Adverse selection in health insurance markets?
Evidence from state small-group health
insurance reforms
Kosali Ilayperuma Simon*
Department of Policy Analysis and Management, MVR Hall, Cornell University, Ithaca, NY 14853, United States Received 6 August 2003; received in revised form 11 June 2004; accepted 1 July 2004
Available online 28 September 2004
Abstract
The past decade witnessed major changes in state laws governing the sale of health insurance to small employers. States took measures to restrict insurers’ ability to distinguish between high and low-risk customers because of concern about the low rate of coverage among workers in small firms, the high prices in the small-group market and the absence of federal health reform. Using both individual-level and employer-level data, I test predictions about the effect of reforms on the employer-provided health insurance market. I estimate these effects for small firms and their workers using large firms and their workers in the same states, as well as large and small firms and their workers in non-reform states, as comparison groups. I find the reforms decreased the rate of employer coverage on average for workers in small firms by less than two percentage points. Within small firms, low-expenditure individuals experienced a larger decline in the rate of coverage through their employer, while the coverage rate of high-expenditure individuals rose slightly in some specifications. There is also evidence that comprehensive reforms increased premiums slightly for small employers, and that most of this increase was passed on to workers through higher employee contributions.
D2004 Elsevier B.V. All rights reserved. JEL classification:I18; I11; J32
Keywords:Adverse selection; Health insurance; Small firms
0047-2727/$ - see front matterD2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jpubeco.2004.07.003
* Tel.: +1 607 255 7103.
E-mail address:kis6@cornell.edu.
Fearing that low health insurance coverage in small firms could be partly due to experience rating and redlining,1many states introduced laws that restricted these practices in the small-group market. Inability to price and issue policies in accordance with risk could worsen informational asymmetry and resulting adverse selection relative to the unregulated market. Adverse selection is thought to generally reduce the insurance consumption of the low-risk groups, to transfer resources from the low-risk group to the high-risk group in cases where subsidized equilibria are sustained, or to result in a market failing to exist altogether. A simple model of insurance where employers buy policies on behalf of their heterogeneous workforce suggests that small group reforms may decrease coverage for the low-risk and increase coverage for the high-risk (seeSimon, 2004), while the predicted changes are ambiguous for the market on average. This comprehensive empirical analysis of state attempts to standardize the terms of insurance across different risk types suggests that reforms have not succeeded in increasing coverage in small firms as a whole. Instead, they may have inadvertently led to a small overall decrease in coverage through an increase in premiums and employee contributions. This analysis also suggests that certain low-cost populations have suffered larger declines in coverage than others.
1. Government involvement in small-group health insurance
Starting in the early 1990s, state legislators took steps tobpromote the availability of health insurance coverage to small employers regardless of their health status. . .and to improve the overall. . .efficiency of the small-group health insurance marketQ.2From 1991 to 1996, 47 states implemented some combination of the small-group reforms described below. Rating restrictions limited the insurers’ ability to use certain predictors of health care use in setting premiums, while guaranteed issue laws banned denial of policies. Some states allowed insurers to market a guaranteeddbare bonesT plan to first-time insurance buyers. Pre-existing conditions exclusion laws and portability laws improved continuity of access to health insurance while working for small firms.3
Most states followed the language of model laws published by theNational Association of Insurance Commissioners (NAIC) (1998). The area in which most variation exists is rating reforms. Most states allowed premiums to vary by certain demographic factors calleddcase characteristicsTand permitted variation around average prices through drate bandsT within which group-specific information could be used. The language of most rating statutes is not straightforward, and many allow insurance commissioners to decide
1 Redlining is the practice of systematically refusing to insure groups in certain high-risk industries or occupations.
2
This quote is taken from Section 2 of the 1992 National Association of Insurance Commissioners (NAIC) Small Employer Health Insurance Availability Model Act.
