State health insurance market reforms and access to insurance for high-risk employees

Journal of Health Economics 24 (2005) 725–750

State health insurance market reforms and access to insurance for high-risk employees Amy Davidoff ∗ , Linda Blumberg 1 , Len Nichols 2 Department of Public Policy, University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, MD 21250, USA Received 2 February 2002; received in revised form 18 December 2003; accepted 10 November 2004 Available online 29 April 2005

Abstract A specific focus for state regulations of the small group insurance market was to increase offers and stabilize premiums for firms with high-risk workers. We examine the effect of reforms implemented from 1993 through 1996 on the likelihood of employer sponsored insurance coverage. We find that packages of reforms that included both guaranteed issue of some products and some form of rate variance restriction had significant positive effects (4.5 percentage points) on ESI coverage for highrisk compared with low-risk workers within small firms and a small negative effect (−1.7 percentage points) on low-risk workers in small compared with large firms. The mechanism for these effects was an increase in take-up, rather than offer. Reform packages that included both guaranteed issue of all products and rate variance restrictions had similar effects overall, although they did not meet criteria for significance. These effects seemed to act through increased offer rather than take-up. © 2005 Elsevier B.V. All rights reserved. JEL classification: I 18 Keywords: Insurance; Regulation; Health status

∗

Corresponding author. Tel.: +1 410 455 6561; fax: +1 410 455 1172. E-mail addresses: [email protected] (A. Davidoff), [email protected] (L. Blumberg), [email protected] (L. Nichols). 1 Principal Research Associate, Health Policy Center, The Urban Institute, 2100 M Street N.W., Washington, D.C. 20037, USA. 2 Director, Health Policy Program at The New American Foundation, 1630 Connecticut Avenue, N.W. 7th Floor, Washington, D.C. 20009, USA. 0167-6296/$ – see front matter © 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.jhealeco.2004.11.010

726

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

1. Introduction Employment based health insurance is the largest source of insurance coverage for Americans. In 2002, 64.2% of non-elderly Americans had insurance through their own employer, or received coverage as a dependent through the employer of their spouse or parent (Fronstin, 2003). However, many Americans who are workers or linked to workers lack insurance coverage, suggesting that the system of employer sponsored insurance (ESI) has important gaps. Overall, 43.3 million Americans lacked health insurance coverage in 2002, and 83% of the uninsured were in a family with a worker (Fronstin, 2003). Workers in small firms are particularly disadvantaged in their ability to obtain ESI. For example, only 54% of workers in the smallest firms (fewer than 10 employees) work for employers that sponsor ESI coverage, compared with 95% for workers in firms with 100 or more employees. Although only 34% of workers are employed at firms with fewer than 100 workers, 62% of uninsured workers are at these smaller firms (Garrett, 2004). In the late 1980s and early 1990s, concerns were raised about insurer practices with respect to small firms that were thought to affect access to coverage. Small firms faced higher premium rates because fixed administrative costs were spread over a small number of persons. However, in the absence of regulation, small firms were also subject to other differences. Many insurers refused to offer insurance to small firms due to the size of the potential risk group and the inherent instability within the risk group over time. When insurance was offered, detailed medical underwriting for small firms was often used to identify workers and dependents with high expected costs. The presence of such workers increased premiums for the firm relative to the average. These elevated premiums discouraged employers from offering coverage and could reduce take-up by employees in small firms that did offer. In addition to relatively high premium rates, small firms were subject to very steep increases in premiums after one or two years, once pre-existing condition restrictions expired. Thus, even when small firms offered coverage, it was common for them to change insurers frequently, and employees may have dropped coverage as premiums increased. In an effort to improve the climate of access to insurance for small firms and their workers, states implemented a variety of regulatory reforms starting in the early 1990s. In most states, firms with fewer than 50 employees were the target for these reforms. A specific focus of the reforms was to improve the ability of firms with high-risk workers to get offers of coverage with relatively stable premiums by broadening the risk pool. By high-risk workers, we mean workers in a family where some member has a chronic health condition likely to result in high health care expenditures. These reforms preceded the small group market reforms that were mandated by the federal Health Insurance Portability and Accountability Act (HIPAA) in 1996, and in many states they continue to be more restrictive than the reforms mandated by HIPAA, and hence are still relevant to current policy. The reforms generally fall into one of two main categories: issue reforms, which affect which firms must be offered insurance and what packages must be offered by insurers, and premium variance restrictions, which affect the prices insurers may charge. Issue reforms are designed to make insurance coverage easier to obtain for small groups. Four major types of issue reforms were implemented during the early 1990s:

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

727

• Guaranteed issue requires insurers to issue some or all health insurance plans to any small employer group willing to purchase, regardless of the medical history or claims experience of the group. During the study period, most states required plans to offer only two health plans, a basic plan (with high deductibles and/or stringent coverage limits) and a standard plan, whereas other states required insurers operating in the small group market to offer all products. The HIPAA legislation passed in 1996 mandated the latter policy. • Guaranteed renewal requires insurers to renew all insurance policies, regardless of the claims experience, as long as premiums are paid in a timely manner. • Limits on pre-existing condition restrictions. Insurers prefer to limit coverage of conditions that were diagnosed prior to purchase of the policy. Policies generally specify a window of time prior to purchase (a “look back” period), and also specify a period of time during which treatment of the identified condition is excluded (a waiting period). Reforms generally limit the length of both the look back and the waiting periods. • Portability reforms allow continuously insured individuals to change group insurance plans without incurring a new period of exclusion for pre-existing conditions. The goal of premium variance restrictions is to achieve more affordable and predictable premiums, particularly for higher risk groups, by limiting the degree to which insurers can vary premiums across groups. Within the industry, the process of setting a premium is called “rating,” and so the terms for specific premium reforms include that term. There are two major types of rating restrictions: • Community rating is the most stringent and requires insurers to offer the same rate to all groups, regardless of demographic composition or medical history. Modified community rating, much more common, permits variation in rates using some demographic characteristics (e.g. age and gender), but not health status or claims experience. • Rate bands allow insurers to set rates on the basis of demographics, health status, or claims history, but limit the ex post amount of variance among premiums within each allowed rating category or overall, for the same product. For example, states might require that policies for a particular product vary by no more than 25% above or below an index rate. States differ in the amount of variation that is permitted, but also in how the index rate is defined, and whether rating is done for the entire insured group or for a subset “book of business” (Hall, 2000). For this reason, it is difficult to compare how binding these rating bands are across states. All of these insurance market reforms have the effect of forcing more risk pooling than the unregulated market would accomplish. Furthermore, reforms, especially premium variance restrictions, are basically symmetric, meaning that they limit equally the degree to which premiums can vary above or below a designated standard, yet actual and expected expenditures are highly skewed, with a small number of high cost persons accounting for a disproportionate share of spending. This has the practical effect of increasing the average premium for plans offered, as it increases access. Premiums are lowered for the relatively few firms with many high-risk workers while premiums are raised for the many low-risk groups and individuals (Nichols, 2000). Thus, reforms naturally produce a tradeoff, about which policy value judgments differ, and the net coverage effects are unpredictable a priori.

728

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Relatively little is known about the impact on insurance coverage of the reforms passed during the early 1990s. Understanding their impact remains highly relevant, particularly for high-risk workers in small firms, as states still vary with respect to the total packages of reforms despite the HIPAA mandates. Early studies, such as the study by Marstellar et al. (1998), Sloan and Conover (1998), and Zuckerman and Rajan (1999) examined effects on insurance coverage specifically for workers or uninsurance rates overall. These studies found little to no effect of reforms overall. However, more recent studies have begun to examine whether there are differential effects on groups with different risk status. We review them in detail in Section 2.3 below. In this study we examine the effect of insurance market reforms on the likelihood that high-risk workers in small firms are offered and are covered by ESI. We pool data from the National Health Interview Survey (NHIS) from 1993 through 1996 to obtain data on health status, insurance coverage, demographic and employment related information for over 100,000 workers. We designate high-risk workers using unique information on the presence of chronic health conditions for the worker and family members. We isolate the effects of market reforms on insurance coverage using a difference-in-differences-in-differences approach, based on the technique used by Gruber (1994). This allows us to estimate the effects of reform on high-risk workers in small firms, using low-risk workers in small firms and workers in large firms as comparison groups. We find that packages of reforms that included both guaranteed issue of some products and some form of rate variance restriction had significant positive effects (4.5 percentage points) on ESI coverage for high-risk compared with low-risk workers within small firms and a small negative effect (−1.7 percentage points) on low-risk workers in small compared with large firms. The mechanism for these effects was an increase in take-up, rather than offer. Reform packages that included both guaranteed issue of all products and rate variance restrictions had similar effects overall, but seemed to act through effects on offer rather than take-up. The remainder of the paper is organized as follows: in Section 2 we provide background, discuss the hypothesized effects of insurance market reforms and summarize findings from the most recent literature; in Section 3 we discuss the estimation approach; Section 4 describes data sources and measurement; Section 5 presents results; Section 6 includes discussion and conclusions.

