforts to educate individuals about which preventive treatments are covered under their insurance make a significant difference in an individuals likelihood of electing the preventive treatment 16

10

The incidence of diabetes is on the rise The nearly 100 billion cost of diabetes and its complications represents only a fraction of the true burden of this disease that is the sixth leading cause of death in the US Believing that this burden is likely to grow, a majority of the states have passed mandates requiring insurers to cover medical treatments for the disease This increased access to treatment could induce a moral hazard problem whereby individuals rationally substitute away from preventive measures such as a healthful diet and exercise routine when the effective price of medical
treatments is lowered However, there is a countervailing force on the part of insurers Because insurers are limited in their ability to price the diabetes risk into their premiums, they have the incentive to mount proactive preventive strategies to limit the number of customers who end up developing diabetes and needing expensive medical treatments Further, among diabetics, mandates have the potential to improve access to self-management supplies and educational resources Thus, the net public health effect of mandates is ambiguous Using individual-level BMI data, we show that while individuals affected by these mandates exhibit significantly lower BMIs than individuals in non-mandate states and unaffected individuals in mandate states, we show that this result is likely an artifact of simultaneity through the inclusion of placebo laws in our regressions However, once we restrict attention to diabetics, affected individuals are shown to have significantly lower BMIs than individuals in the control groups, and analysis of placebo laws in these regressions suggests that the relationship is causal These results imply that mandates do improve access to self-management supplies and
educational resources for diabetics and these things improve the health behavior of diabetics It is unclear, however, why mandates are necessary to improve access to these resources

17

References ADA 2003 Economic Costs of Diabetes in the US in 2002 Diabetes Care, 263: 917-932 Benjamin, Stephanie, Rodolfo Valdez, Linda Geiss, Deborah Rolka, and Venkat Narayan 2003 Estimated Number of Adults with Prediabetes in the US in 2000 Diabetes Care, 263: 645649 CDC 2002 National Diabetes Fact Sheet http://wwwcdcgov/diabetes/pubs/estimateshtmprev4 CDC 2003 Diabetes: Disabling, Deadly, and on the Rise http://wwwcdcgov/nccdphp/aag/aag_ddthtm Chou, Shin-Yi, Michael Grossman, and Henry Saffer 2002 An Economic Analysis of Adult Obesity: Results from the Behavioral Risk Factor Surveillance System NBER Working Paper: 9247 Cutler, David, Edward Glaeser, and Jesse Shapiro 2003 Why Have Americans Become More Obese? NBER Working Paper: 9446 Fuchs, Victor 1982 Time Preference and Health: An Exploratory Study in Victor Fuchs, ed, Economic Aspects of Health Chicago: University of Chicago Press Guglielmo, Wayne 2001 Does Mandated Diabetes Coverage Boost Compliance? Medical Economics, 7822: 61-62, 65 Hu,
Frank, JoAnn Manson, Meir Stampfer, Graham Colditz, Simin Liu, Caren Solomon, and Walter Willett 2001 Diet, Lifestyle, and the Risk of Type 2 Diabetes Mellitus in Women New England Journal of Medicine, 34511: 790-797 Kahn, Matthew 1999 Diabetic Risk Taking: The Role of Information, Education, and Medication Journal of Risk and Uncertainty: 182: 147-164 Lakdawalla, Darius, and Tomas Philipson 2002 The Growth of Obesity and Technological Change: A Theoretical and Empirical Examination NBER Working Paper: 8946 Mokdad,Ali, Earl Ford, Barbara Bowman, William Dietz, Frank Vinicor, Virginia Bales, and James Marks 2003 Prevalence of Obesity, Diabetes, and Obesity-Related Health Risk Factors JAMA, 2891: 76-79 Moulton, Brent 1990 An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Unit Review of Economics and Statistics, 722: 334-338