3
The regulatory information used in this analysis was gathered through a careful primary investigation of state statutes and bills checked against all available secondary data sources, and from personal communication with almost all state insurance departments (Simon, 2000). SeeHall (1999)for detailed discussions of these reforms.
on the specific restrictions upon reviewing the pricing structure submitted by insurers. The definition of a small firm usually varies from one with 2–25 employees to ones with 1–50 employees. Most states fit naturally into a three-category definition of reform. dFull reformTrefers to states with the two most binding laws, guaranteed issue and rating reform, together with the weaker portability and pre-existing conditions laws. dPartial reformT
refers to states with rating reforms that did not guarantee the issue of health insurance, and states with no issue or rating laws are called dno reformTstates.4Table 1 describes the regulatory regime in each state from 1991 to 1996.
2. Previous studies of small-group reform
The existing literature can be separated into studies using individual or employer level data. Cross-sectional studies on the effect of reforms on the employer’s decision to purchase insurance find some positive effects (Jensen and Morrisey, 1999; Hing and Table 1
Timing and nature of state reforms: 1991–1996 State Full reform Partial reform Bare bones
plan laws
State Full reform Partial reform Bare bones plan laws AK 1994–1996 MT 1994–1996 1992–1996 AL NC 1992–1996 1993–1996 AR 1992–1996 1993–1996 ND 1995–1996 1994–1992 1992–1996 AZ 1994–1996 1992–1996 NE 1995–1996 1992–1994 1992–1996 CA 1994–1996 NH 1996 1994–1995 CO 1996 1995 1992–1996 NJ 1995–1996 1992–1996 CT 1992–1996 1992–1996 NM 1996 1992–1995 1992–1996 DC NV 1992–1996 DE 1994–1996 1992–1993 1994–1996 NY 1994–1996 FL 1994–1996 1992–1993 1994–1996 OH 1993–1996 GA 1992–1996 1994–1996 OK 1995–1996 1993–1994 1991–1996 IA 1993–1996 1992 1992–1996 OR 1992–1996 1992–1996 ID 1994–1996 1996 PA IL 1995–1996 1992–1994 RI 1993–1996 1991–1996 IN 1993–1996 SC 1996 1992–1995 KS 1993–1996 1992 1993–1996 SD 1996 1992–1995 KY 1996 1991–1996 TN 1994–1996 1994–1996 LA 1995–1996 1992–1994 TX 1995–1996 MA 1992–1996 1992–1996 UT 1996 MD 1995–1996 1992–1996 VA 1994–1996 1991–1996 ME 1994–1996 1991–1993 VT 1993–1996 MI WA 1994–1996 1993–1996 MN 1994–1996 1994–1996 WI 1993–1996 1993–1996 MO 1995–1996 1994 1992–1996 WV 1992–1996 1992–1996 MS 1996 1993–1996 WY 1993–1996 1993–1996 Source:Simon (2000). 4
No state enacted guaranteed issue laws without rating reforms. Since bare-bones plan laws were treated as distinct from the other small-group laws by legislators, they constitute a separate reform category in my analysis.
Jensen, 1999), but this may reflect pre-existing differences. Marquis and Long (2001/ 2002)use employer data from 1993 and 1997 to analyze the effects of reform in nine separate states. Compared with 11 states that did not enact reform, no systematic pattern emerges in these nine to suggest that reforms have had clear effects on offer rates, enrollments or premiums, even when comparing small firms to medium firms. The papers using individual-level data generally find no discernible impact of reform on insurance coverage rates.Zuckerman and Rajan (1999)use national CPS data to look for general effects on insurance rates; thus we do not know whether these reforms have had a differential effect on workers in small firms, the group we expect to be affected.Sloan and Conover (1998)find that, as a result of age rating bans, insurance coverage improved for older workers in small firms in New York and New Jersey. They use national CPS data at the individual level through the March 1994 survey, thus their findings pertain to the early reform period. Kaestner and Simon (2002) find some effects of small group reform on firms with under 100 workers, but their study does not examine firms with fewer than 25 workers separately or use a with-state control group of large firms since the focus of their study is the impact of state insurance mandates, which affect all commercially insured firms. Two individual level analyses look at the effect of reform in small firms, using large firms as a further control group. Using the March CPS,Buchmueller and DiNardo (2002) construct a three-state (PA, NY and CT) case study of the effects of community rating. The authors find statistically insignificant effects of reform on insurance coverage rates.5 Monheit and Schone (2004) use household level surveys from a period years before reforms (the 1987 National Medical Expenditure Survey, NMES) and a period just after them (the 1996 Medical Expenditure Panel Survey, MEPS) to look at the effect of reform on the health insurance status of workers.6 They find no evidence that coverage was affected on average. When the authors look at workers by risk status, they find mixed evidence. In the main specification, the only statistically significant effect is an unexpected negative impact on coverage for high-risk workers as a result of one of two definitions of stringent reform. In an alternative specification, there is a negative effect on high-risk workers from one stringent reform but a positive impact as a result of the other stringent reform. Overall, the sign of the coefficients on the effect of reform on offer rates in small firms is negative, but the sign corresponding to the effects on being covered by that employer’s policy is positive.