2. Background and theory of insurance market reforms 2.1. Small group insurance markets In the absence of perfect and symmetric information, adverse selection is a feature of every voluntary insurance market. Insurers must assume that individuals willing to pay average prices for insurance are likely to have higher than average health risks. Groups with different health risks will prefer different degrees of coverage. In theory, this can lead to extreme market segmentation and even collapse (Rothschild and Stiglitz, 1976). The assumption that insurers are risk neutral is essential to the standard theory of insurance markets. However, observed behavior in the small group market does not always support that assumption. Information is costly and the exact probabilities of bad events

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

729

cannot be determined a priori. In response, insurers take explicit and sometimes costly action to protect themselves from risk. In the absence of regulation, insurers may use detailed applications and claims histories to identify health risks among the employees and dependents in a small firm. This helps them underwrite, or tailor premiums to specific groups’ relative health risk. Insurers typically set small group premiums based on weighted averages of similar groups’ claims experience with the insurer and the recent experience of the particular small group. Insurers also adopt various techniques to protect themselves from groups with disproportionately high-risk, such as pre-existing condition restrictions and refusal to renew coverage. The insurer may offer different types of insurance plans at different costs for each group, or refuse to sell a product to some groups at any price, presumably because uncertainty is so great that an actuarially fair price is difficult to estimate. 2.2. Expected effects of small group insurance market reforms The two major categories of insurance market regulation may each independently affect the insurance products offered and premium set. Issue reforms such as limits on pre-existing condition exclusions and portability are likely to increase premiums to a small extent for insurance products already offered to firms, because the reforms force insurers to cover costs associated with more of the health conditions experienced by the worker risk pool. Guaranteed issue, and to a lesser extent guaranteed renewal, are likely to have larger effects on premiums, since they force insurers to provide coverage to groups that would be deemed too risky in the absence of the regulation. The groups newly offered insurance as a result of this regulation will likely face higher premiums than those offered in the absence of reforms, and average premiums for the product will increase. Even low-risk firms previously offered insurance may face increased premiums as a result of issue reforms. The transaction costs associated with determining expected costs for the new high-risk groups are non-trivial, and may lead to more pooling of risks (Newhouse, 1996) than predicted by the stylized Rothschild–Stiglitz model (1976). Premium variance or rating restrictions increase the degree of risk pooling by constraining the extent to which insurers may use health status and demographic information to set premiums. This process essentially forces low-risk firms and individuals to cross-subsidize premiums for high-risk firms and individuals. Thus, low-risk firms will pay higher premiums whereas high-risk firms pay lower premiums than in the absence of the regulation. The size of the increase in premiums for low-risk firms under the scenario of increased risk pooling should be much smaller than the decrease in premiums faced by the high-risk firms because there are many more low-risk than high-risk firms (Nichols, 2000). The expected effect of these changes in premiums, at least among supporters of rating restrictions, was that the increases in premiums among the low-risk groups would not be large enough to discourage employers from offering or employees from taking up coverage, but that the decrease in premiums for the high-risk would be sufficiently large so as to encourage employer sponsorship and employee take-up of health insurance. The effect of variance or premium rating restrictions is blunted to some extent when these reforms are implemented without issue reforms. Insurers may still refuse to offer coverage to high-risk groups, which limits risks to insurers. However, when issue reforms

730

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

are combined with premium variance or rating restrictions, the effect on premiums is likely to be much greater because much greater pooling of risks is required. The net effect of small group reforms on the average health risk of insured small group pools is ambiguous though, as several mechanisms are possible. High-risk workers in small firms may experience improved access to health insurance, either through increased offers, lower premiums, or both. High-risk workers may perceive that they have greater access to insurance in a small group, and so may be less inclined to seek out jobs with large firms—which already have de facto guaranteed issue and considerable intra-firm risk pooling. In this way, small group health insurance market reforms may actually increase cross-firm size mobility among high-risk workers. However, reforms may also reduce the likelihood that low-risk workers take up coverage. Low-risk workers may balk at paying higher premiums out of pocket. If their wages suffer as employer health insurance costs go up, low-risk workers may persuade employers to drop coverage or to work for firms that do not offer insurance. A firmor industry-wide decline in offered health insurance could reduce coverage among some currently insured high-risk workers as well. This becomes more likely if there are limits on reducing the generosity of the package, either from state mandates or market norms among large firms who compete for labor. So, although advocates expected that more high-risk workers would gain access than would lose it, the net coverage impact of reforms, even for high-risk workers, is ambiguous. 2.3. Patterns of implementation Between the years 1990 and 1996, 46 states implemented some form of regulation in the small group insurance market. Implementation of the four types of issue reforms and two types of rating restrictions was not random, rather states tended to implement several reforms jointly. For example, Zuckerman and Rajan (1999) report that by 1995, 36 states had implemented all four of the issue reforms and some form of rating variance restriction, six states had implemented all of those except guaranteed issue, and three states had guaranteed renewal and some form of rating restriction. 2.4. Previous studies The existing literature on the effects of small group insurance market reforms on insurance coverage have examined whether there is an overall effect of reforms. Most of these studies found no discernable effect of reforms on coverage rates. For example, Sloan and Conover (1998) used the Current Population Survey (CPS) to examine the effect of reforms on individual insurance coverage from 1989 to 1994—whether the person had any, the type and source. They tested the effects of each reform separately, and failed to find evidence of any effect, with the exception of a positive effect of community rating on the likelihood of having group coverage among the near elderly. Marstellar et al. (1998) used the CPS from 1989 to 1995, to construct state level measures of the percent uninsured, privately insured and Medicaid enrolled. They tested the effects of packages of issue reforms and rating reforms in the small group market. They found a strong negative effect on the proportion uninsured that was associated with implementation of the four issue reforms, but it was counterbalanced by a strong positive effect on the proportion uninsured asso-

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

731

ciated with rating reforms. Zuckerman and Rajan (1999) also used the CPS to construct state level measures of the proportion uninsured. They tested the effects of packages of reforms that combined issue and rating restrictions, and failed to find any effects on coverage rates. More recent studies have begun to examine the effects of reforms on high-risk workers in small firms, the specific target groups for the reform efforts. Both Percy (2001) and Simon (2002) used CPS data from early to mid-1990s to examine the effects of packages of insurance reforms. Simon defined “full reform” as a state having both guaranteed issue and rating restrictions. The dependent measure was whether a worker had coverage through their employer. Simon identified high cost demographic groups using the Medical Expenditure Panel Survey (MEPS), and then used those demographic characteristics to identify “highrisk” workers on the CPS. The effects of reforms were tested for low versus high-risk workers, and in small versus large firms. Simon found that full reforms had a negative effect on coverage for low-risk workers in small firms, but no effect on coverage for highrisk workers. Simon also analyzed the 1993 National Employer Health Insurance Survey and the 1996 Medical Expenditure Panel Survey to estimate the effects of reforms on offers of ESI and premiums. Full reform packages resulted in premium increase of $7.80 per month per person, relative to an overall small firm average premium of $195. A major limitation of studies using the CPS is that they lack specific information on health status of workers and dependents, relying instead on broad demographic characteristics to identify high and low-risk workers. This approach is not consistent with the underwriting practices used by insurers in the small group market, absent reform, that may involve detailed questions on the health history and current health status. Monehit and Schone (2003) examined the effects of small group market reforms using data from the 1987 National Medical Expenditure Survey (NMES) and the 1996 and 1997 Medical Expenditure Panel Survey (MEPS). To identify high-risk workers, they estimated an expenditure model that incorporated information on chronic medical conditions, and then predicted expenditures to worker families, assuming private insurance coverage, selecting those in the top 25th percentile as high-risk workers. They find that reforms had little effect on offer rates, but that in states with the most stringent reforms, employment based coverage and policyholder rates increased for high-risk relative to low-risk workers. Monheit and Schone’s approach to identifying high-risk workers represents a major contribution, and demonstrates the importance of health status in evaluating the effects of reforms. In this study we take a somewhat different approach, using detailed self-reported information on chronic medical conditions available on the National Health Interview Survey to identify high-risk workers, and estimate the effects of insurance reforms on their insurance coverage.