18

Parente, Stephen, David Salkever, and Joan DaVanzo 2003 The Role of Consumer Knowledge of Insurance Benefits in the Demand for Preventative Health Care Among the Elderly NBER Working Paper: 9912 Peele, Pamela, Judith Lave, and Thomas Songer 2002 Diabetes in Employer-Sponsored Health Insurance Diabetes Care, 2511: 1964-1968
Peltzman, Sam 1975 The Effects of Automobile Safety Regulation Journal of Political Economy, 834: 677-726 Philipson, Tomas, and Richard Posner 1999 The Long-Run Growth in Obesity as a Function of Technological Change NBER Working Paper: 7423 Viscusi, Kip 1984 The Lulling Effect: The Impact of Child-Resistant Packaging on Aspirin and Analgesic Ingestion American Economic Review, 742: 324-327

19

Table 1 Mandate Adoption State Alaska Arizona California Colorado Connecticut Delaware Florida Georgia Hawaii Illinois Indiana Iowa Kentucky Louisiana Maine Massachusetts Michigan Minnesota Mississippi Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York Year 2000 1998 1981 1998 1997 2000 1995 1998 2000 1998 1997 1984 1998 1997 1996 2000 2000 1994 1998 2001 1999 1997 1997 1996 1997 1993

North Carolina Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

1997 1996 2001 1998 1996 1999 1999 1997 1997 2000 1997 1998 1997 1996 1987 2001

Table 2 Descriptive Statistics Variable BMI Total Sample BMI Diabetics Excluded BMI Diabetics Only Effective Mandate Ineffective Mandate Effective
Trend Ineffective Trend Placebo for Effective Mandate Placebo for Ineffective Mandate Description Body Mass Index Body Mass Index Body Mass Index Indicator 1 if individual lives in a mandate state and mandate applies Indicator 1 if individual lives in a mandate state and mandate does not apply 1 in 1st year of mandate, 2 in 2nd, etc if mandate applies to individual 1 in 1st year of mandate, 2 in 2nd, etc if mandate does not apply to individual Indicator 1 for 2 years prior to mandate adoption if mandate applies to individual Indicator 1 for 2 years prior to mandate adoption if mandate does not apply to individual Age in years Income in 1,000s Education level reported on scale of 1-6 Indicator 1 if individual indicated he is currently unemployed Indicator 1 if individual indicated he is insured Number of alcoholic beverages the individual consumed during last month Indicator 1 if individual currently smokes Indicator 1 if individual is currently married Indicator 1 if individual is divorced or separated Number of children ages 18 and under individual has Indicator 1 if individual is female Indicator 1 if individual is currently pregnant Mean 26052 25850 29611 0031 0532
0116 1801 0019 0244 Std Dev 5117 4957 6415 0174 0499 0977 3211 0136 0429

Age Income Education Unemployed Insured Drinks Smoker Married Separated or Divorced Children Female Pregnant

46669 38649 4665 0034 0877 9080 0237 0542 0157 0734 0590 0014

17378 21643 1097 0181 0329 23987 0447 0498 0364 1138 0492 0119

Table 3 Effect of Diabetes Mandates on BMI for Full Sample standard errors in parentheses Variable Effective Mandate Ineffective Mandate Effective Trend Ineffective Trend Placebo for Effective Mandate Placebo for Ineffective Mandate Age Age2 Income Income2 Education Unemployed Insured Drinks -0611 0070 0071 0043 0330 0004 -0003 0000 -0025 0002 0009 0003 -0327 0011 0351 0058 0132 0032 -0005 0000 -0078 0013 0014 0007 0329 0004 -0003 0000 -0024 0002 0008 0003 -0328 0011 0349 0058 0116 0032 -0005 0000 No Clustering -0533 0100 0014 0035 0328 0004 -0003 0000 -0024 0002 0008 0003 -0328 0011 0348 0058 0110 0032 -0005 0000 -0610 0087 0067 0068 -0501 0108 0028 0054 0330 0004 -0003 0000 -0025 0002 0009 0003 -0327 0011 0352 0058 0139 0032 -0005 0000 -0611 0077 0071 0045 0330 0008 -0003 0000 -0025 0003 0009 0003 -0327 0022 0351 0100 0132 0045 -0005 0001 Standard Errors
Clustered by State -0078 0028 0014 0003 0329 0008 -0003 0000 -0024 0003 0008 0003 -0328 0022 0349 0100 0116 0045 -0005 0001 -0533 0107 0014 0035 0328 0008 -0003 0000 -0024 0003 0008 0003 -0328 0022 0348 0100 0110 0044 -0005 0001 -0610 0084 0067 0056 -0501 0113 0028 0040 0330 0008 -0003 0000 -0025 0003 0009 0003 -0327 0022 0352 0100 0139 0045 -0005 0001