These studies do not paint a clear picture of the comprehensive effects of small-group reform. My study adds to the literature in many ways. First, no nationally representative study to date has analyzed the effect of reform on the employer and employee prices of health insurance and other outcomes at the employer level which we expect to be directly impacted by reforms.7 Studies to date that look at indirect outcomes, such as whether
5 This is somewhat at odds with the findings inSloan and Conover (1998)for New Jersey and New York combined.
6
Both Buchmueller and DiNardo and Monheit and Schone use the same DDD method as in this paper, although their definitions of small firms include some sizes not characterized as intended targets in some states’ statutes.
7
There are descriptive comparisons of premiums and offer rates by state, adjusted for plan characteristics, in
workers in small firms are insured, could benefit from additional analysis for a number of reasons. They either use aggregate data and do not identify a separate effect on small firms, or are not nationally representative, or use data from time periods separated by almost a decade. The identification strategy in this paper differs fromMonheit and Schone (2004) in that I exploit important variation in the timing of reforms in different states. The policy details such as content and effective dates were collected from primary sources and checked rigorously. Between 1987 and 1996, all but three states enacted reforms, and these states are likely to differ from the rest of the nation in many ways. A further difference is the definition of the treatment group. In their study, workers are defined to be in small firms if in establishments with fewer than 26 workers, or in single-location establishments with 26–50 workers. This classification may not completely accurately measure the treatment group as defined in these regulations to the extent that establishments with fewer than 26 workers belong to larger firms, and to the extent that many states defined a small firm as one with fewer than 25 workers, not 50 workers.8
3. Methods and data
I estimate the effect of reform by comparing the changes in insurance outcomes for workers in small firms before and after reform, to changes in outcomes for workers in non-reform states, controlling for health insurance outcomes of workers in large firms. This method produces consistent results if the within-state control group picks up exogenous factors that simultaneously affect small firms, and if policies are not endogenously adopted. The data best suited to conduct tests about coverage at the individual level are the March Current Population Surveys (CPS) from 1992 to 1997.9I code a state as having a certain reform if it is effective by the last day of year t2 in order to apply to health insurance sold int1 and reported in the CPS oft.
Using these data, the effect of treatment can be identified through the following probit equation: Yi¼U b1þb2Xiþb3Siþb4TtAiþb5AiSiþb6TtSiþ X3 j¼1 pjRjPOSTijtSi ! ð1Þ
In this equation,Ustands for the normal distribution function,Yirepresents the probability that a full-time workeri received health insurance from his/her own employer.Rj=1 if a state ever had full reform (j=1) partial reform (j=2), or bare bones laws (j=3). Each of these three indicator variables is 0 otherwise.Si=1 if workeriis employed in a small firm
8Although most states shifted to defining a small firm as one with fewer than 50 workers by 1996, 11 states continued to use the cutoff of 25 workers to define small firms in their regulations. One state used 35 and another used 40, while the rest almost all used 50 (Simon, 2000).