3. Methods We assess whether and to what extent state regulation of the small group insurance market improved access to ESI coverage for high-risk workers. We use repeated cross sectional measures from a household survey of insurance coverage and health status, during a period when many states were implementing or strengthening insurance market reforms, and estimate the effects of reforms on coverage. Because the reforms are targeted at small

732

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

employers, we expect a greater effect on workers who work for small firms, thus our estimation strategy specifically tests for such an effect. Workers in large firms provide an important comparison group, because they are subject to, and therefore control for state specific trends that may affect small firms. However, we recognize that the reforms may have affected job mobility between large and small firms, with a secondary effect on coverage for high-risk individuals, thus the differences between the firm size groups may be biased downward to some degree. Because insurance coverage is also affected by individual and family characteristics, characteristics of employment, the availability of publicly subsidized insurance, characteristics of the local labor market and economy, and temporal trends, we use multivariate techniques to control for these in our analysis. 3.1. Estimation strategy To estimate the effects of state insurance market reforms on insurance coverage of workers, and to test for differential effects on high-risk workers employed in small firms, we employ a difference-in-difference-in-differences (DDD) technique, based on the technique described by Gruber (1994). The primary model that we estimate is: ESI = ΦESI (β0 + β1 TIME + β2 STATE + β3 X + β4 REFORMS + β5 HIGHRISK + β6 SMALLFIRM + β7 HIGHRISK × REFORMS + β8 SMALLFIRM × REFORMS + β9 HIGHRISK × SMALLFIRM + β10 HIGHRISK × REFORMS × SMALLFIRM + εESI )

(1)

The dependent variable, ESI, is one of three measures that capture access to ESI and ESI coverage. The key explanatory variables of interest include REFORMS, a vector with two binary indicators for whether a state had implemented either a highly restrictive or somewhat less restrictive combination of insurance market reforms in a given year. HIGHRISK measures whether any family member is predicted to have fair or poor health, based on demographic characteristics and reported chronic conditions.1 SMALLFIRM is an indicator for a family in which none of the workers are employed in large firms. The model includes a full set of two and three-way interaction terms between REFORMS, HIGHRISK, and SMALLFIRM. Measurement of these and other variables in the model is described in detail in Section 4. It is important to note that the measures of ESI coverage, firm size, and health problems used in the study explicitly account for family level decision making by workers and spouses. The model also includes a series of control variables (X) for the characteristics of the worker, spouse and/or child. A vector with county level measures that capture the average price of insurance, and factors likely to affect offers of insurance, is included, as are state level measures of generosity of Medicaid coverage relevant to family members in the state and year. The model also includes controls for temporal trends and fixed state effects. We do 1

We defined families as married couples and their biologic, adoptive or step children. This definition is intended to group workers and dependents who would be eligible for insurance coverage through the worker’s ESI.

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

733

not include state-year interactions to capture the unique economic circumstances that a state may face in a particular year. Instead we allow the county level measures of unemployment to capture economic shocks at the local level. We use various combinations of the coefficients estimated from the model to arrive at estimates of the effects of reforms on workers in small firms, for high and low-risk workers. Table 1 illustrates the various comparisons that we use to generate these results. We assume for purposes of exposition that all variables are simple binary indicators, so that when a condition is not met (not a small firm, no reform, not high-risk), the indicator value equals zero, and the coefficient does not contribute to a predicted probability of ESI coverage. We also assume that the estimated model is linear, so that the coefficient equals the marginal effect. Each cell in the figure shows the coefficients from the key variables in Eq. (1) that would contribute to the predicted probability of ESI under various scenarios. Column 1 indicates those coefficients that would contribute to the predicted probability under a scenario of insurance reforms. Column 2 indicates the coefficients that would contribute to the predicted probability under a scenario without reforms. Column 3 is the difference between column 1 and column 2, and captures the effect of reforms on the predicted probability of ESI. We are interested in both the DD and DDD estimates. The DD effects essentially capture the effects of reform for the target group of interest (small firm, high-risk workers) compared with either but not both of the comparison groups (large firm, high-risk workers or small firm, low-risk workers). The DDD estimate captures the effect of reform on small firm, high-risk workers, relative to effects on large firm workers and lowrisk workers. We clarify further by describing a specific example from Table 1. Cell D2 indicates that the probability of ESI, pre-reform (REFORM = 0), for workers in small firms (SMALLFIRM = 1) who have a family health problem (HIGHRISK = 1), is equal to β0 + β1 + β2 + β3 + β5 + β6 + β9. The probability of insurance, post-reform, for this same group of workers equals β0 + β1 + β2 + β3 + β4 + β5 + β6 + β7 + β8 + β9 + β10 (Cell D1). The effect of reforms for high-risk workers in small firms equals the latter minus the former, as indicated in Cell D3, or β4 + β7 + β8 + β10. Similarly, the probability of insurance coverage for low-risk workers (HIGHRISK = 0) in small firms (SMALLFIRM = 1) pre-reform equals β0 + β1 + β2 + β3 + β6, as indicated in Cell E2. The post-reform probability of coverage for these workers is measured by β0 + β1 + β2 + β3 + β4 + β6 + β8 (Cell E1). Thus, the effect of reforms for workers in small firms without family health problems is β4 + β8 (Cell E3). Finally, the DD estimator, the difference between the effect of reforms on probability of coverage for workers in small firms with family health problems versus no health problems is Cell D3 − Cell E3, or β7 + β10, as shown in Cell F3. The DDD estimator is the difference between the effects of reform on high-risk compared with low-risk workers in small firms, and relative to the effects of reform on high-risk compared with low-risk workers in large firms. This DDD estimator is computed in row M of Table 1. In the empirical analysis, we estimated equations as probits. We determine the magnitude of the DD results that involve multiple coefficients by calculating the difference in predicted probabilities of the dependent measure when the variable of interest is alternately set to a value of one or zero, and the interaction terms are recomputed. We calculate a Wald test on the linear sum of the relevant coefficients to determine the significance of the DD effects.

734

1 Reform (β0 + β1 + β2 + β3 + . . . A B C D E F G H I J K L M

2 No Reform (β0 + β1 + β2 + β3 + . . .

High-risk, small firm β4 + β5 + β6 + β7 + β8 + β9 + β10 β5 + β6 + β9 High-risk, not small firm β4 + β5 + β7 β5 DD effect of reform on high-risk workers, small vs. not small firm (Cell A3 − Cell B3) High-risk, small firm β4 + β5 + β6 + β7 + β8 + β9 + β10 β5 + β6 + β9 Not high-risk, small firm β4 + β6 + β8 β6 DD effect of reform on high-risk workers vs. low-risk workers, in small firms (Cell D3 − Cell E3) Not high-risk, small firms β4 + β6 + β8 β6 Not high-risk, not small firms β4 Effect of reform on low-risk workers in small vs. large firms (Cell G3 − Cell H3) β5 High-risk, not small firm β4 + β5 + β7 Not high-risk, not small firms β4 DD effect of reform on high-risk vs. low-risk workers in large firms (Cell J3 − Cell K3) DDD effect of reforms on high-risk workers in small firms (F3 − L3) = (C3 − I3)

3 Effect of Reform (col 1 − col 2) β4 + β7 + β8 + β10 β4 + β7 β8 + β10 β4 + β7 + β8 + β10 β4 + β8 β7 + β10 β4 + β8 β4 β8 β4 + β7 β4 β7 β10

ESI = ΦESI (β0 + β1 TIME + β2 STATE + β3 X + β4 REFORMS + β5 HIGHRISK + β6 SMALLFIRM + β7 HIGHRISK × REFORMS + β8 SMALLFIRM × REFORMS + β9 HIGHRISK × SMALLFIRM + β10 HIGHRISK × REFORMS × SMALLFIRM + εESI ).

