How has small group market reform affected employee health insurance coverage?

Journal of Public Economics 88 (2003) 237 – 254 www.elsevier.com/locate/econbase

How has small group market reform affected employee health insurance coverage? Alan C. Monheit a, Barbara Steinberg Schone b,* a School of Public Health, Division of Health Systems and Policy, University of Medicine and Dentistry of NewJersey, Piscataway, NJ 08854, USA b Center for Cost and Financing Studies, Agency for Healthcare Research and Quality, 2101 E. Jefferson St., Suite 500, Rockville, MD 20852, USA

Received 6 January 1999; received in revised form 28 May 2002; accepted 4 June 2002

Abstract In the early 1990s, over 40 states passed legislation designed to limit a number of exclusionary practices by insurers in the small group market in order to improve the availability and affordability of health insurance to employees in small firms. In this paper, we address the effects of reform on the likelihood that workers are offered insurance, have employment-based coverage, or are policyholders of an employment-based plan. We use differences-in-differences (DD) and differences-indifferences-in-differences (DDD) estimators to evaluate the differential effects of alternative reform measures on high and low risk workers. We generally find little effect of reform on offer rates and find that, in states with the most stringent reform, employment-based coverage and policyholder rates increased for high risk workers relative to low risk workers. Our results also indicate that the effects of reform varied significantly by the extent to which states adopted guaranteed issue. D 2002 Elsevier B.V. All rights reserved. JEL classification: I10; I18 Keywords: Health insurance; Small group market reform

1. Introduction Improving the availability and affordability of health insurance to employees in small firms has been a focal point of efforts to reform the employment-related health insurance system and to reduce the number of uninsured Americans. This focus on the small group * Corresponding author. Tel.: +1-301-594-2059; fax: +1-301-594-2166 E-mail address: [email protected] (B. Steinberg Schone). 0047-2727/$ - see front matter D 2002 Elsevier B.V. All rights reserved. doi:10.1016/S0047-2727(02)00133-0

238

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

market is a consequence of the low insurance coverage rates of employees in small firms. It also reflects a concern that market failure—driven by insurer issuance, underwriting, and rating practices—has limited the availability of insurance coverage and made the costs of coverage excessive even for small groups of standard health risks. In the early 1990s, over 40 states passed legislation designed to limit a number of exclusionary practices by insurers in the small group market. Included in such legislation were measures to guarantee the issue and renewal of health insurance policies to small firms, to regulate premiums and their rate of increase, to constrain the time limits associated with preexisting condition exclusions, and to ensure the continuity of coverage for small firm employees and their families. Although such measures were intended to assure access to coverage in the small group market and to control its price, other changes that occurred during this period, such as increased employee premium sharing and stagnant real wages, may have constrained the ability of some workers to take advantage of coverage made available through small group reform. In this paper, we assess the degree to which small group market reform has affected employee health insurance coverage. We examine the effect of reform on three key aspects of insurance coverage: (1) the likelihood that a worker is offered coverage; (2) the likelihood that a worker is covered by employment-based insurance; and (3) the probability that a worker is a policyholder of such coverage. Since the expected impacts of reform are likely to be different depending on the health risk of individuals, we focus on differential effects of reform by health status. Our analysis, which uses both differences-indifferences (DD) and differences-in-differences-in-differences (DDD) estimators, exploits the variation in reform provisions across states, over time, and by firm size (in the DDD specification) to isolate the impact of reform.

2. The market for small group health insurance and the role of market reform Provision of health insurance through the workplace is typically based upon an evaluation of the health care experience and risk profile of well-defined employment-related groups. When such groups are large, the risk of unanticipated and expensive medical care events can be spread over a sizeable, and typically stable, risk pool. As a result, per worker administrative costs are low, and the availability of coverage and associated experiencerated premiums generally display little year-to-year variation. Thus, employees of large firms can usually be assured of obtaining offers of coverage. Moreover, they generally face premiums that reflect a reliable and fair actuarial assessment of their group’s health care experience and encounter little difficulty in renewing coverage in subsequent periods. In contrast, employees of small firms face far greater uncertainty regarding health benefits. Small groups may be unable to establish a sufficiently large and stable risk pool to absorb the costs of expensive, low probability medical events and are less able to take advantage of scale economies in the administration of health insurance benefits. These considerations, as well as concerns regarding adverse risk selection (i.e., when small firms seek coverage they do so because an employee or dependent is in ill health), have led insurers to offer coverage to small employers at premiums that exceed those charged to large firm employees with comparable risk profiles.

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

239

The underwriting, insurance, renewal, and rating practices of private insurers in the small group market have also contributed to a perception that insurers engage in exclusionary practices that at best yield favorable risk selection, and at worst result in market failure.1 For example, small firms may be subject to the kinds of risk selection and medical underwriting procedures that are common in the market for individual coverage— the use of questionnaires and medical exams to determine acceptable risks, exclusions from coverage altogether, or whether an excessive premium should be charged. In fact, insurers in the small group market have been criticized for engaging in ‘cherry picking’ or ‘cream skimming’ by selecting the good risks among small firms. They have also been accused of redlining firms and employees with particular characteristics (e.g., those in specific industries or occupations) and making liberal and aggressive use of pre-existing condition clauses. Finally, the use of durational rating, whereby discounted premiums are initially offered and then observed to increase dramatically over time or with any unfavorable claims experience, and tier rating, in which insurers reclassify groups with unexpected medical care costs to a higher rating class, have led to concerns about the rating practices used by insurers in the small group market. 2.1. The content of small group market reform A variety of legislative policies were enacted in the early 1990s to reform the small group health insurance market.2 Although the policies varied substantially across states, all states that adopted reforms included some type of restriction on premiums that was intended to decrease premium variability across insurance groups. States generally imposed one of two types of premium restrictions: (1) rating bands, which require premiums to be within a certain prescribed range (e.g., 20 –33%) of an average premium based on experience with a particular class or ‘book’ of business (such as small businesses);3 or (2) some form of community rating—either pure community rating, where all enrolled groups within an insurance pool face the same premium regardless of their perceived health risk, or adjusted or modified community rating, which allows for premium variation within a specific insurance pool according to a few specific group or enrollee characteristics (such as age, gender, industry) that may be associated with differential health risk. In addition, all states adopted some combination of ‘access’ reforms that were designed to improve coverage in the small group market.4 These measures typically included (1) Guaranteed issue, which requires all insurers who provide health insurance in the small