9
I select full-time private-sector workers who are between 17 and 65 years of age. I exclude information on workers in firms with 25–99 employees since some states considered them to be small-firm workers and others did not. Hawaii is excluded due to its employer mandate. Further details on the data set construction are provided inSimon (1999).
(less than 25 workers), and 0 otherwise. POSTijt=1 if reform j is effective that year in that state, and 0 otherwise. The regression specification includes state by year fixed effects
TtAi (as well as first-level effects). The coefficient on the three-level interaction term,pj, represents the effect of reform on small-firm workers.10
The March CPS does not contain information on prices and employer decisions to offer coverage, thus I draw on two employer-level surveys to test other predictions about the effects of reform. These employer data also contain independent measures of coverage of workers that allow for a second test of the predictions to compare with CPS results. The data best suited to answer questions about employer level effects of reform are the 1996 Medical Expenditure Panel Survey Insurance Component and the 1993 National Employer Health Insurance Survey (NEHIS) that collected information about health insurance provision and other employer characteristics for 23,000 and 34,604 private employers, respectively. This is the only pair of surveys designed to produce bnational and state estimates of the supply of private health insurance available to American workers and to evaluate policy issues pertaining to health insurance.Q (Sommers, 1999, p. iii). The empirical method followed in this analysis is similar to that used in the CPS analysis, except that now the method of estimation is a linear probability model even for binary outcomes because confidentiality restrictions prevented the two surveys from being analyzed simultaneously.11I estimate regression models of the following outcomes: the premium paid by the firm per worker, the employee contribution paid by the worker, whether the firm offers health insurance, and the fraction of workers who are covered.12
4. Results of individual level analysis
To address the adequacy of the control groups, I first rule out policy endogeneity and ensure that the trends picked up through reform variables did not exist prior to reform.
10 The vectorX includes controls for the following explanatory variables, with excluded categories where appropriate to accommodate an intercept in the model: age, age2, indicators for gender, marital status (five categories), and interactions between gender and marital status, education (nine categories), two indicators of poor health of a family member, hours worked per week, log weekly wage, number of people in the household, categorical variables for firm size, occupation (13 categories), and industry (13 categories).
11The surveys were carried out by two separate federal agencies (the MEPISC by the Census Bureau funded by AHRQ and NEHIS by NCHS) under guarantees of confidentiality, which prevent any transfer of micro data between the two locations that are 25 miles apart. Despite the fact that this prevents analysis by conventional means, I am able to compute regression coefficients, standard errors, and F tests statistics using the basic matrix algebra properties (e.g. partitioning) of the OLS estimator, while still abiding by the confidentiality protocols. This procedure involves creating theX’X matrix components necessary from the first data set at one location, showing that the results contained cross products, which prevent confidentiality breaches, and transferring them to the second site to be combined with similar cross product matrices that could then produce the vector of coefficients, etc. For further details on the estimation method and the data sets used, seeSimon (2004).
12
Since 25–50 full-time worker firms are in the treatment group in some states and control in others, I remove them from my sample. I used 30–60 total workers as an approximation to the 25–50 full time worker category because I know only the total number of workers in the firm. The control variables in these regressions are ones appropriate for explaining an establishment’s insurance outcome such as industry, wage distribution, age of the business, etc. A full list of control variables appears in footnote 16.
With additional data going back to the March 1988 CPS, I test whether the trend in the small-firm vs. large-firm gap in coverage rates is different in reform vs. non-reform states13in the period before reform, and found no statistically significant differences. I also tested whether a state’s adoption of reform in a certain year was influenced by differences in the insurance time trend between small and large firms prior to that year and found no statistical evidence that this was the case. As a further check, states separated into categories based on the date and the type of reform enacted, and coverage rates plotted from the 1988 through 1997 CPS by keeping the groups of states fixed. There was no obvious difference in the pattern between small and large-firm coverage related to the date of enactment. Thus, even though on a national scale insurance reform may have been prompted by a deterioration of the small-group insurance market, it does not seem to explain the pattern of reform adoption across states, which is the source of variation in this study.