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Table 1 Difference-in-differences-in-differences model: computation of selected DD and DDD results

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

735

3.2. Estimating insurance coverage versus offer and take-up We expect that reforms would have different effects on the probability that a worker is offered ESI and that the worker takes up ESI, because reforms affect both issue of insurance to small firms as well as the premiums that are charged. Thus, we decompose the dependent variable ANYESI into offer of ESI and take-up of ESI to determine whether there are different effects of reforms on each of these stages. The take-up model is conditioned on the worker having an offer of ESI in the family, thus it is estimated on a sub-sample of workers. To the extent that workers sort into firms that do and do not offer ESI based on their expected participation, then the sub-sample of workers with ESI offers will be selected endogenously, and the estimates of take-up may be biased and inefficient. To address this issue, we estimate models of offer and take-up as a simultaneous bivariate probit system, using the HECKPROB procedure in Stata. We identify the take-up equation by excluding three county level measures of industry structure (percent of firms in the service industry, percent of firms with less than 50 employees, unemployment rate) that are included in the offer equation. An individual’s decision to take-up an employer offer should not be a function of the composition of the labor market—it is a function of the individuals’ preferences and circumstances. However, the decision to offer will be related to the probability that other employers competing for similar workers offer insurance coverage. The measure of correlation between the equations (rho) tests for sorting of workers into firms based on whether they offer ESI. 4. Data and measurement issues 4.1. Sources of data The principal source of data for this study was the NHIS, a continuous national household survey sponsored by the National Center for Health Statistics (Adams and Marano, 1995; Benson and Marano, 1998). The NHIS collects information on demographics, labor force participation, income, health status, health insurance and use of health care services through a core survey instrument and a series of annual supplements. We used NHIS data from 1993 through 1996. The 1993 NHIS was the first year that workers were asked routinely about employer offers of insurance. We discontinued the series after 1996 since the survey design changed substantially.2 The period from 1993 through 1996 was also the time when many states were implementing or altering their small group reforms. The annual sample size for the NHIS is approximately 45,000 households and 110,000 persons.3 Data on state level regulations or health insurance market reforms were assembled from three existing 2 The survey instrument and design for collecting data on health conditions changed dramatically mid-year 1996, affecting the format of publicly available data starting in 1997. In the newly designed survey, the medical conditions check list included a subset of the most common conditions, and the full checklist was asked of a sample adult per family. In the survey instrument relevant to this study, the condition checklist was extensive, but persons in each household were asked about only one sixth of the conditions. 3 However, health insurance data were collected for only half the sample in 1993, and the change in survey instrument in mid-year 1996, resulted in a 3/8 reduction in sample size for that year as well.

736

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

collections. Data collected by Simon (1999) through surveys of state insurance departments and review of state statutes, indicate the month and year in which various regulations were implemented in each state. The organization of Simon’s data allowed for annual snapshots of the regulations in place for each year, and we used those data for 1993 through 1995. For 1996, we used a cross sectional snapshot that was assembled by Curtis and colleagues for Long and Marquis (1998). We also drew on information concerning passage of legislation assembled by the Blue Cross and Blue Shield Association (2000). When the Simon data were not fully consistent with Long and Marquis, we gave precedence to Long and Marquis. 4.2. Sample Our primary analytic sample consists of workers, aged 18 through 64. We defined workers based on data collected by the NHIS on employment in the past two weeks. We excluded persons who reported 0 h of work during the past month and workers who were primarily self-employed. Our sample size over the four year period was 106,335 workers. 4.3. Dependent variables We used three key measures of ESI offer and coverage for workers in our analysis: whether the worker is covered by own or spouse ESI (AnyESI); whether any adult worker in the family is offered ESI (AnyOffer); and whether the worker takes-up ESI, conditional on an adult worker in the family having an offer (Take-Up). Among all workers, across the four years of data, 86.5% had an offer of ESI in the family and 77.2% had ESI coverage. Among those with offers in the family, take-up was 92.4%.4 The dependent variables are designed to account for family level decision making in two-worker families. The vast majority of employers that offer ESI to their workers also offer dependent coverage, and many two-earner couples have access to two offers of health insurance. However, to provide family coverage, families with two workers need only ensure that one worker has an offer of dependent coverage, thus it is important to incorporate information on ESI offers and coverage from the employer of a working spouse. The NHIS collects information on prior month insurance coverage. For private plans, the NHIS collects information on whether the policy was sponsored by an employer or union and whether the policy is held by the worker. We treated union sponsored coverage as if it were employer sponsored. We identified workers with coverage sponsored by their own or a spouse’s employer. We constructed a measure of offer based on data from the Health Insurance Supplement on the NHIS. The offer question on the NHIS asks whether the worker was offered insurance coverage by her/his employer. The question is skipped for workers who reported a private insurance policy in their own name. We logically imputed an employer offer to those workers.5 4 The estimation sample for the Any ESI model includes some workers for whom the family offer measure is missing. Likewise, the Family Offer model includes workers for whom the Any ESI measure is missing. 5 The offer question was also skipped for workers with a non-group policy in their own name. In the current analysis we set the offer variable to missing, although we know from other sources that very few workers with non-group coverage received and declined ESI offers.

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

737

4.4. Measuring key explanatory variables The key explanatory variables of interest are measures constructed from the complex set of insurance regulations, worker and family health status, and information on firm size for the worker and spouse. There are many ways to describe alternative insurance reform regimes. As discussed in Section 2, some states have implemented packages of reforms hypothesized to be sufficiently restrictive to have an effect on insurance coverage, whereas other states have implemented very limited packages of reforms that are unlikely to meet a threshold of activity. In between are states with reform packages that may have smaller or even different effects. Some states implemented no reforms. In this paper we employ a three-tiered summary measure. STRONG reform indicates that in a given year a state had guaranteed issue of all products in the small group market and some restrictions on premium variance across different risk groups.6 MEDIUM reform states were defined as having either guaranteed issue of all products or guaranteed issue of at least some products and some sort of premium variance restrictions.7 It should be noted that states that met these criteria also had passed portability and guaranteed renewal requirements, and limits on pre-existing condition restrictions, so their set of insurance issue reforms was relatively complete. Those states that had only implemented guaranteed issue of some products, limits on pre-existing condition exclusions, and guaranteed renewal were grouped with states that had no reforms. Table 2 identifies states that implemented specific reforms by type and year. Over the four-year period, 19 states implemented guaranteed issue of some products, and 9 states newly implemented guaranteed issue of all products or switched from some to all products. Twenty states implemented some type of rate variance reforms. Though not part of health insurance market reform legislation, the existence of a highrisk pool, a partially subsidized source of coverage for those who have been turned down by private insurers because of their health risk, can be an important element in a state’s insurance market environment (Communicating for Agriculture, 2003). We hypothesize that insurers in states with a high-risk pool are less concerned about adverse selection, since high-risk persons have an alternative source of insurance. Thus, we expect states with high-risk pools to have lower average premiums in the commercial market, and more private coverage, ceteris paribus, than states without high-risk pools. We included a dummy indicator for whether a state had a high-risk insurance pool in operation during each year. In a sensitivity analysis we interacted reforms with the presence of a high-risk pool. Health status is even more multidimensional than health insurance market reforms, but we require a single measure to work with the DDD regression model that has multiple interaction terms. Following Bound et al. (1999), we estimated a model of health risk,

6 Ideally we might create a hierarchy based on the tightness of the restrictions on rating that are implied by the various state laws and use that in our algorithm to describe the strength of reforms. However, the laws use different bases from which rates can vary, and provide insurers with much leeway in how to operationalize the rating restrictions. Thus, creating such a hierarchy or index would be a complex process outside the scope of this project. Furthermore it is not essential to test the impact of rating restrictions on insurance coverage. 7 In fact, all states with guaranteed issue of all products had premium variance restrictions, and thus were classified as strong reform states.

738

Year

Specific reform implemented

Analytic reform groups

Guaranteed Issue—any products

Guaranteed Issue—all products

Rating reforms

Strong

Medium

1993

AK, CA, ID, MI, MT, TN

MN, NY

MN, NY

AK, CA, ID, MT, TN

1994 1995 1996

AZ, CO, MD, MO, NJ, ND, NE, OK KY, MS, NM, SC, SD

ME, TX CA, MD, NH KY, OR

AK, CA, ID, IL, MN, MO, MT, NH, NY, TN AZ, MD, MO, NJ, TX KY, CO, MS, UT NV

ME, TX CA, MD, NH KY, OR

AZ, MD, MO, ND, NE, NJ, OK CO, KY, MS, NM, SC, SD

Source: Urban Institute analysis of plan implementation data from Long and Marquis (1998) and Simon (1999).