1

The discussion that follows is drawn from the following sources: Hall (1992, 1994); Zellers et al. (1992); Leibowitz et al. (1992); US General Accounting Office (1992), and Blumberg and Nichols (1995). 2 See Hall, 1994; US General Accounting Office, 1992, 1995; Markus et al., 1995 for a discussion of these reform provisions. 3 Rating bands typically allow for premium variation across a number of factors such as industry or location. 4 In addition, a small number of states exempted small firms from mandated benefit requirements through legislation permitting the sale of ‘bare bones’ health plans, prohibited insurers from redlining firms in specific industries, and established health insurance purchasing cooperatives so that small firms could band together and obtain health insurance plans similar to those available to larger firms.

240

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

group market to make coverage available to any small group that applies, regardless of the group’s health status or claims experience; (2) Guaranteed renewal, which requires insurers to renew the coverage of all enrolled groups; (3) Limitations on pre-existing condition exclusions, whereby insurers can no longer exclude employees or their dependents from coverage because of a particular health condition nor can they impose unusually long waiting periods before the onset of coverage;5 and (4) Portability of coverage provisions, which ensure that enrolled workers can, without being subject to a new exclusion period, obtain new coverage when they change jobs or when they change insurers within the same workplace. Table 1 displays the various combinations of reform measures adopted by states since 1987 and through the middle of 1996.6 The table reveals variation in the number of states with specific types and combinations of reform provisions. Three states (Alabama, Michigan, and Pennsylvania) were without any small group reform legislation.

3. A conceptual and empirical framework for analyzing the effects of reform A priori, it is difficult to predict theoretically the impact of small group reform on insurance coverage. Ultimately, the overall effects of reform will depend on the behavior of individuals who initially lack coverage relative to workers who currently have coverage. For example, rating restrictions (i.e., rate bands or community rating) that condense premium variation across workers with different risk profiles will produce subsidies from relatively healthy to less healthy workers. Among currently insured workers, we might observe some healthy individuals dropping coverage because insurance is not valued at its additional cost. In addition, some unhealthy workers, who faced prohibitively high premiums before the reforms and lacked coverage, may now find that the cost of insurance has dropped sufficiently to induce them to obtain coverage. If reform attracts more high risk workers to the small group market relative to the number of low risk workers that leave, then premiums will also be expected to rise. Thus, the effects of rating restrictions on insurance coverage and premiums will depend on the proportion of healthy workers who drop coverage relative to unhealthy workers who gain coverage. The stricter the restrictions on premium variation, the larger the potential subsidies from low risks to high risks (and the less attractive insurance will be to low risk individuals).7 Similarly, the impact of reforms designed to affect access to health insurance (i.e., guaranteed issue, guaranteed renewal, portability, and limits on pre-existing condition 5

Typical waiting periods are 6 or 12 months. A variety of sources provide information on small group reform legislation (e.g., US General Accounting Office, 1995; the George Washington Intergovernmental Health Policy Project, 1995, 1994a,b; and the Urban Institute’s New Federalism database). For the purpose of our analysis, we primarily rely on information from Simon (1998) and the Urban Institute’s New Federalism project. 7 Contrary to the discussion above, Hall (2000/1) argues that reform may not have significantly lowered premiums for high risk individuals, especially in states with relatively weak rating restrictions. He notes that insurers took advantage of allowable flexibility in rating restrictions to engage in risk selection under small group reform. 6

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

241

Table 1 Frequency distribution of reform measures adopted by states

1 2 3 4 5 6 6A 6B 7 7A 7B

Guaranteed issue

Guaranteed renewal

Portability

Pre-existing conditions

Rating restrictions

No No No No No Yes 1 Plan >1 Plan Yes 1 Plan >1 Plan

No No Yes Yes Yes Yes Yes

No No No Yes Yes Yes Yes

No Yes No Yes Yes Yes Yes

None Relatively Relatively Relatively Tight Relatively Relatively

Yes Yes

Yes Yes

Yes Yes

Tight Tight

weak weak weak weak weak

No. of States (%)

Sample size

3 1 1 5 1 21 2 19 16 3 13

2070 741 161 1548 180 6674 1024 5650 8831 1014 7817

(6.3) (2.1) (2.1) (10.4) (2.1) (43.8) (4.2) (39.6) (33.3) (6.3) (27.1)

Three states (Hawaii, Washington and the District of Columbia) are excluded from the analysis (see text for explanation). See the text for a complete definition of the reform measures. Sample size refers to the number of individual workers in each type of state that are not in medium-sized firms. Source: Round 1 data from the 1996 and 1997 Medical Expenditure Panel Survey (MEPS) and 1987 National Medical Expenditure Survey (NMES). See Footnote 6 of the text for reform sources.

restrictions) is not clear.8 Improved access may raise the value of health insurance since workers may feel more confident that they will keep their coverage in the future. The ultimate success of access measures in improving coverage depends on the effect of these reform measures on premiums.9 If improved access causes relatively unhealthy workers to obtain coverage, then premiums may rise on average. As a consequence, there may be some healthy workers who decide to forego coverage. Finally, the presence and extent of guaranteed issue, in particular, may be important for determining the overall effect of reform on coverage. Since guaranteed issue may have a greater impact on the likelihood that relatively unhealthy people will obtain coverage, we might observe larger increases in premiums in states with such a provision. Thus, we might expect stronger positive effects of reform on coverage for relatively unhealthy workers and stronger negative effects for low risk workers in states with guaranteed issue. The nature of guaranteed issue also varies across states with five of the 37 states with such a provision only extending it to a single plan. In these states, insurers may have an incentive to provide guaranteed issue on a relatively unattractive plan to avoid high risk

8 Analyzing the impact of reform is also complicated by the precise nature of the policies enacted. Even states that adopted policies with similar components may have substantially different reform packages when the details of the legislation are considered. 9 Simon (2000) finds that extensive small group reform (e.g., guaranteed issue and renewal, rating reform, and pre-existing condition constraints) was associated with an increase in health insurance premiums for small firms of between 4 and 6%. However, little evidence is available regarding the differential effects of reform on premiums by risk level. For example, Simon’s analysis does not distinguish between tight and weak rating restriction states.