Second, I present an empirical exercise inTable 2corresponding to the univariate DDD estimator. This shows the CPS mean coverage rates for workers in small and large firms, before and after reform, in states that enacted reforms and in states that did not, and indicates that full reform decreased the probability that a full-time worker in a small firm received health insurance by over two percentage points. Partial reforms have no discernible impact. If one simply uses 1996 data to compare the coverage rate for small firms in states with full reform to that in states without reform, it would appear that full reform decreased coverage by 10 percentage points. However, looking at the correspond-ing coverage rates for large firms suggests that part of this difference is due to underlycorrespond-ing level differences between the two groups of states. These numbers underline the usefulness of these control groups in accounting for secular trends and level differences between states that arise for extraneous reasons.
Table 2
Effect of full and partial reform on coverage rates: DDD calculations from means, 1991 and 1996 Effect of full reform
Before reform After reform
Reform Small 39.36 37.39
Large 75.79 73.71
No reform Small 47.18 47.04
Large 79.61 77.36
DDD estimate =2.01 percentage points Effect of partial reform
Before reform After reform
Reform Small 40.56 0.40949
Large 76.41 0.7486
No reform Small 47.18 47.04
Large 79.61 77.36
DDD estimate=0.18 percentage points
13I define reform/non-reform states according to the state’s status in 1993, 1995 and 1997 separately. More details on these tests are available upon request.
I next turn to the multivariate estimates that control for other determinants of insurance coverage.Table 3 shows the CPS probit coefficients, the associated marginal effects for the three variables of interest, and standard errors for both the coefficients and the marginal effects.14Most findings are consistent with the theoretical predictions of the effect of reform on coverage rates. On average, full reform carries a coefficient of0.057 with a standard error of 0.026. This translates into a statistically significant marginal effect of1.9 percentage points on the coverage rate of small-firm workers.15Given that 39% of full-time workers in small firms receive health insurance from their employer, this means that full reforms caused on average a 5% decrease in the rate of employer-provided coverage. Coverage rates declined by a larger magnitude for workers whom insurers considered low risk, while those considered high risk were not significantly affected and thus, fared better than the average small-firm worker. Partial reform has an insignificant effect in almost all cases, and a marginal effect that is generally smaller in magnitude than Table 3
Probit results, individual level (dependent variable=1 if worker received health insurance from employer) Sample Observations Sample
mean
Small*Full*Post Small*Partial*Post Small*BBP*Post Probit coefficient Marginal effect Probit coefficient Marginal effect Probit coefficient Marginal effect Whole sample 222,032 0.64 0.0566 [0.0263] 0.0185 [0.0086] 0.0057 [0.0285] 0.0019 [0.0094] 0.0225 [0.0251] 0.0074 [0.0082] Never married males b35 years 23,256 0.51 0.2305 [0.0765] 0.0635 [0.0201] 0.0054 [0.0875] 0.0016 [0.0252] 0.0179 [0.0745] 0.0050 [0.0207] Married women b41 years with kids 20,837 0.53 0.1008 [0.0930] 0.0308 [0.0291] 0.0666 [0.0983] 0.0194 [0.0281] 0.0570 [0.0875] 0.0171 [0.0259] Standard errors in parentheses. Marginal effects are calculated by first establishing a baseline at 1991 by changing all the variables involving time, then setting the relevant three level interaction to 1, and 0, and computing the average difference in predicted probability of the dependent variable over all small-firm workers. Standard errors for marginal effects are calculated using a Taylor Series approximation (theddelta methodT). Bold font indicates significance at least at thep=0.10 level.
Other variables included in this regression for which results are not reported are: worker’s age, age squared, gender, marital status (five categories), interactions between gender and marital status, education (nine categories), indicators of poor health of a family member, hours worked per week, log weekly wage, number of people in the household, firm size, occupation (13 categories) and industry (13 categories), state effects (49), year effects (6), state by year interactions (300), year by small firm size (6), and state by small firm size (50).
14
Removing potentially endogenous variables such as hours worked and wages did not affect the results substantially. The control variables in the model showed plausible effects, and their coefficients and standard errors are available from the author upon request.