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Table 2 States with newly implemented insurance market reforms by year

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

739

with an indicator for whether any member of the family was in fair or poor health as the dependent measure. The explanatory variables included indicators for whether any member of the family had a series of common chronic conditions. The estimated model was used to predict the probability that a worker or their family members were in fair or poor health, which was used to construct the health risk indicator in our models.8 The model included chronic conditions that had a prevalence of at least 2% among workers in the U.S. in 1997. Our set of conditions included arthritis, slipped disc, dermatitis, bursitis, diabetes, migraine, hypertension, heart disease, varicose veins, chronic bronchitis, asthma and sinusitis. Since these conditions are chronic, their reporting is less subject to endogeneity bias than more temporary acute conditions, where current insurance coverage would increase the likelihood of diagnosis and respondent awareness of the condition. Since most uninsured are uninsured for less than a year (Swartz and McBride, 1990), the likelihood of a chronic condition being known to someone who is currently uninsured is higher than that of a temporary acute condition. Essentially, the predicted probability of fair/poor health that results from this approach is one that reflects the objective components of self-reported health status. One of the strengths of the NHIS survey during this period is the broad array of medical conditions that are included in the instrument. However, use of the condition data is complicated by the survey design. Conditions on the survey are divided into six groups or condition lists, and one list is assigned randomly to each household in the survey. Thus, each household responds to questions about one sixth of the conditions. To accommodate this feature of the condition data, we controlled for the condition list assigned to the family, using a series of dummy indicators. Each condition specific indicator was interacted with the indicator for the condition list on which it appeared. We also included worker age, race, education, marital status, and interactions between each of the conditions and worker gender and age. Variable means, and results from this model are available from the authors upon request. The model was highly significant, with a pseudo R-squared of 0.18. In the DDD models, we included measures of worker demographics, such as age, gender, race and Hispanic ethnicity, an indicator for persons not born in the U.S., and highest educational attainment. We also included family earnings for the prior month (in $1000s) and level of labor force participation, with variables indicating part-time work at 20–35 h and a second indicator for working less than 20 h weekly. We included a series of dummy indicators for the worker’s industry (using the service industry as the reference category). We estimated the probability of family offer of ESI, and coverage through either own or spouse ESI, thus, we controlled for characteristics of a married worker’s spouse that would be correlated with whether the spouse worker had an ESI offer or ESI coverage. We included an indicator for marriage to a non-worker, marriage to a worker in an industry with a low likelihood of offering ESI, and marriage to a worker in a high-offer industry. 9 8 The predicted probability of fair/poor health is transformed into a yes/no binary variable. A random number from a uniform distribution is assigned to each worker. If the predicted probability is greater than or equal to the random number, the binary variable is set equal to 1. If the random number exceeds the predicted probability, the binary variable is set equal to zero. We took this approach in order to be able to draw conclusions about the healthy relative to the not healthy, a comparison that requires a binary measure. 9 Workers who reported being separated were dropped from the sample because we were uncertain whether they would have access to a working spouse’s offer of ESI.

740

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

As discussed previously, workers in small firms are at a disadvantage with respect to ESI offers and premiums. However, the presence of a second worker in the family who is employed by a large firm increases the range of insurance choices, beyond those of workers who only have access to insurance through a small firm. Our measure of firm size captures whether all workers in a family were employed in a small firm, defined as having less than 50 employees. The reference category captures families where at least one worker is employed at a large firm. To capture the effect of public insurance availability on offer and take-up, we included the income threshold that would be used to determine whether the worker’s family, given the observed family size, would be eligible to receive cash assistance through the Aid to Families with Dependent Children (AFDC) program in the observed state and year. Eligibility for AFDC is linked to eligibility for family coverage through the Medicaid program. We also included the Medicaid poverty expansion threshold that would be relevant to the youngest child in a family, in the given state and year. In all models, we also controlled for the potential effects of HMOs on insurance premiums by including a measure of private HMO penetration. This county level measure is based on enrollment data from InterStudy that is parsed to the level of the county.10 In our models of any ESI and in family offer of ESI we controlled for the proportion of firms in the county that have less than 50 employees, and for the proportion of firms in the service industry. Data for both measures are taken from County Business Patterns data collected by the Bureau of the Census.11 These variables were excluded from models that estimate ESI take-up conditional on offer. To construct our analysis files, we linked state and county level variables to the worker level records in the NHIS. County and state level identifiers are not available on the public use NHIS files. We were able to create this linkage through use of the Research Data Center at the National Center for Health Statistics. Our analyses were adjusted for intra-familial correlation. However, the NHIS uses a complex multistage sample design. Due to computational constraints, we were unable to perform a direct adjustment of the estimated standard errors to correct for the sample design. As a result, the unadjusted standard errors may be understated.

5. Results 5.1. Descriptive analyses Workers in the target group for reforms differ from workers in the various control groups along a number of dimensions that are likely to affect both offer and take-up of ESI. Table 3 presents mean values of the dependent and explanatory variables for all workers, and then for high-risk-small firm workers, low-risk-small firm workers, and high-risk-large firm workers. Relative to the policy target group, low-risk-small firm workers are more likely to 10

We are grateful to Dr. Douglas Wholey of the University of Minnesota for providing these data to us. Data on firm and industry composition by county and year downloaded from www.census.gov/pub/download/ cbpdownload, access February 2001. 11

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

741

Table 3 Characteristics of workers in policy target and comparison group All workers (n = 108,664)

High-risk, small firm workers (n = 3005)

Low-risk, small firm workers (n = 22,258)

High-risk, large firm workers (n = 9311)

Dependent variables Worker has any ESI coverage Any offer in the HIU Worker has any take-up/offer

77.2% 86.5% 92.4%

48.3% 60.0% 84.2%

54.5%* 65.0%* 87.4%*

82.5%* 91.6%* 92.3%*

Worker age and sex 18–24 25–34 35–44 45–54 55–64 Female

12.5% 28.7% 29.1% 20.5% 9.2% 46.6%

3.4% 16.5% 30.8% 25.7% 23.5% 50.4%

20.0%* 30.8%* 24.7%* 16.0%* 8.6%* 42.3%*

2.2%* 15.2% 30.7% 32.0%* 19.9%* 53.3%*

Race/ethnicity/immigrant status Black, not hispanic Other, not hispanic Hispanic White, not hispanic Immigrant

10.3% 4.2% 8.6% 76.9% 11.5%

11.0% 3.6% 12.8% 72.6% 12.5%

7.1%* 4.3% 11.0%* 77.7%* 14.7%*

15.3%* 4.5% 9.4%* 70.8% 9.9%*

Education College graduate Some college High school graduate Less than high School

27.5% 24.7% 37.3% 10.5%

5.5% 14.8% 48.1% 31.6%

22.1%* 24.5%* 39.5%* 13.9%*

8.9%* 17.5%* 53.7%* 19.8%*

Reforms High-risk pool Strong reform Medium reform

53.6% 29.8% 41.5%

54.5% 28.1% 42.9%

54.5% 30.7%* 40.8%

52.3% 28.3% 44.2%

Firm size All workers in small firm

23.2%

100.0%

100.0%

0.0%

Health status Predicted HIU fair/poor health

10.8%

100.0%

0.0%

100.0%

Family status HIU size Not married Married to non-worker Married to worker in low offer industry Married to worker in high offer industry HIU earnings

2.5 31.9% 20.2% 26.8% 21.1% 3.3

2.8 17.6% 52.7% 18.6% 11.2% 2.0

2.2* 46.7%* 25.8%* 19.0% 8.5%* 2.4*

2.9 11.8%* 34.5%* 27.3%* 26.5%* 3.2

MSA size Greater than 1 million 250,000–999,999 Under 250,000 Non-MSA

48.6% 26.7% 8.5% 16.2%

40.7% 27.8% 9.0% 22.5%

47.9%* 24.8%* 8.6% 18.6%*

44.1%* 28.1% 8.2% 19.6%*

742

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Table 3 (Continued ) All workers (n = 108,664)

High-risk, small firm workers (n = 3005)

Low-risk, small firm workers (n = 22,258)

High-risk, large firm workers (n = 9311)

Industry Agriculture Mining Construction Manufacturing Transportation Wholesale trade Retail trade FIRE public administration Services