242

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

individuals. If insurers are successful in avoiding high risks, then the impact of reform may be weaker in states without comprehensive guaranteed issue.10 3.1. Empirical implementation With a few notable exceptions, most studies of small group reform have identified its effects by comparing observations that were affected by reform with those that were unaffected at a single point in time.11 In general, these studies find mixed but not especially pronounced effects of reform on employer offer rates and employee coverage rates. However, an important concern with this approach is that other contemporaneous differences across observations that could affect the outcomes of interest may be attributed to insurance market reforms. In response to such concerns, several papers have attempted to identify reform effects by controlling for unobservable factors using DDD estimators to construct a variety of comparisons. Buchmueller and DiNardo (2002) consider whether implementation of community rating in New York state led to an adverse selection death spiral and find no impact of reform on coverage rates of employees in small firms. Simon (1999) finds a modest and negative overall effect of reform on the likelihood that an individual holds employment-based coverage; however, she finds a larger negative effect for single males less than 36 years of age, who presumably have low expected health care costs. Our work follows in the spirit of research that has used a DDD approach, but we employ different data, different insurance outcome measures, and somewhat different classifications of the reform measures. Our reliance on data from the Medical Expenditure Panel Survey (MEPS), in particular, provides much more precise measures of employee health risk, which enables us to test directly the effects of reform by individuals’ health riskiness. Specifically, we use detailed health conditions and health care expenditure information in MEPS to construct expected health expenditures for each worker and associated dependents. We then assess health risk by evaluating the position of each worker and family in the expected expenditure distribution. We use DD and DDD econometric tests to investigate the effects of reform. The DD specifications use variation in reforms across states and changes in the insurance outcomes over time to identify the effects of the policy among employees in small firms. The DDD models use large firm employees as an additional control group under the assumption that they are unaffected by the reforms. Further, since the effects of reform are likely to vary by workers’ health status and by the intensity of reform efforts, we model several variants of reform policy and interact them with health status. Table 2 provides classifications of the reform measures used in our analysis. In addition to extensive control variables, our econometric model includes indicators for time period (POST REFORM = 1 in the post-intervention time period). The DDD

10

We thank an anonymous reviewer for raising this issue. For example, see the studies by Jensen and Morrisey (1997), Hing and Jensen (1999), Sloan and Conover (1998) and Zuckerman and Rajan (1999). Studies of the impact of reform in specific states include the review by Nichols et al. (1997), Buchmueller and Jensen (1997), and the case studies of Hall (1999). 11

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

243

Table 2 Specifications of reform variables for empirical tests Variables

Row (Category) from Table 1

Main specification Measures reform intensity

No reform Moderate reform Stringent reform (Weak Rating) Stringent reform (Tight Rating)

1 2–5 6 7

Alternative 1 Based on rating restriction

No reform Weak rating restrictions Tight rating restrictions

1 2 – 4, 6 5, 7

Alternative 2 Based on presence of guaranteed issue & pricing

No guaranteed issue Guaranteed issue with weak rating restrictions Guaranteed issue with tight rating restrictions

1–5 6 7

Alternative 3 Based on extent of guaranteed issue & pricing

No guaranteed issue or guaranteed issue of one plan Guaranteed issue with weak rating restrictions Guaranteed issue with tight rating restrictions

1 – 5, 6A, 7A 6B 7B

Tight rating restrictions are designated for those states identified in the Urban Institute’s New Federalism database as having either community rating or very tight rate bands. Relatively weak rating restrictions refer to all other rate band states.

specifications also include an additional control for firm size (SMALL = 1 for working in a small firm). To capture reform effects, we allow for k different reform provisions, REFORMk, to denote reform intensity in different states (no reform or the least extensive reform serves as the reference category in all specifications). These variables are then interacted with time (in the DD and DDD models); a three-level interaction of reform, time, and firm size is also included in the DDD specification. Since we are primarily interested in identifying the effects of reform on individuals by their health risk, we calculate separate effects of reform for each risk type (H = high risk; L = low risk). Our DD model (estimated only on workers in small firms) becomes: Yi ¼ Zi a þ d1  H þ þ

X X risk¼L;H

þ

d2;risk Post Reform  Risk

risk¼L;H

d3;k;risk Reformk  Risk

k

X X risk¼L;H

X

d4;k;risk Post Reform  Reformk  Risk

k

and our DDD model is: Yi ¼ Zi c þ b1  H þ þ

X

risk¼H;L

X

X X

b2;r Small  Risk þ

risk¼H;L

b3;risk;k Reformk  Risk

risk¼H;L k

b4;r Post Reform  Risk þ

X X

risk¼H;L k

b5;risk;k Reformk Post Reform  Risk

244

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

þ

X X

b6;risk;k Reformk  Small  Risk þ

risk¼H;L k

þ

X X

X

b7;r Small  Post Reform  Risk

risk¼H;L

b8;risk;k Small  Post Reform  Reformk  Risk;

risk¼H;L k

where i indexes individual, Z is a vector of control variables, and Y is the insurance outcome of interest. The d4;k;risk coefficients provide our DD estimates of reform, and the b8;k;risk coefficients provide the DDD estimates.