15
Marginal effects are calculated to simulate the effect on coverage rates for small-firm workers if all states reformed, relative to remaining without reform. I chose 1991 as the baseline year at which to evaluate the effects of reform by changing all the appropriate year by state and year by small-firm worker interactions. I calculated the predicted probability of coverage under each type of reform for each small-firm worker by first setting the appropriate three-level interaction to one, and then calculated the probability of coverage under no reform by setting the relevant three-level interaction(s) to zero. The average difference between these two predicted probabilities over all small-firm workers is the marginal effect reported inTable 3. The appropriate standard errors are computed using a Taylor Series approximation.
full reform. Although states hoped that bare bones plans would provide low-cost alternatives to uninsured small firms by removing mandated benefits, the empirical results here suggest that they had a statistically insignificant impact on the small-group market. Ideally, one would categorize people based on specific health risks and prior medical use that insurers could detect through experience rating but are unable to use because of rating reform. Since the CPS does not contain such detailed variables, I obtained medical utilization information from the 1996 Medical Expenditure Panel Survey (MEPS) to investigate the link between demographic factors and expected health costs. These data show that married women of childbearing age (between 16 and 41 years) are more than five times as likely to be admitted to hospital, have almost three times the number of doctors visits and spend twice as many nights in hospital as never-married men under 35 years of age. Absent reforms, there is evidence that insurers sometimes refused to cover businesses employing a high proportion of childbearing-aged women for this reason (Zellers et al., 1992). It is unfortunate that the CPS lacks detailed health data because these are demographic characteristics that about half of all states did not restrict through reform. Only one state fully banned the use of age rating (New York).
When I limit my sample to young never-married men under 35 years of age, a relatively low-cost demographic group, the marginal effect of full reform is 6.4 percentage points and statistically significant atp=0.01. The effects of partial reform and of bare bones laws are negative but statistically insignificant. However, when I use the relatively high expenditure sample, married female full-time workers of childbearing age (17–41 years) with young children, the effect of full-reform is positive but statistically insignificant. For similar reasons, I expect that those men likely to include women of childbearing age under their policies should also be affected in a similar way. When I restricted the sample to married men under 45 years of age with children, I found that the effect due to full reform was similar to the effect for married women with children in unreported results.
Since many states allowed pricing according to characteristics that define the high and low expenditure groups, part of the result above is likely due to forces other than price differentials caused by reform. One possibility is that since these groups differ in their need for health care, they also may differ in their need for health insurance and sensitivity to premium changes. Unfortunately, the employer level data used are not rich enough to discern the differential premium increases for these groups. Even though married individuals would in general have better outside options (and thus be more price sensitive), those in the high expenditure group who had insurance in their name may have been less likely to have a readily available source of spousal coverage given that dual coverage is not very common. I explicitly tested the hypothesis that reforms lead to a greater reliance on spousal health insurance by the high-risk group, and found no evidence of this in unreported results. In summary, the differential in results for the two groups is larger than expected, and further work using better health data is needed to explore this issue.
I have attempted to reconcile my findings with other individual level studies. When I limit my sample to just Pennsylvania, New York and Connecticut, I find insignificant effects of reforms, consistent with the findings in Buchmueller and DiNardo (2002). To test the stability of my findings, I undertook several specification checks. I first establish
that the results are not driven by the control group by investigating whether small-group reforms had any bearing on large-firm worker coverage rates. I found no apparent effects of the treatment on the control group. I then changed the definition of a small firm to one with fewer than 10 workers. I estimated this alternative model because smaller groups were considered higher risk pre-reform and might have been able to increase coverage (or not drop coverage as much) by being grouped together with somewhat larger firms. I find that effects of reform are smaller in magnitude and statistically insignificant for workers in smaller firms. I re-estimated the main models using data just on the outgoing rotation group to see whether the re-appearance of the same individuals in my sample, as they rotate through CPS, affects the estimates. The results indicate that the coefficient on full reform is slightly smaller when I focus on the outgoing rotation group. When using the complement of the outgoing rotation group, the coefficient is slightly larger and the standard error is the same as for the outgoing rotation sample. I also separated the sample into odd and even year observations, and found that the coefficients did not change substantially as a result of this sample split. As a further specification check I investigated whether part-time workers (workingb20 and b15 h/week separately) who are generally not eligible for health benefits were affected by reform. I find that as expected, the coefficients are smaller and statistically insignificant for these workers.