1.4% 0.4% 5.6% 18.8% 8.0% 3.5% 15.3% 6.8% 5.5% 34.6%

3.0% 0.5% 10.1% 14.8% 5.4% 4.1% 19.4% 6.2% 2.6% 33.8%

3.4% 0.3% 11.0% 13.1%* 5.2% 4.4% 20.3% 6.0% 1.5%* 34.7%

0.7%* 0.4% 4.0%* 23.2%* 9.9%* 2.9%* 13.7%* 5.6% 7.2%* 32.3%

Work hours 20–34 h Under 20 h >35 h

10.9% 3.5% 85.6%

18.8% 7.8% 73.3%

14.2%* 4.8%* 81.1%*

11.9%* 3.9%* 84.2%*

4.5 0.5

4.7 0.4

4.4* 0.4*

22.1% 5.6 94.9% 36.1%

20.0% 6.0 95.1% 35.3%

21.7%* 5.9* 95.0%* 35.9%*

Medicaid eligibility Payment standard Poverty threshold County level measures HMO penetration Unemployment rate Pct of est with firm size <50 Pct of est in service industry

4.6 0.4 21.3%* 5.7* 94.9%* 35.7%*

Source: Authors’ analysis of the 1993–1996 National Health Interview Survey. ∗ Indicates significant difference from high-risk workers in small firms at 0.05 level.

have an ESI offer, and more likely to take-up an offer of ESI. These workers are younger, less likely to be black or Hispanic, have higher educational attainment, are more likely to be unmarried and less likely to be married to a non-worker, and are more likely to work full time. High-risk, large firm workers are much more likely to have ESI coverage, an offer of ESI, and have much higher take-up rates, when compared with high-risk, small firm workers. They tend to be older, more likely to be black, have higher educational attainment, are more likely to be married to workers in high-offer industries, and have higher earnings. A descriptive comparison across the health risk and firm size groups, suggest that reforms had a negative effect on low-risk, but no significant effects on high-risk workers in small firms. Table 4 presents means for each of the three dependent variables (any ESI coverage, ESI offer, and ESI take-up conditional on offer), for target workers and workers in the comparison groups, comparing workers subject to either strong or medium reform regimes to workers not subject to reforms. Both strong and medium reforms were associated with higher rates of coverage for high-risk workers in small firms, compared to the “no reform” workers, but the differences were not significant. Low-risk workers had

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

743

Table 4 Percentage of workers with ESI coverage, offers and take-up, by small group market reform, for policy target and comparison groups Any ESI coverage

Reform regime Strong

Medium

None

Target group High-risk, small firms High-risk, large firms Low-risk, small firms Low-risk, large firms

47.2%†,§ 83.3% 52.6%*,§ 83.9%*

50.0%§ 82.1%† 53.4%*,§ 84.3%*

46.1%†,§ 82.3%† 58.0%§ 85.5%

Any offer High-risk, small firms High-risk, large firms Low-risk, small firms Low-risk, large firms

59.3%§ 90.7%† 62.9%*,§ 92.2%*

61.7%§ 92.1% 64.2%*,§ 92.6%*

57.8%†,§ 91.8%† 68.1%§ 93.2%

Take-up High-risk, small firms High-risk, large firms Low-risk, small firms Low-risk, large firms

83.3%†,§ 93.4% 87.9%§ 93.8%

85.6%§ 91.6%† 85.8%*,§ 93.1%*

82.7%†,§ 92.2%† 88.9%§ 93.9%

Source: Authors’ analysis of 1993–1996 National Health Interview Survey. † Indicates a significant difference between high-risk and low-risk at the 0.05 level. § Indicates a significant difference between small firms and large firms at the 0.05 level. ∗ Indicates a significant difference between strong/medium and no reform at the 0.05 level.

lower rates of any ESI coverage, ESI offer, and take-up under medium reforms relative to no reforms, and had even lower rates of any ESI coverage and ESI offer under strong reforms compared to no reforms. These results are suggestive, but adjustment for the different characteristics across the target and comparison groups is essential to test these relationships. 5.2. Estimated effects of reforms on any ESI coverage for workers in small firms Table 5 presents the DD and DDD estimate of the effects of reforms on various target and comparison groups. The three key DD results are the effects of reforms on (1) highrisk workers in small versus large firms; (2) low-risk workers in small versus large firms, and (3) high-risk versus low-risk workers in small firms. The DDD results measure the effects of reform on high-risk workers in small versus large firms, compared with the effects of reforms on low-risk workers in small versus large firms. The first columns in the table provide the effects of reform on Any ESI coverage. The top section shows the estimated effects of strong reforms and the bottom section, the estimated effects of medium reforms. The DD and DDD estimates of the effects of reforms on various target and comparison groups reveal that many of the group differences are explained by the differences in observable characteristics, leaving relatively few significant results. The results are qualitatively similar for strong and medium reforms, but estimated effects for medium reforms

744

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Table 5 Estimated effects of small group market reforms on worker ESI coverage, offers, and take-up: DD and DDD results Any ESI coverage Marginal effect

Any offer

ESI take-up

P-value marginal effect

P-value Marginal effect

P-value

Strong reform High-risk, small firms 0.030 Low-risk, small firms −0.009 High-risk, large firms 0.019 Low-risk, large firms 0.006 High-risk, small vs. large firms (DD) 0.011 Low-risk, small vs. large firms (DD) −0.015 Small firms, high vs. low-risk (DD) 0.039 Large firms, high vs. low-risk (DD) 0.014 High- vs. low-risk, small vs. large firms (DDD) 0.026

0.211 0.472 0.146 0.462 0.913 0.070 0.086 0.257 0.452

0.040 0.004 0.002 0.005 0.037 −0.001 0.035 −0.003 0.038

0.070 0.734 0.819 0.407 0.153 0.564 0.084 0.736 0.122

−0.004 −0.005 0.010 0.000 −0.014 −0.005 0.001 0.010 −0.009

0.881 0.626 0.328 0.998 0.463 0.533 0.915 0.257 0.643

Medium reform High-risk, small firms 0.037 Low-risk, small firms −0.008 High-risk, large firms 0.014 Low-risk, large firms 0.008 High-risk, small vs. large firms (DD) 0.023 Low-risk, small vs. large firms (DD) −0.017 Small firms, high vs. low-risk (DD) 0.045 Large firms, high vs. low-risk (DD) 0.006 High- vs. low-risk, small vs. large firms (DDD) 0.039

0.094 0.463 0.238 0.239 0.477 0.026* 0.034* 0.615 0.143

0.027 −0.002 0.013 0.003 0.014 −0.005 0.029 0.010 0.019

0.195 0.849 0.165 0.556 0.935 0.346 0.141 0.250 0.677

0.025 −0.015 0.003 0.000 0.022 −0.015 0.040 0.003 0.037

0.221 0.118 0.721 0.971 0.343 0.056 0.045* 0.661 0.116

Source: Authors’ analysis of 1993–1996 National Health Interview Survey. Note: Marginal effect calculated as difference in predicted values of dependent variables when reform indicators are alternatively set to one or zero, and the remaining interaction terms are re-computed. The P-value reflects the level of significance of the linear sum of coefficients that comprise each DD result. Strong reforms include states with guaranteed issue of all products and some type of rate variance restriction Medium reforms include states with guaranteed issue of some products and some type of rate variance restriction. ∗ Indicates significance at the 0.05 level.

are slightly larger and are more likely to meet standards of significance. We find a weakly significant positive effect of strong reforms on any ESI coverage for high versus low-risk workers in small firms, and a weak negative effect of reforms on low-risk workers in small versus large firms. Medium reforms have a weakly significant positive effect on any ESI coverage for high-risk workers in small firms. We find a significant positive effect (4.5 percentage points) of medium reforms on any ESI coverage when high-risk workers are compared with low-risk workers in small firms. We also find a small negative effect (−1.7 percentage points) of reforms on low-risk workers in small firms compared with large firms. Neither of the estimated DDD effects were significant, although the DDD result for medium reforms suggests a positive effect (P = 0.143). Overall the magnitude of the effects is small, and the direction of the effects consistent with a positive effect of reforms on high-risk workers and a negative effect on low-risk and small firm workers.12 12

In sensitivity analyses we combined the strong and medium reforms and find slightly smaller effects with similar patterns of significance to those of medium reforms.