4. Data and methods Our data are drawn from the first rounds of the 1996 and 1997 Medical Expenditure Panel Survey (MEPS)—our post-reform sample, and Round 1 of the 1987 National Medical Expenditure Survey (NMES)—our pre-reform sample. Each of these surveys are nationally representative samples of households that provide detailed information on health care use and expenditures, health insurance coverage, health status, medical conditions, and basic socioeconomic information.12 Our samples for each time period consist of employed wage earners between ages 21 and 64.13 Estimating the effects of reform requires information on whether an individual worked in a small firm. Since household respondents in both NMES and MEPS report establishment size rather than firm size, we classified a worker as being in a small firm if the worker was employed in an establishment with fewer than 26 workers or the worker was in a firm with 26 –50 workers and the firm had no other locations.14 Our large firm category includes establishments with 50 or more workers. Given the ambiguity of measuring firm size for our purposes, 661 workers who were employed in establishments with 26– 50 workers for which the firm had more than one location were excluded from the analysis. Our overall sample consists of 20,192 workers. An important feature of our analysis is investigating the differential effects of reform by health riskiness. Unfortunately, relying on self-reported health status or health limitations is not likely to provide a sufficient measure of riskiness due to the low prevalence of poor health among workers in this age range. Instead, we take advantage of health care expenditure data from the NMES and MEPS to differentiate our sample. We use a two-part model of health care expenditures that is similar to that of Pauly and Herring (1999) and Herring (2000) to form predicted expenditures. We then define individuals as being high

12 Additional information on the MEPS surveys can be found in Cohen (1997). Information regarding the design of the 1987 NMES can be found in Edwards and Berlin (1989). 13 Workers from Hawaii, the District of Columbia and Washington were excluded from our analysis. Both Hawaii and Washington adopted legislation that went beyond small group reform (i.e., some form of employer mandate), while the District of Columbia did not adopt reform but is dominated by government workers. 14 Workers in establishments with less than one hundred workers in New Hampshire and Virginia were also classified as being in small firms since small group definitions in these states were more expansive.

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

245

risk if their predicted expenditures are in the top 25% of the expected expenditure distribution.15,16 To estimate the effects of reform, we use linear probability models corrected for heteroskedasticity to estimate the likelihood that a worker is offered insurance, the likelihood that a worker is covered by employment-based health insurance, and the likelihood that a worker is a policyholder of employment-based coverage.17 In each model we control for a rich set of socioeconomic variables that includes age, race/ethnicity, family size, education, an indicator of fair or poor health status, marital status, and sex. Our regressions also control for geographic differences (an indicator for residing in a metropolitan area and a measure of the local unemployment rate), employment characteristics (industry and whether the worker is in a union) and state fixed effects.

5. Results Table 3 shows DD and DDD estimates of reform for a number of different specifications derived from linear probability models. As described above, we calculate separate DD and DDD estimates of reform for high and low risk workers. On net, we expect relatively beneficial effects of reform for high risk individuals and detrimental effects for low risk individuals, with the largest differences observed in the states with the most stringent reform (i.e., states that should have experienced the strongest compression of the premium distribution). Consistent with other researchers (Buchmueller and DiNardo, 2002; Simon, 1999), our DDD estimates use large firm employees as an additional control group to isolate the effects of reform since they may eliminate any state-specific (and firm-size invariant) differences in changes in offer rates and coverage over time. However, large firm employees may represent an inadequate control group if there are firm-size specific and state-specific differences in rates of growth of offer and coverage rates. An additional maintained assumption in the DDD specifications is that large firm employees are unaffected by reform. This assumption might be violated if couples (with one worker in a small firm and another in a large firm) adjust their insurance choices due to reform or if firms change the characteristics of their insurance offers to large firm employees. Unfortunately, there are no viable alternatives to using large firm employees as an additional control.18 Thus, in addition to the DDD estimates, we also report DD estimates of reform based on the sample of small firm workers.

15

In our empirical models, the top 25% of individuals account for approximately 50% of total expenditures. Details about the estimation procedure and the full set of regression results for the expenditure models are available upon request from the authors. 17 As in work by Simon (2000), Buchmueller and DiNardo (2002), and Gruber (1994b), we focus on linear probability models because they provide a straightforward interpretation of the effects of reform, which are not sensitive to nonlinearities or other covariates. 18 It would be preferable to focus our analyses exclusively on single workers since they are most likely to be affected by reform (i.e., they cannot rely on a spouse for health insurance). Unfortunately, our sample of single workers stratified by firm size and reform type was not sufficiently large to allow such an analysis. 16

246

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

Table 3 Main effects of small group market reform, linear probability model specifications Offer rates

Employmentsponsored coverage rates

Employmentsponsored policyholder rates

Estimated coefficients (Percentage point change) Main Specification DDD 96/97*Small Firm*Moderate Reform*Low Risk 96/97*Small Firm*Moderate Reform*High Risk 96/97*Small Firm*Stringent (Weak Rating) Reform*Low Risk 96/97*Small Firm*Stringent (Weak Rating) Reform*High Risk 96/97*Small Firm*Stringent (Tight Rating) Reform*Low Risk 96/97*Small Firm*Stringent (Tight Rating) Reform*High Risk

 4.0 2.2  5.7  0.05  5.8  1.9

 4.2  6.1  5.6  12.2#  6.2 1.0

 0.3 5.2 2.2 13.9 1.4 7.4

DD (Small Firm Workers) 96/97*Moderate Reform*Low Risk 96/97*Moderate Reform*High Risk 96/97*Stringent (Weak Rating) Reform*Low Risk 96/97*Stringent (Weak Rating) Reform*High Risk 96/97*Stringent (Tight Rating) Reform*Low Risk 96/97*Stringent (Tight Rating) Reform*High Risk

 1.5 1.2  0.2 0.005 0.5  0.9

 3.0  2.2  2.9  7.4#  0.9 5.1

6.1 19.2  1.0 16.4 3.3 15.2*

Alternative 1 DDD 96/97*Small Firm*Weak Rating*Low Risk 96/97*Small Firm*Weak Rating*High Risk 96/97*Small Firm*Tight Rating*Low Risk 96/97*Small Firm*Tight Rating*High Risk