Smoker Married Separated or Divorced Children Female Pregnant Race Effects State Effects Year Effects Observations Adjusted R2

-0864 0024 0102 0029 -0401 0036 0048 0010 -1615 0024 0969 0083 Yes Yes Yes 212,082 0091

-0863 0024 0107 0029 -0397 0036 0049 0010 -1612 0024 0970 0083 Yes Yes Yes 212,082 0091

-0863 0024 0108 0029 -0396 0036 0049 0010 -1611 0024 0972 0083 Yes Yes Yes 212,082 0091

-0866 0024 0100 0029 -0404 0036 0048 0010 -1616 0024 0969 0083 Yes Yes Yes 212,082 0091

-0864 0050 0102 0038 -0401 0043 0048 0015 -1615 0046 0969 0149 Yes Yes Yes 212,082 0091

-0863 0050 0107 0038 -0397 0043 0049 0015 -1612 0046 0970 0148 Yes Yes Yes 212,082 0091

-0863 0050 0108 0038 -0396 0043 0049 0015 -1611 0046 0972 0148 Yes Yes Yes 212,082 0091

-0866 0050 0100 0038 -0404 0043 0048 0015 -1616 0046 0969 0149 Yes Yes Yes 212,082
0091

Note: The dependent variable is Body Mass Index BMI as reported in the Behavioral Risk Factor Surveillance Survey System for the years 1996-2000 Effective Mandate represents an indicator which takes the value of 1 if the insured individual resides in a state with a mandate and is not subject to ERISA preemption of state mandates Ineffective Mandate represents an indicator taking the value of 1 if the individual resides in a state with a mandate but the mandate does not apply either because of ERISA preemption or because the individual has no insurance The coefficient for Income2

Table 4 Effect of Diabetes Mandates on BMI for Sample with Diabetics Excluded standard errors in parentheses Variable Effective Mandate Ineffective Mandate Effective Trend Ineffective Trend Placebo for Effective Mandate Placebo for Ineffective Mandate Race Effects State Effects Year Effects Observations Adjusted R2 -0560 0069 0074 0043 Yes Yes Yes 203,749 0087 -0069 0012 0014 0007 Yes Yes Yes 203,749 0087 No Clustering -0510 0099 0010 0035 Yes Yes Yes 203,749 0087 -0610 0087 0067 0068 -0501 0108 0028 0054 Yes Yes Yes 212,082 0091 -0560 0074 0074 0042 Yes Yes Yes 203,749 0087
Standard Errors Clustered by State -0069 0027 0014 0003 Yes Yes Yes 203,749 0087 -0510 0109 0010 0033 Yes Yes Yes 203,749 0087 -0558 0087 0072 0068 -0475 0107 0028 0054 Yes Yes Yes 203,749 0088

