Effect of Recent Federal Health Insurance Regulations on Uninsurance in Small Firms CONFERENCE DRAFT-PLEASE DO NOT CIRCULATE OR CITE Kosali Simon Assistant Professor Department of Policy Analysis and Management Cornell University, Ithaca NY 14850, and NBER Abstract: Lack of health insurance in the US tends to be a problem disproportionately concentrated among low skilled small firm workers and their families. Policies aimed at assisting small firms offer health insurance have the potential to help millions of uninsured Americans. This paper examines the effect of recent federal health insurance reform contained in the 1996 Health Insurance Portability and Accountability Act (HIPAA) that specifically targets small firms. This federal policy in combination with a patchwork of other state small-group health insurance reforms already on the books created substantial variation in effective regulations across states. Preliminary findings indicate that these laws have had no detectable effect on uninsurance or employer based insurance in small firms.
The latest statistics on the uninsured released by the Census Bureau shows that the share of the American population without health insurance rose in the year 2003 due to a decline in the likelihood of employer based health insurance coverage, which dropped from 61.3 percent in 2002 to 60.4 percent in 2003 (US Census Bureau, 2004). From 1994 to 2000, employer based coverage increased steadily, but has been on a continuous decline since then. 1 The slowdown of the economy and the continued rise in health care costs are two forces that plausibly help explain this trend. Looking closer at the time series of employer health insurance offer rates shows that the rise during the mid to late 1990s and the fall during the most recent years are more pronounced in small firms than larger firms (see Figure 1). Another perspective on this trend comes from looking at the share of workers who receive health insurance from their employers. Author calculations from the Survey of Income Program Participation (SIPP) show that the share of employees in small firms (fewer than 25 workers) who reported health insurance from their own employer rose from 29% in 1996 to 33% in 2003, while for large firms (100 or more workers), this went from 67% to 68%. 1 Employer offers of insurance have been rising from 1996 to 2002 from 52.9% of private firms offering health insurance to 59.3% in 2000. Since then, this number has declined to 57.2 in 2002. (AHRQ, 2004a).
Fraction of Private Sector Establishments Offering Health Insurance by Firm Size 1.20 1.00 Fraction 0.80 0.60 0.40 Under 50 Employees Over 50 Employees 0.20 Figure 1 0.00 1994 1996 1997 1998 1999 2000 2001 2002 Year More striking than the difference in insurance trends between small and large firms is, of course, the difference in the levels. In 2002, 37% of firms with fewer than 10 workers offered health insurance, while in firms with 50 or more workers, 97% did so (AHRQ 2004 a). Among workers offered health insurance, take-up rates do not differ to a great extent by firm size, and if anything, the secular decline in take-up has been slightly more among larger firm employees. From 1996 to 2002, the take-up of health insurance among eligible employees changed from 86.7% to 81.7% in firms with greater than 50 employees, while in firms with fewer than 50 workers, this decline has been from 81.1% to 78.5% (AHRQ 2004 b). Small firm workers and their families are less likely to be covered by health insurance for three notable reasons: adverse selection problems that worsen as group size reduces; limited ability to bargain and spread fixed costs over more covered lives; and features of small firm employment such as a concentration in lower skilled occupations which are associated with a lower demand for health benefits. The statistics noted above point to large differences in the incidence of employment related health insurance in ones
own name by firm size, but this does not necessarily translate to differences in uninsurance rates by firm size among families due to several features of the US health insurance structure, including the availability of family based coverage as well as sources of health insurance not tied to employment. However, statistics of family uninsurance rates by firm size show that there are large differences. Author calculations from the SIPP 2 shows that the uninsurance rate among individuals under 65 year of age who are connected to small firms 3 fell from 28% in 1996 to 24% in 2000 before rising again to 26% in 2003. Among those connected to larger firms (100 or more workers), the rate of uninsurance fell from 9.1% in 1996 to 8% in 2000 before rising again to 9.3% in 2003. Employer based health insurance is estimated to cover 50.8% of people connected to small firms and 82% of individuals tied to large firms in 1996, while for 2003 the numbers are 52% and 80.6% respectively. These numbers, together with the information on offers of health insurance by employers above, suggest that take-up rates have increased slightly among small firms relative to large firms over this time period. 4 However, large gaps persist by firm size. 5 The high uninsurance rates in the small group market attracted attention from 2 Note that although the estimates use longitudinally corrected SIPP weights, there is still reason to expect differences across panels due to the nature of the questionnaire used changing, e.g. the order of the questions may not be the same. While this is not a problem in regression analysis since panel fixed effects are used, this means that raw differences in insurance rates may reflect more than just changes over time. However, what is important here is the relative difference between small and large firms, and the change in panels should not be of much concern in this regard. 3 I first sort individuals into health insurance units (nuclear families consisting of parent(s) and any children who are dependents) or married couples without children, or single individuals) and then designate one adult within a unit to be the main worker based on hours worked a week. This worker s firm size is noted, and all individuals within that health insurance unit are regarded to be tied to employment in that firm size. 4 The firm size definitions used in this discussion are different when referring to firms vs individuals due to the measures publicly available (firm size 100 or more workers is not available for employers). 5 Fronstin (2004) shows using March CPS data that in 2003, the rate of uninsurance among nonelderly Americans whose family head worked in a small firms (fewer than 10 workers) was 31.8%, which was even higher than the uninsurance rate among Americans whose family head did not work (27.8%). The uninsurance rate among those attached to a firm with 10-24 workers was 26.4%.