Within the group I call full-reform states, there is variation in the number of plans that insurers were required to issue on a guaranteed basis, and in thebtightnessQof the rating reform measured by the health band. Unreported estimates suggest that the strength of the guaranteed issue law was important, and that the variation in rating reform is insignificant when entered as a continuous variable, but significant when coded as above and below the median value.16 In states with all plans guaranteed issue (as opposed to select plans issued on a guaranteed basis), the effect of full reform was a two percentage point decline in coverage rates on average, statistically significant at the 1% level. The effect of this type of reform on the low-risk group was a statistically significant decline in coverage rates of 6.2 percentage points. The most interesting result from this exercise is that guaranteeing the issue of all plans in full reform states had a weak but positive and statistically significant (at the 10% level) effect of 6.7 percentage points on the high-risk coverage rates.
Table 4
OLS results, establishment level: the impact of full reform
N Mean Small*Full*Post
Premiums 26,651 181.1 7.8 (4.2)
Employee contribution 28,052 32 5.1 (2.4)
Decision to offer 50,485 0.66 0.01 (0.01)
Coverage rate 47,598 42.9 2.12 (1.29)
Standard errors in parentheses. Bold font indicates significance at least at thep=0.10 level. See footnote 16 for a full explanation of control variables included.
16
This measure is highly correlated with the variable measuring the strength of full reform, hence I use just the latter.
5. Results of establishment level analysis
Table 4shows the effect of reform on establishment level outcomes utilizing the matrix algebra explained earlier.17Given the smaller sample size and shorter time period available in the establishment data compared to the CPS, I look at the effect of full reform, including no reform and partial reform in the omitted category. This specification indicates that full reform increased premiums on average by a statistically significant US$7.80 a month per person, that employee contributions rose a statistically significant US$5.10, that employer decisions to offer coverage were not affected, but that the percent of workers covered at the firm fell by a statistically significant 2.12 percentage points, a magnitude that is very close to that from the individual level analysis. Further investigations testing whether there is a differential impact on firms by predicted workforce characteristics, or by whether the industry is red-lined or not (Zellers et al., 1992) showed results that are not statistically significant, but whose magnitudes are generally consistent with predictions of an increase of premiums, employee contributions and decreases in coverage rates for lower-risk firms.
6. Discussion and conclusion
The empirical results in this paper suggest that stringent small-group reforms have spread the costs associated with health risks more evenly across the market and may have unintentionally reduced insurance coverage through increased premiums and employee contributions. The analysis of both employer and individual data yields fairly consistent answers. Given that employers pass on about 75% of the premium hike as increased employee contributions, it appears that employee contributions serve the role of allowing employers to compensate workers efficiently. The results also suggest that young single men were particularly sensitive to premium changes in their take-up decisions, although the magnitude is larger than expected since the combination of gender, age and marital status were observable to insurers in half the reform states. Further research is needed with more detailed health risks than available in the CPS to fully understand the impact of reform by risk level.