 5.5 0.8  5.4  1.7

 5.4  10.9#  6.0 1.0

1.4 11.1 1.5 8.8

DD (Small Firm Workers) 96/97*Weak Rating*Low Risk 96/97*Weak Rating*High Risk 96/97*Tight Rating*Low Risk 96/97*Tight Rating*High Risk

 0.9 0.3 0.8  0.4

 3.1  6.2#  0.7 5.0

0.9 16.6* 3.6 16.5*

 3.4  1.0  3.4  2.9

 3.1  8.2#  3.8 5.0

2.6 11.1 1.8 4.6

0.6  0.9 1.3  1.9

 1.4  5.9# 0.6 6.7

 4.0 6.4 0.3 5.2

 5.5

 4.2

 1.3

Alternative 2 DDD 96/97*Small Firm*Guaranteed 96/97*Small Firm*Guaranteed 96/97*Small Firm*Guaranteed 96/97*Small Firm*Guaranteed

Issue Issue Issue Issue

(Weak Rating) Reform*Low Risk (Weak Rating) Reform*High Risk (Tight Rating) Reform*LowRisk (Tight Rating) Reform*High Risk

DD (Small Firm Workers) 96/97*Guaranteed Issue (Weak Rating) Reform*Low Risk 96/97*Guaranteed Issue (Weak Rating) Reform*High Risk 96/97*Guaranteed Issue (Tight Rating) Reform*Low Risk 96/97*Guaranteed Issue (Tight Rating) Reform*High Risk Alternative 3 DDD 96/97*Small Firm*>1 Plan Guaranteed Issue (Weak Rating) Reform*LowRisk 96/97*Small Firm*>1 Plan Guaranteed Issue (Weak Rating) Reform*High Risk

 11.3

 15.2**#

4.5

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

247

Table 3 (continued) Offer rates

Employmentsponsored coverage rates

Employmentsponsored policyholder rates

Estimated coefficients (Percentage point change) Alternative 3 DDD 96/97*Small Firm*>1 Plan Guaranteed Issue (Tight Rating) Reform*Low Risk 96/97*Small Firm*>1 Plan Guaranteed Issue (Tight Rating) Reform*High Risk DD (Small Firm Workers) 96/97*>1 Plan Guaranteed 96/97*>1 Plan Guaranteed 96/97*>1 Plan Guaranteed 96/97*>1 Plan Guaranteed

Issue Issue Issue Issue

(Weak Rating) Reform*Low Risk (Weak Rating) Reform*High Risk (Tight Rating) Reform*Low Risk (Tight Rating) Reform*High Risk

 2.4

 3.3

 5.4

 14.4*

 2.3

1.4

 2.1  8.3 0.6  11.5

 1.6  13.9**# 1.1  1.9

 6.9 4.5  4.9 7.2

DDD estimates are based on linear probability models on small and large firm workers and control for demographic, employment characteristics, state fixed effects. DD estimates are for small firm workers only. Coefficients are statistically significant at: **P < 0.05; *P < 0.10; #reflects statistically significant differences ( P < 0.10) between the marked category and high risk individuals in stringent reform states. Source: Round 1 data from the 1996 and 1997 Medical Expenditure Panel Survey (MEPS) and 1987 National Medical Expenditure Survey (NMES).

The top panel of Table 3 shows the effects of reform for high and low risk workers for our main specification, which distinguishes between workers in states that did not adopt any reform (the reference category), workers in states that did not adopt guaranteed issue (moderate reform), workers in states that adopted guaranteed issue with weak rating restrictions (stringent reform with weak rating), and workers in states with the most stringent reform including guaranteed issue and tight rating restrictions.19 In both the DD and DDD specifications, we find mostly small (and in several cases, negative) point estimates of the effect of reform on offers, although none are statistically significant. However, it is interesting to note that we observe small positive (or less negative) changes in offer rates for high risk workers relative to low risk workers (except for workers in stringent reform states with tight rating restrictions in the DD specification). Although not statistically significant, these results are consistent with the predictions of theory that suggest that reform may have aided high risk workers at the expense of low risk workers. The results from the main specification also show that employment-sponsored insurance coverage rates fell for most groups as a result of reform. Although the point estimates are statistically insignificant, they do tend to be somewhat larger in the DDD specifications, suggesting that large firm employees in reform states did not face similar declines in coverage. The results also indicate that coverage improved for high risk workers relative to low risk workers in the states with stringent reform and tight rating restrictions, although the differences are not statistically significant. In addition, we 19 Table 3 reports the key DDD and DD coefficients from the regressions (the d4;k;risk and b8;k;risk coefficients from the models described above). The coefficient estimates for all variables in the model are available upon request from the authors.