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

745

5.3. Effects of reform on the probability of an ESI offer and on accepting or taking up an ESI offer When we estimate the system of offer and take-up equations, we find evidence of worker sorting as a function of their preferences for an ESI offer, and different effects of reforms on the two stages associated with obtaining ESI. The estimated correlation across the equations, as measured by rho, was highly significant (Chi-squared = 9.95; P = 0.003), providing evidence of sorting. We failed to find significant effects of strong reforms on either offer or ESI take-up (second and third sets of columns in Table 5). There were weakly significant positive effects of strong reforms on the likelihood of offer to high-risk compared with low-risk workers within small firms, and the positive marginal DDD effect (0.038) of strong refoms on offer for high-risk small firm workers is suggestive (P = 0.122). The effects of medium reforms on any ESI coverage appear to work through an effect on take-up rather than offer. Medium reforms had a significant positive effect on take-up for high versus low-risk workers within small firms, and a weakly significant negative effect on take-up for low-risk workers in small versus large firms. The positive DDD effect (0.037) of medium reforms on take-up for high-risk, small firm workers was suggestive (P = 0.116). The complete set of estimated coefficients from the model are provided in Appendix A (Table A.1). 5.4. Effects of reforms in the presence of high-risk pools We tested whether the presence of a high-risk pool would alter the effects of reforms on workers by adding an interaction term between risk pool and reforms into models of any ESI coverage, estimated separately for small and large firms. The results suggest that the presence of a high-risk pool has a protective effect for low-risk workers in small firms under strong reforms (data not shown). From Table 5 we recall that there was no effect of strong reforms on low-risk workers in small firms overall. However, under a regime of strong reforms where there is no high-risk pool, low-risk workers in small firms face a decreased likelihood of any ESI coverage (−0.119, P = 0.005). There are no comparable effects on high-risk workers, workers in large firms, or for medium reforms.

6. Discussion In this analysis, we test whether state private insurance market reforms had a differential effect on coverage for workers with high-risk families in small firms. Our findings suggest that reforms had small positive effects on ESI coverage for the target group of high-risk small firm workers, but had unintended negative side effects for low-risk workers in small firms. Reform packages that combined guaranteed issue of some products with premium variance restrictions had positive effects on high-risk workers in small firms, relative to low-risk workers. However, reforms also had an unintended negative effect on coverage for low-risk workers in small firms, particularly relative to workers in large firms. Reform packages that combined guaranteed issue of all products with premium variance restrictions had effects that were in the same direction, but smaller in magnitude and were not significant. In general, the magnitude of the effects was quite small.

746

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Reform packages that include premium variance restrictions and guaranteed issue of some products tended to have significant effects on take-up of ESI. In contrast, the strong reform package, with guaranteed issue of all products, results in significant effects on offer. This pattern is consistent with the expectation that strong reforms force insurers to raise premiums more than medium reforms, and this reduces take-up. Our findings are roughly consistent with Simon (2002) and Monehit and Schone (2003). In addition, our findings on the protective effects of high-risk pools for low-risk small firm workers are consistent with results from the Marstellar et al. (1998) study. The results suggest that the high-risk pool acts as a safety valve for insurers, reducing the number of very high cost individuals in the private market. These results have implications for the impact of HIPAA on access to insurance for small employers. HIPAA required states to mandate that insurers offer all insurance products to small employers. This requirement would have resulted in a shift from medium to strong reforms for 27 states, according to our hierarchy; eight states would have shifted from having only weak or no reforms to having strong reforms, and four states would shift from weak or no reforms to medium reforms. Since our results suggest that strong reforms have smaller effects than medium reforms, HIPAA may actually have resulted in some loss of coverage for small firm workers. Offer likely increased, but takeup likely suffered. Chollet et al. (2000) come to a similar conclusion about HIPAA’s effects. A limitation to our analysis is that we use discrete groupings for the reform measures. Since states vary dramatically with respect to how they specify the permitted width of rating bands, our estimates represent an average effect of the reform packages actually implemented by states. With a more continuous measure of rate band widths, we might be able to draw some conclusions about the relationship between this aspect of risk pooling and coverage. Creating a meaningful index of rating restrictions is exceedingly difficult to do well, however, since allowable rating factors as well as the width of bands vary considerably by state. Another potential limitation is that our analysis presumes that state policies are exogenous; failure of this assumption would result in biased estimates of the effects of reforms. Simon (2002) addresses this issue explicitly, finding no correlation between state specific levels of ESI coverage for small firm workers and the likelihood that various regulations would be implemented. A related limitation is that we measure policy implementation by the states, but do not qualify with information on enforcement of policies. If enforcement of policies is randomly allocated across states that have implemented policies, measurement error is introduced, and the estimated effects of reforms will be biased downwards. However, if states with greater concern about access to ESI for high-risk workers are more likely to enforce regulations, then Simon’s analysis concerning endogeneity of state policy does not adequately address the issue. One of the difficulties faced in attempting to conceptualize and evaluate the effects of insurance market reforms is the absence of information on how high-risk workers are distributed across individual small firms. High-risk workers may be distributed randomly across firms, or there may be some sorting of workers by risk type, so that high-risk workers are clustered in a subset of firms. Firms might attract (or discourage) high-risk workers through various benefit policies, such as flexible leave, or through informal policies that make a work

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

747

environment more or less supportive. Furthermore, workers may sort themselves based on preferences for ESI. This issue of homogeneity of workers within firms is important because the degree of sorting may determine how disadvantaged high-risk workers are in the absence of reforms, and the effect of reforms on the likelihood of ESI coverage. To the extent that risk is distributed randomly, then the number of high-risk workers will be very limited in most small firms, and these workers may not be at a particular disadvantage relative to low-risk workers in the same firms. To the extent that reforms increase premiums for small firms generally, then these isolated high-risk workers will be hurt to the same extent as low-risk workers. However, if high-risk workers are clustered within firms, then they are more likely to be disadvantaged in the absence of reforms, and reforms are more likely to have a positive effect on offers and reduce premiums relative to the pre-reform state. In this analysis, we have information on worker risk status and firm size, but we do not know whether that worker is in a firm with many other high-risk workers, or may be the isolated high-risk worker, and no data set contains such information. Since one can imagine opposite effects of reforms depending on how the high-risk workers are distributed, this may explain why we do not find larger effects on high-risk workers. This analysis may also indicate that while small group insurance market reforms forced additional pooling across small firms, the degree of pooling remained insufficient to protect the insurance opportunities for high-risk workers. It may be that expanded pooling across a broader population, i.e., including small firms in the large firm rating pool, would have been more successful at increasing access for the high-risk worker while imposing more modest premium increases on the healthy. Based on the results for low-risk workers in small firms, it seems that a fundamental failing of the reforms is that they did not address the higher premiums faced by all small firms relative to large firms. Increasing risk pooling across small firms does not address the higher administrative costs, nor the risk associated with insuring a particular firm. The development of alternative strategies for more broadly spreading risk across the full population of insured workers and dependents could be useful in addressing these problems and should be a focus for policy related research in the future. Acknowledgements This research was conducted while Amy Davidoff was a Research Associate at The Urban Institute. We wish to thank Jack Hadley, Stephen Zuckerman, and Bowen Garrett and two anonymous reviewers for insightful comments on an earlier version of the paper. Stacey McMorrow and Sarbajit Sinha provided excellent research assistance. The research was funded by grant No. 039135 from the Robert Wood Johnson Foundation through its initiative on Health Care Financing and Organization. Opinions expressed are those of the authors and do not necessarily reflect the positions of the University of Maryland Baltimore County, The Urban Institute or its funders. Appendix A The complete set of estimated coefficients from the model are provided in Table A.1 .