Note: The dependent variable is Body Mass Index BMI as reported in the Behavioral Risk Factor Surveillance Survey System for the years 1996-2000 Effective Mandate represents an indicator which takes the value of 1 if the insured individual resides in a state with a mandate and is not subject to ERISA preemption of state mandates Ineffective Mandate represents an indicator taking the value of 1 if the individual resides in a state with a mandate but the mandate does not apply either because of ERISA preemption or because the individual has no insurance Although the coefficients for the non-policy covariates are not reported, each regression includes: Age, Age2, Income, Income2, Education, Unemployed, Insured, Drinks, Smoker, Married, Separated or Divorced, Children, Female, and Pregnant The coefficients for these variables are similar in sign, magnitude, and statistical significance to those presented in Table 3

Table 5 Effect of Diabetes Mandates on BMI for Diabetics standard
errors in parentheses Variable Effective Mandate Ineffective Mandate Effective Trend Ineffective Trend Placebo for Effective Mandate Placebo for Ineffective Mandate Age Age2 Income Income2 Education Unemployed Insured Drinks -1053 0543 -0154 0278 0558 0031 -0006 0000 -0064 0014 0054 0016 -0238 0063 0543 0381 0034 0237 -0000 0003 -0176 0099 -0015 0042 0558 0031 -0006 0000 -0063 0013 0054 0016 -0236 0063 0550 0381 0023 0237 -0000 0003 No Clustering -0306 0836 0133 0222 0557 0306 -0006 0000 -0063 0014 0053 0016 -0236 0063 0539 0381 0015 0237 -0000 0003 -1048 0647 -0148 0449 -0432 0887 0022 0358 0559 0031 -0006 0000 -0064 0014 0054 0016 -0237 0063 0550 0381 0037 0237 -0000 0003 -1053 0428 -0154 0283 0558 0029 -0006 0000 -0064 0013 0054 0015 -0238 0057 0543 0426 0034 0247 -0000 0003 Standard Errors Clustered by State -0176 0036 -0015 0017 0558 0029 -0006 0000 -0063 0013 0054 0014 -0236 0057 0550 0425 0023 0246 -0000 0003 -0306 0736 0133 0225 0557 0029 -0006 0000 -0063 0013 0053 0014 -0236 0057 0539 0423 0015 0246 -0000 0003 -1048 0491 -0148 0335 -0432 0728 0022 0270 0559 0029 -0006 0000 -0064 0013 0054 0015 -0237 0057 0550 0425 0037 0247 -0000 0003

Smoker
Married Separated or Divorced Children Female Pregnant Race Effects State Effects Year Effects Observations Adjusted R2

-1370 0166 -0145 0178 -0678 0214 0115 0077 0921 0200 0339 1054 Yes Yes Yes 8,333 0084

-1372 0166 -0147 0178 -0677 0214 0119 0077 0922 0200 0326 1054 Yes Yes Yes 8,333 0084

-1374 0166 -0142 0178 -0675 0215 0119 0077 0929 0200 0314 1055 Yes Yes Yes 8,333 0084

-1370 0166 -0147 0178 -0678 0215 0115 0077 0922 0200 0352 1055 Yes Yes Yes 8,333 0084

-1370 0240 -0145 0205 -0678 0291 0115 0082 0921 0268 0339 1566 Yes Yes Yes 8,333 0084

-1372 0241 -0147 0206 -0677 0291 0119 0083 0922 0268 0326 1574 Yes Yes Yes 8,333 0084

-1374 0240 -0142 0206 -0675 0292 0119 0082 0929 0267 0314 1580 Yes Yes Yes 8,333 0084

-1370 0239 -0147 0206 -0678 0292 0115 0082 0922 0268 0352 1568 Yes Yes Yes 8,333 0084

Note: The dependent variable is Body Mass Index BMI as reported in the Behavioral Risk Factor Surveillance Survey System for the years 1996-2000 Effective Mandate represents an indicator which takes the value of 1 if the insured individual resides in a state with a mandate and is not subject to ERISA preemption of state mandates Ineffective Mandate represents an indicator taking
the value of 1 if the individual resides in a state with a mandate but the mandate does not apply either because of ERISA preemption or because the individual has no insurance The coefficient for Income2 has been multiplied by 100 for presentation