248

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

observe a small statistically significant improvement in coverage for high risk workers in stringent reform states with tight rating restrictions (1.0 percentage point improvement in coverage in the DDD specification) relative to high risk workers in stringent reform states with weak rating restrictions (who experienced a 12.2 percentage point decline in coverage in the DDD specification). These findings are consistent with the notion that reform should help high risk workers more in states with tight rating restrictions, since tight rating restrictions should lead to relatively lower premiums for high risk individuals. However, it is unclear why high risks in states with weak rating restrictions experienced declines in coverage.20 Our multivariate results also indicate that for those offered health insurance coverage, reform generally had a positive effect on the likelihood of being a policyholder. In both the DDD and DD specifications, we find that high risk workers in states with the most stringent reform had increases in the likelihood of being a policyholder (7.4 percentage points in the DDD specification and 15.2 percentage points in the DD specification); the point estimates for low risk workers were small but positive (1.4 percentage points in the DDD specification and 3.3 percentage points in the DD specification). The DD estimates also indicate sizeable increases in the likelihood of being a policyholder for high risk workers in all states with reform, including those states without guaranteed issue. Since guaranteed issue is likely to be the most expensive aspect of reform, we would expect premiums to be less sensitive to reform in states that did not adopt it. The fact that the DDD estimate is so much smaller suggests that this estimate may be reflecting other trends in these states over the time period. Finally, although policyholder rates tended to improve as a result of reform, this did not generally translate into improvements in employmentsponsored insurance coverage rates, except for high risks in stringent reform states. 5.1. Other specifications To assess the robustness of our results, the remaining panels of Table 3 show the DDD and DD estimates of the effects of reform using slightly different reform classifications. Given the potential importance of guaranteed issue and rating restriction provisions, we consider different representations of reform along these dimensions. The second panel in Table 3 (Alternative 1) focuses solely on differences in rating restrictions (no reform, weak and tight rating restrictions). As before, we find small and statistically insignificant effects of reform on offer rates and generally negative and statistically insignificant effects on coverage rates, with low risk workers in tight rating restriction states facing a decline in coverage relative to high risk workers in the same states. High risk workers in states with weak rating restrictions continue to fare worse than their high risk counterparts in those states with tight rating restrictions. In terms of policyholder rates, we observe a large and statistically significant positive effect on high risk workers in states with tight rating restrictions in the DD specification; the DDD estimates are somewhat smaller and not statistically significant. Consistent with the main specification, our results continue to show improvements in policyholder rates for high risks relative to low risks. 20 Consistent with Hall’s (2000/2001) observations, these results suggest that insurers may have used the allowable premium flexibility in states with weak rating restrictions to engage in risk selection.

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

249

The final two specifications classify reform by the presence of (and extent of) guaranteed issue and the degree of rating restrictions. In Alternative 2 we contrast states according to the presence of guaranteed issue and the degree of rating restrictions. We group states without guaranteed issue along with states that did not adopt any reform under the assumption that guaranteed issue might have the largest effect on behavior. As in our other specifications, we observe generally small and statistically insignificant effects of reform on offer rates. We continue to find that high risk workers in states with guaranteed issue and tight rating restrictions were more likely to have coverage after reform relative to high risk workers in states with guaranteed issue and weak rating restrictions (an increase of 5.0 percentage points for the former relative to a decrease of 8.2 percentage points for the latter in the DDD specification, P < 0.10). In addition, coverage rates for high risk workers improved relative to low risk workers in states with guaranteed issue and tight rating restrictions, although these differences are not statistically significant. Finally, point estimates also indicate that policyholder rates for high risk workers improved in all states relative to their low risk counterparts but these findings are not statistically different. If guaranteed issue is an important component of reform, then the extent to which guaranteed issue is adopted may matter for the ultimate effectiveness of small group reform proposals. Several states that adopted reform only required insurers to guarantee issue for one plan. Therefore, it is possible that insurers behaved strategically by offering guaranteed issue on a relatively unattractive plan. Thus, the potential benefits of guaranteed issue may only apply to workers in states that adopted a more expansive provision. Alternative 3 allows us to evaluate this possibility. In this case, we contrast workers in states with guaranteed issue of more than one plan by the level of rating restrictions relative to all other workers (those in states without reform, without guaranteed issue, or with guaranteed issue of a single plan). As in earlier specifications, we generally find statistically insignificant and negative effects of reform on offer rates, although the point estimates are larger in magnitude. Unlike the other specifications, high risk workers in states with guaranteed issue of at least two plans and tight rating restrictions had a statistically significant 14.4 percentage point decline ( P < 0.10) in offers in the DDD specification. In fact, the results indicate that offers fell for high risks relative to low risks. Such findings suggest that, in the presence of extensive guaranteed issue, insurers may engage in risk selection strategies to limit high risk employees’ access to coverage, especially when they are unable to limit guaranteed issue to one relatively unattractive plan. Such behavior may have lead some firms to stop offering coverage. With respect to coverage, we observe large negative effects of extensive guaranteed issue for high risk workers in states that have weak rating restrictions. Although theory suggests that high risk workers in states with tight rating restrictions should fare better than high risk workers in states with weak rating restrictions, it is unclear why reform would have such detrimental effects on coverage for high risk workers in states with weak rating restrictions. However, this finding is consistent with the large declines in offers noted above and may result from risk selection on the part of insurers (Hall, 2000/2001). Finally, although statistically insignificant, in states with the strictest reform, there was also a decline in policyholder rates (among those offered insurance) for low risk workers (  5.4 and  4.9 percentage points in the DDD and DD specifications, respectively) relative to their high risk counterparts, who obtained gains in policyholder status (1.4 percentage

250

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

points and 7.2 percentage points in the DDD and DD specifications, respectively). Similar patterns also exist for workers in states with weak rating restrictions. In sum, using a variety of reform specifications, we find no evidence that, overall, small group reform improved insurance offer rates. Moreover, our empirical work does not provide much evidence that reform generally improved employment-based coverage for small firm employees. Although our estimates lack precision, our results do suggest that stringent reform (e.g., guaranteed issue, renewal, and portability; limits on preexisting conditions and tight rating restrictions) was generally associated with small increases in coverage rates and larger gains in policyholder rates for high risk employees relative to low risk employees. These findings are consistent with our expectations regarding the differential effect of reform on workers by risk status that arise from insurance market theory. In all specifications, we also find evidence that high risks in states with the strongest reform measures gained relative to high risk workers in states with less strict reforms. Finally, our specifications focusing on the extent of guaranteed issue reveal that offer rates declined substantially for high risk workers in states where guaranteed issue was extended to more than one plan, resulting in significant declines in coverage. These results are consistent with the notion that insurers may have engaged in more aggressive risk selection when guaranteed issue was extended to multiple plans. 5.2. Other considerations While our analysis has emphasized the impact of small group market reform on employee access to coverage and employment-based coverage rates, the period we study was also marked by other important changes in the health care system that may have also affected the health insurance outcomes of interest. To the extent that other changes were correlated with small group reform, our estimated effects of reform could be biased. We consider two such changes—Medicaid expansions and changes in self insurance in small firms—and conclude that it is unlikely that these particular factors are confounding our estimates. During the late 1980s and early 1990s, there was significant expansion in Medicaid eligibility due to both federal mandates and state options that relaxed income and family structure standards, extended coverage to targeted groups of low income children and pregnant women, and resulted in considerable variation in eligibility requirements across states. In light of these changes, a substantial literature has developed that suggests that the expansions may have affected individuals’ decisions to obtain employment-based coverage and firms’ decisions to offer such coverage.21 From the perspective of our analysis, if the Medicaid expansions were associated with the presence and content of small group reform across states, then enhanced Medicaid access for small group employees could confound our findings regarding the reform’s impact on offer and coverage rates. To assess the role played by the Medicaid expansions, we examined whether a systematic relationship exists between changes in state-specific measures of 21 See the literature on Medicaid crowd out of private insurance (e.g., Cutler and Gruber, 1996; Dubay and Kenney, 1996; Yazici and Kaestner, 2000 and Shore-Shepard et al., 2000).