748

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Table A.1 Estimated coefficients from DDD models of ESI, offer, and take-up Any ESI

Heckprob model Offer

Take-up/Offer

Coefficient

Std. error

Marginal effect

Coefficient

Std. error

Coefficient

Std. error

Worker age and sex 18–24 25–34 35–44 45–54 Female

−0.574 −0.432 −0.268 −0.137 0.062

0.024 0.022 0.022 0.022 0.010

−0.174* −0.119* −0.072* −0.036* 0.016*

−0.726 −0.358 −0.269 −0.138 0.034

0.030* 0.029* 0.029* 0.029* 0.012*

−0.811 −0.522 −0.326 −0.153 0.024

0.041* 0.035* 0.035* 0.034* 0.014

Race Black, not hispanic Hispanic Other, not hispanic Immigrant

−0.234 −0.141 −0.172 −0.258

0.017 0.019 0.029 0.018

−0.064* −0.038* −0.047* −0.071*

−0.116 −0.147 −0.090 −0.314

0.022* 0.023* 0.037* 0.022*

−0.211 −0.053 −0.184 −0.076

0.024* 0.028 0.043* 0.028*

Education College graduate Some college High school graduate

0.541 0.490 0.364

0.020 0.018 0.016

0.121* 0.109* 0.088*

0.546 0.397 0.333

0.025* 0.022* 0.019*

0.481 0.347 0.250

0.031* 0.027* 0.024*

Reforms High-risk pool Strong reform Medium reform

0.179 0.027 0.039

0.105 0.037 0.033

0.046 0.007 0.010

0.218 0.041 0.026

0.144 0.049 0.044

0.187 0.000 −0.002

0.147 0.052 0.046

Firm size All workers in small firm

−0.634

0.023 −0.185*

−0.809

0.029*

−0.186

0.040*

Health Status Predicted fair/poor health

−0.062

0.037 −0.016

−0.077

0.051

−0.077

0.048

0.016

0.166

0.107

−0.064

0.139

0.058 −0.058 0.122

0.051 0.014 0.032 −0.015 0.083 0.029

−0.023 −0.023 −0.080

0.068 0.039 0.079

0.077 −0.031 −0.120

0.068 0.050 0.098

0.024 −0.066 −0.141

0.047 0.006 0.030 −0.017* 0.064 −0.038*

0.043 0.074 −0.035

0.102 0.065 0.037

0.205 0.027 −0.089

0.130 0.061 0.047

0.029 −0.015 0.243

0.008 0.007* 0.017 −0.004 0.021 0.057*

0.003 0.005 0.539

0.011 0.022 0.031*

−0.009 −0.112 −0.270

0.012 0.027* 0.030*

Interactions Small firm* strong* fair/poor health Strong* fair/poor health Small firm* strong Small firm* medium* fair/poor health Medium* fair/poor health Small firm* medium Small firm* fair/poor health Family status HIU size Married to non-worker Married to worker in high offer industry Married to worker in low offer industry HIU earnings

0.067

0.089

0.048

0.019

0.012*

0.252

0.026*

−0.375

0.027*

0.145

0.005

0.037*

0.167

0.008*

0.140

0.008*

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

749

Table A.1 (Continued ) Any ESI

Heckprob model Offer

Coefficient

Std. error

Marginal effect

0.023 0.021 0.025

0.003 0.007 0.001

Take-up/Offer

Coefficient

Std. error

Coefficient

Std. error

−0.032 0.000 −0.014

0.029 0.027 0.031

0.022 −0.012 −0.013

0.028 0.028 0.033

MSA Size Greater than 1 million 250,000–999,999 Under 250,000

0.011 0.028 0.003

Industry Agriculture Mining Construction Manufacturing Transportation Wholesale trade Retail trade FIRE Public administration

−0.296 0.438 −0.195 0.325 0.178 0.190 −0.152 0.200 0.427

0.038 −0.085* 0.078 0.088* 0.021 −0.053* 0.016 0.074* 0.021 0.042* 0.029 0.044* 0.014 −0.041* 0.023 0.047* 0.029 0.089*

−0.317 0.310 −0.231 0.349 0.146 0.223 −0.175 0.233 0.581

0.042* 0.108* 0.026* 0.021* 0.027* 0.038* 0.017* 0.031* 0.043*

−0.101 0.353 −0.077 0.231 0.180 0.152 −0.117 0.200 0.284

0.061 0.113* 0.032* 0.023* 0.029* 0.041* 0.021* 0.033* 0.039*

Work hours 20–34 h Under 20 h

−0.410 −0.458

0.015 −0.119* 0.025 −0.138*

−0.611 −0.851

0.017* 0.028*

−0.257 −0.196

0.026* 0.045*

Medicaid eligibility Payment standard Poverty threshold

−0.027 0.013

0.004 −0.007* 0.014 0.003

−0.046 0.055

0.005* 0.018*

0.025 0.028

0.007* 0.018

County level measures HMO penetration Unemployment rate Pct of est with firm size < 50 Pct of est in service industry

0.188 −0.010 −2.866 −0.643

0.063 0.048* 0.003 −0.003* 0.598 −0.726* 0.203 −0.163*

0.210 −0.022 −3.893 −0.872

0.080* 0.004* 0.771* 0.262*

0.153 n/a n/a n/a

0.085 n/a n/a n/a

Note: Model also includes vector of state and year indicators. Source: Authors’ analysis of the 1993–1996 National Health Interview Survey. ∗ Indicates significance at the 0.05 level.

References Adams, P.F., Marano, M.A., 1995. Current Estimates from the National Health Interview Survey, 1994. Vital and Health Statistics 10, 130–134. Benson, V., Marano, M.A., 1998. Current Estimates from the National Health Interview Survey, 1995. Vital and Health Statistics 10, 125–129. Blue Cross Blue Shield Association, 2000. State Legislative Health Care and Insurance Issues: 2000 Survey of Plans. Bound, J., Schoenbaum, M., Stinebrickner, T., Waidmann, T., 1999. The dynamic effects of health on the labor force transitions of older workers. Labour Economics 6 (2), 179–202. Chollet, D., Simon, K., Kirk, A. 2000. What Impact HIPAA? State Regulation and Private Health Insurance Coverage Among Adults. Office of the Assistant Secretary for Health and Human Services, Washington, DC. Accessed at URL: http://aspe.hhs.gov/health/reports/hippaimpact/index.html).

750

A. Davidoff et al. / Journal of Health Economics 24 (2005) 725–750

Communicating for Agriculture, 2003. Comprehensive Health Insurance for High-risk Individuals—A State by State Analysis, sixteenth ed., 2002/2003. CA, Inc. Fergus Falls, MN. Fronstin, P. 2003. Sources of health insurance and characteristics of the uninsured: analysis of the March 2003 current population survey. EBRI Issue Brief Number 264. Employee Benefits Research Institute, Washington, DC. Garrett, B. 2004. Employer-sponsored health insurance coverage: sponsorship, eligibility, and participation patterns in 2001. Report prepared for the Kaiser Commission on Medicaid and the Uninsured. The Kaiser Family Foundation, Washington, DC. Gruber, J., 1994. State mandated benefits and employer-sponsored health insurance. Journal of Public Economics 55, 433–464. Hall, M.A., 2000. The structure and enforcement of health insurance rating reforms. Inquiry 37, 376–388. Long, S.H., Marquis, M.S., 1998. Baseline information for evaluating the implementation of the health insurance portability and accountability act of 1996. Final Report Prepared for the Health Care Financing Administration under Contract 500-95-0056, Task Order 4. October. Marstellar, J.A., Nichols, L.M., Badawi, A., Kessler, B., Rajan, S., Zuckerman, S., 1998. Variations in the Uninsured: State and County Analyses. Urban Institute. Online at http://www.urban.org/health/variatfr.html. Monehit, A.C., Schone, B.S., 2003. How has small group market reform affected employee health insurance coverage? Journal of Public Economics 88, 237–254. Newhouse, J.P., 1996. Reimbursing health plans and health providers: efficiency in production versus selection. Journal of Economic Literature 34, 1236–1263. Nichols, L.M., 2000. State regulation: what have we learned so far? Journal of Health Politics, Policy and Law 25, 175–196. Percy, A.M., 2001. Community rating and regulatory reform in the small group and individual health insurance markets. Paper presented at the American Economic Association Annual Meeting. New Orleans, LA. Rothschild, M., Stiglitz, J., 1976. Equilibrium in competitive insurance markets: an essay on the economics of imperfect information. Quarterly Journal of Economics 90, 629–650. Simon, K.I., 1999. The Impact of small-group health insurance reform, 1990–1996. Dissertation. Department of Economics, University of Maryland. Simon, K.I., 2002. Adverse Selection in Health Insurance Markets: Evidence from State Small-Group Health Insurance Reforms. Working Paper. Department of Policy Analysis and Management, Cornell University. Sloan, F.A., Conover, C.J., 1998. Effects of state reforms on health insurance coverage of adults. Inquiry 35, 280–293. Swartz, K., McBride, T.D., 1990. Spells without health insurance: distributions of durations and their link to point-in-time estimates of the uninsured. Inquiry 27, 281–288. Zuckerman, S., Rajan, S., 1999. An alternative approach to measuring the effects of insurance market reforms. Inquiry 36, 44–56.