Specification checks indicate that the relationships emerging from my analysis represent more than chance correlation. Economic theories of insurance markets warn
17The explanatory variables included in the model include: a linear and quadratic term for the number of employees at the firm, 10 industry indicators, an indicator variable for the presence of unionized workers at the establishment, age of the firm and its square, the fraction of low-wage workers at the establishment level (defined as earning below US$6.50 an hour in 1996 and below US$5 an hour in 1993), the fraction of high wage workers (defined as earning above US$15 an hour in 1996 and in 1993). In regressions that involve premiums and employee contributions, I also control for the type of plan (whether HMO, conventional, or a mixed plan, and whether self insured in the case of large firms), the amount of the total deductible, whether there is a lifetime maximum benefit and its amount, the co-payment required for outpatient treatment, and the coinsurance rate for inpatient and outpatient treatment to proxy for the quantity of coverage provided by the plan. Additionally, I include 14 indicators for specific covered services such as outpatient prescription drugs. When control variables contain missing values, I create separate indicator variables that take a value of 1 for a valid number and 0 for a missing value, and I replace missing values by zero.
us that preventing insurers from distinguishing between different risk groups may worsen the availability of insurance for healthier individuals but not for those who are considered medically expensive. This analysis of small-group health insurance reform reveals the complexity of regulating a market subject to adverse selection. By changing the rules of conduct for insurers, states did not directly address what many argue is the single most important reason why small firms are uninsured—high prices. Several questions about the effects of government intervention in health insurance markets remain unanswered. If forcing insurers to treat all customers alike does not produce optimal results, then what will? Should government policy encourage small employers to band together in purchasing pools to take advantage of size as large firms do? The optimal design of insurance regulations that takes adverse selection behavior into account, and other solutions to reducing the medical uninsurance problem remain areas of high priority for future research.
Acknowledgements
I am grateful to William Evans, Judith Hellerstein, Steve Coate and Jonathan Gruber for valuable suggestions. I would also like to thank seminar participants at several universities and conferences for helpful comments. I am grateful to the National Center for Health Statistics for allowing me to conduct a portion of this research at their offices. Part of this work was conducted while the author was a Research Associate at the Census Bureau. Dissertation support from the former Health Care Financing Administration is graciously acknowledged for a portion of this research. All errors and opinions expressed are mine.
References
Buchmueller, T., Jensen, G., 1997. Small group reform in a competitive managed care market: the case of California, 1993–1995. Inquiry 34, 249 – 263.
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.
Hall, M., 1999. An Evaluation of Health Insurance Laws. Wake Forest University School of Medicine. Various reports available atwww.phs.wfubmc.edu/insure(Access date 2003).
Hing, E., Jensen, G., 1999. Health insurance portability and accountability act of 1996: lessons from the states. Medical Care 37 (7), 692 – 706.
Jensen, G., Morrisey, M., 1999. Small group reform and insurance provision by small firms, 1989–1995. Inquiry 36 (2), 176 – 187.
Kaestner, R., Simon, K., 2002. Labor market consequences of state health insurance reforms. Industrial Labor Relations Review 56 (1), 136 – 160.
Marquis, M.S., Long, S., 2001/2002. Effects of dSecond GenerationT small group health insurance market reforms, 1993–1997. Inquiry 38, 365 – 380.
Monheit, A., Schone, B., 2004. How has small group market reform affected employee health insurance coverage? Journal of Public Economics 88 (1–2), 237 – 254.
National Association of Insurance Commissioners (NAIC), 1998. Model Laws and Regulations. Kansas City, MO.
Simon, K. 1999bThe Impact of Small-Group Health Insurance Reform.QDissertation. Department of Economics, University of Maryland.
Simon, K., 2000. State Profiles of Small Group Health Insurance Reform, 1990–1999. University of Maryland. Typescript.
Simon, K., 2004. Adverse Selection in Health Insurance Markets? Evidence from State Small Group Health Insurance Reforms. Cornell University. Typescript.
Sloan, F., Conover, C., 1998. Effects of state reforms on health insurance coverage of adults. Inquiry 35, 280 – 293.
Sommers J., 1999. List Sample Design of the 1996 Medical Expenditure Panel Survey Insurance 1999 Component.QRockville (MD): Agency for Health Care Policy and Research. MEPS Methodology Report #6. Zellers, W., McLaughlin, C., Frick, K., 1992. Small-business health insurance: only the healthy need apply.
Health Affairs 11 (1), 174 – 178.
Zuckerman, S., Rajan, S., 1999. An alternative approach to measuring the effects of insurance market reform. Inquiry 36, 44 – 56.