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

251

expected Medicaid eligibility and the components of small group market reform used in our analysis. To evaluate this relationship, we used information on predicted Medicaid eligibility from two sources: (1) tabulations by Gruber (2000, Table 7) of the proportion of children ages 15 or younger in each state likely to be eligible for Medicaid in 1983 and 1996;22 and (2) 1987 and 1996 estimates of each state’s expected proportion of eligible children derived by using a standard population of children less than age 18 from the Medical Expenditure Panel Survey and state Medicaid eligibility criteria in both years.23 Using each set of data, we regressed the change in eligibility (absolute and percentage changes) on the degree of small group reform adopted in states to determine if there was any evidence of an association. We found no systematic relationship between changes in state-specific expected Medicaid eligibility and the classes of state small group reform used in our study. On this basis, it is unlikely that our estimates of the relationship between small group reform and insurance offer and coverage rates will be biased due to the impact of expanded Medicaid eligibility over our study period. A second issue of concern is the prevalence of self insurance in small firms. Since selfinsured firms are exempt from state health insurance regulation, failure to control for any changes in self insurance in small firms (especially differential changes in self insurance across states by the extent of reform adopted) over our study period could affect our estimates of the impact of small group reform. Analyzing the effect of reform on self insurance with our household data (as well as household data used in other studies of small group reform) is not possible since the data are not able to identify workers employed by self-insured small firms. However, other employer-based data suggest that (1) the prevalence of self-insured small firms declined over our study period and that (2) small group reform may not have had an impact on the decision by small firms to self insure. As regards the former, Marquis and Long (1999) use data from the Robert Wood Johnson Employer Health Insurance Surveys for 1993 and 1997 to track trends in self insurance by firm size. They find that between 1993 and 1997, the percent of self-insured establishments in firms with fewer than 100 employees actually declined from 9 to 3% and the percent of enrollees in self-insured plans in such establishments declined from 11 to 4%. Park (2000), using data from the 1993 National Employer Health Insurance Survey, reports that only 4.5% of establishments in firms with less than 50 employees self insured in 1993 and that only 5.4% of employees in establishments with firm size less than 50 employees self insure. Thus, self insurance appears to play a very small role among small firms. Marquis and Long also find a weak association between the presence of self-insured small firms and small group reform: only 3 –4% of establishments with firm size less than 50 employees in states with small group reform self insure. In addition, multivariate work by Park reveals no association between small group reform and the likelihood that establishments in firms with 50 or fewer employees would self insure. Thus, it would appear that the presence of self insurance among small firms is not especially pronounced and that small group reform had little if any impact on the decision by such firms to self insure. 22 Gruber’s estimates are derived from CPS data on children in each state in each year and state Medicaid eligibility criteria in each year. 23 We thank Tom Selden of the Agency for Healthcare Research and Quality for providing these data.

252

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

6. Conclusions Reform of the small group health insurance market represents an incremental regulatory approach designed to address a number of perceived exclusionary and selective pricing practices of insurers. At the same time, such reform has sought to target a group of employees that are most likely to lack health care coverage. Despite the intent of small group insurance reform to lower premiums and improve access to coverage for small firm employees, the ultimate effects of reform may vary across individuals depending upon whether they are at high risk or low risk to incur above average health care expenses. The existing body of research regarding the impact of reform on coverage and premium levels (e.g., Jensen and Morrisey, 1997; Buchmueller and DiNardo, 2002; Simon, 1999, 2000), provides little evidence that reform has precipitated an adverse selection death spiral in the small group market. Although our point estimates suggest somewhat more sensitivity of insurance coverage to the features of small group reform than others have found, we still find the overall effects of reform to be relatively modest. However, our findings, along with those of Simon (1999) also suggest that the modest effects of reform found by other researchers (Nichols et al., 1997; Buchmueller and Jensen, 1997; Hall, 1999; Jensen and Morrisey, 1997; Hing and Jensen, 1999; Sloan and Conover, 1998; Zuckerman and Rajan, 1999) may actually mask the differential outcomes experienced by specific population subgroups. Taken together, both studies suggest the importance of evaluating the impact of reform across groups of varying health risk. Moreover, our results regarding the presence of guaranteed issue of more than one health plan suggest that efforts by policymakers to ensure access to coverage for high risk employees may require additional initiatives to ensure that insurers are not engaging in risk selection. Finally, it is important to note that many of the reforms were implemented in the early 1990s. As reforms remain in existence for longer periods, and as more data become available, researchers will have the opportunity to examine adjustments to small group reform beyond the initial responses of employers and employees. For example, little is known about the effects of reform on two-worker families who may adjust their insurance choices. Additional data may also permit an evaluation of the contribution of individual reform components so that the effectiveness of specific types of reform packages can be better assessed. In the meantime, it is clear from the existing body of research that the beneficial effects of reform have been relatively modest and that small group reform may have had a differential impact on low risk and high risk workers.

Acknowledgements We thank Tom Buchmueller, Bob Haveman, Len Nichols, James Poterba, seminar participants at the Agency for Healthcare Research and Quality, the American Economic Association meetings, the University of North Carolina at Chapel Hill, the International Health Economics Association, and the Southern Economic Association meetings and two anonymous referees for helpful comments. We are also grateful to Sam Zuvekas for compiling the state regulation data, to Kosali Simon for sharing her information on state

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

253

reforms, and to Chao-Sung Yu for assistance in constructing the analysis file. The views expressed in this paper are those of the authors and no official endorsement by AHRQ or the Department of Health and Human Services is intended or should be inferred.

References Buchmueller, T., DiNardo, J., 2002. Did community rating induce an adverse selection death spiral? Evidence from New York, Pennsylvania and Connecticut. American Economic Review 92 (1), 280 – 294. Buchmueller, T.G., Jensen, G.A., 1997. Small group reform in a competitive managed care market: the case of California, 1993 to 1995. Inquiry 34, 249 – 263. Blumberg, L.J., Nichols, L.M., 1995. Health Insurance Market Reforms: What They Can and Cannot Do. Urban Institute Press: Washington, DC. Cohen, J., 1997. Design and Methods of the Medical Expenditure Panel Survey Household Component. Rockville, MD: Agency for Health Care Policy and Research. MEPS Methodology Report, No. 1. AHCPR Pub. No. 97-0026. Cutler, D., Gruber, J., 1996. Does public insurance crowd out private insurance? Quarterly Journal of Economics 111 (2), 391 – 430. Dubay, L.C., Kenney, G.M., 1996. The effects of medicaid expansions on insurance coverage of children. The Future of Children 6 (1), 152 – 161. Edwards, W.S., Berlin, M., 1989. Questionnaires and Data Collection Methods for the Household Survey and the Survey of American Indians and Alaska Natives. DHHS Publication No. (PHS) 89-3450. National Medical Expenditure Survey Methods 2, National Center for Health Services Research and Health Care Technology Assessment. Public Health Service, Rockville, MD. Gruber, J., 1994b. State-mandated benefits and employer-provided health insurance. Journal of Public Economics 55, 433 – 464. Gruber, J., 2000. Medicaid. NBER Working Paper W7829. August. Hall, M.A., 1992. The political economics of health insurance market reform. Health Affairs 11 (2), 7 – 27. Hall, M.A., 1994. Reforming Private Health Insurance. American Enterprise Institute, Washington, DC. Hall, M.A., 1999. An Evaluation of Health Insurance Reforms: Summary of Findings. http://www.phs.bgsm.edu/ insure/summary.html. Hall, M.A., 2000/1. The structure and enforcement of health insurance rating reforms. Inquiry 37, 376 – 388. Herring, B.J., 2000. Access to Free Care for the Uninsured and Its Effect on Private Health Insurance Coverage. University of Pennsylvania Doctoral Dissertation. Hing, E., Jensen, G., 1999. Health Insurance Portability and Accountability Act of 1996: Lessons from the States. Unpublished manuscript. Intergovernmental Health Policy Project, 1995. Fifty State Profile: Health Care Reform, 1995. George Washington University, Washington, DC. Jensen, G.A., Morrisey, M.A., 1997. A Small Group Reform and Insurance Provision by Small Firm, 1989 – 1995. Unpublished manuscript. Leibowitz, A., Damberg, C., Eyre, K., 1992. In: Multiple Employer Arrangements, in US Department of Labor, Pension and Benefits Administration, Health Benefits in the Workforce. US Government Printing Office, Washington, DC, pp. 37 – 42. Markus, A.R., Ladenheim, K.E., Atchison, L., 1995. In: Small Group Market Reforms: States’ Experiences. The Commonwealth Fund, New York. Marquis, M., Long, S.H., 1999. Recent trends in self-insured employer health plans. Health Affairs 18 (3), 161 – 166. Nichols, L.M., Blumberg, L.J., Acs, G., Uccello, C.E., Marsteller, J.A., 1997. In: Small Employers: Their Diversity and Health Insurance. The Urban Institute Press, Washington, DC. Park, C.H., 2000. Prevalence of self-insured health benefits. Medical Care Research and Review 57 (3), 340 – 360. Pauly, M., Herring, B., 1999. Pooling Health Risks. The American Enterprise Institute Press, Washington, DC.

254

A.C. Monheit, B. Steinberg Schone / Journal of Public Economics 88 (2003) 237–254

Shore-Shepard, L.D., Buchmueller, T.C., Jensen, G.A., 2000. Medicaid and the crowding out of private insurance: A re-examination using firm level data. Journal of Health Economics 19 (1), 61 – 91. Simon, K., 2000. The Effect of State Insurance Regulation on Price and Availability of Health Benefits in Small Firms. Econometrics and Economic Theory Working Paper No. 2001, Department of Economics, Michigan State University. Simon, K., 1999. Did Small-Group Health Insurance Reform Work? Evidence from the March Current Population Survey 1992 – 1997, University of Maryland working paper. Simon, K., 1998. State Profiles of Small-Group Health Insurance Reform 1990 – 1996, University of Maryland working paper. Sloan, F.A., Conover, C.J., 1998. Effects of state reforms on health insurance coverage of adults. Inquiry 35 (3), 280 – 293. United States General Accounting Office, 1992. Access to Health Insurance Regulation: State Effort to Assist Small Businesses. GAO/HRD-92-90. United States General Accounting Office, 1995. Health Insurance Regulation: Variation in Recent State Small Employer Health Insurance Reforms. GAO/HEHS-95-161FS. Yazici, E.Y., Kaestner, R., 2000. Medicaid expansions and the crowding out of private health insurance among children. Inquiry 37 (1), 23 – 32. Zellers, W.K., McLaughlin, C.G., Frick, K.D., 1992. Small business health insurance: Only the healthy need apply. Health Affairs 11 (1), 174 – 180. Zuckerman, S., Rajan, S., 1999. An alternative approach to measuring the effects of insurance market reforms. Inquiry 36 (1), 44 – 56.