policy makers in the early 1990s, and lead to the enactment of legislation that has come to be known as small-group health insurance reform. 6 These laws sought to address insurance market practices used by insurers to differentiate prices among small firms, and to prevent redlining businesses (whereby insurers refused to offer coverage to firms that they considered the most unprofitable clients). These policy initiatives can be separated into rating reforms (which restricted the range of premiums charges based on health or other demographic factors), guaranteed issue (which stipulated whether any (and how many) plans should be offered to all comers), guarantees renewability, and preexisting conditions exclusions. The evidence to date shows that these laws have done little to improve the uninsurance problems in small firms (See Simon 2004 for a review of the recent literature). Similar legislation in individual markets has also not produced higher rates of coverage either (See Chollet 2004, for a review of the recent literature). More recent legislation that has focused on small firms is Title 1 (the insurance provision section) of the Health Insurance Portability and Accountability Act of 1996 (HIPAA) that became effective July 1, 1997. The effect of these regulations on insurance outcomes in small firms has not been analyzed to date, despite a large number of papers that analyze pre-hipaa small-group market laws. One section of HIPAA establishes rules to protect workers who leave their current employer to accept a job with another employer or to enter non-employment or unemployment by addressing portability and pre-existing conditions stipulations in policies sold in the group insurance or individual health insurance market. Plans may not establish pre-existing conditions exclusions periods longer than 12 months after the Act 6 Concerns about the functioning of the individual health insurance markets also lead to reforms in that market.
becomes effective. The look-back period is limited to 6 months. Employers are also required to use periods of continuous health insurance coverage from prior sources towards fulfillment of the pre-existing conditions period ( portability ). All group health insurance is required to be sold on a guaranteed renewable basis, except in obvious cases such as when an employer fails to pay premiums due. A second section of this law that applies only to firms with 2 to 50 or fewer employees is the guarantees issue requirement. This means that insurers selling in the small-group market must offer all policies to all small employers who wish to buy (and to all eligible workers within those firms). This precludes insurance companies from denying applications from clients they do not wish to insurance. Even though the HIPAA small employer provisions by themselves do not appear binding in the sense that insurers are free to use price as the selection mechanism, the existence of state-level small-group reforms that restrict rate setting creates variation in the degree to which we expect HIPAA to exert an effect. Rating laws, which by themselves may not have been binding either because of the ability of insurers to refuse to sell policies to certain firms, are made more binding by the imposition of federal guaranteed issue requirements since insurers exposed to both laws now need to sell policies to all firms who apply, and are restricted in their ability to differentiate price across customers based on health and/ or other factors. This combination of state and federal laws provides researchers with a quasi natural experiment to study the effects of insurance reform on small firms using richer variation than would have existed had HIPAA not been superseded by state small-group laws. Specifically, the questions addressed in this paper are the following:
How does requiring all plans to be issued to all comers, combined with rating laws, affect the probability that small firm workers and their families are insured by their employer? How does this combination of laws affect the probability that workers and their families are uninsured? Data I use two main sources of information on health insurance coverage of individuals linked to firms of varying size. One is the Survey of Income and Program Participation. This survey is conducted by the Census Bureau in panels that follow cohorts of individuals over time. I use the two most recent panels of the SIPP, the 1996 panel that goes through 2000, and the 2001 panel that goes through 2003. This survey has been used in prior health insurance research and has some notable features which distinguish it from the March Current Population Survey, including the ability to follow the same individuals over a long period of time and to know insurance status at each point in time during the year. A new feature added to these panels relative to earlier panels is firm size (asked in the core questionnaire and thus collected for every month) and whether the firm offered workers health insurance (in a topical module, asked once per panel). I plan to use data from the Current Population Survey March Supplement for the 1996-2003 years in the future to check the robustness of the SIPP results. The advantage of the March CPS survey is its larger size and more detailed breakdown of firm size. I classify small firms in the SIPP as those with 25 or fewer workers. 7 A large firm is defined as one with 100 or more workers. Although this does not allow one to test whether the smallest firms are differentially affected, among all small firms, these cutoffs are useful for this research, as all states included incorporated firms with 25 or fewer
workers in setting the applicability of small group reform (for example, in 1996, 11 states used 25 as the maximum size of a small firm, and thus a definition that included all firms with fewer than 50 workers would not be ideal). The sample is selected in two further ways. Those over 65 years of age are not included in the sample, and observations from Maine, Vermont, North Dakota, South Dakota, and Wyoming are excluded since they are not separately identified in the 1996 and 2001 SIPP panels. I merge the household level data in these two surveys with state and federal insurance reform policies gathered from statutes of each state during this time period. A search of legal documents was conducted in the summer of 2003 to document the small group laws and HIPAA implementation in effect in states during 1996-2003 as they applied to the main areas of small group reform. More information was coded and will be used in future work than is used in the current version of the paper, including the specific range of premiums allowed based on commonly used factors (including, where applicable, age, gender, health status, smoking status, industry, geographic area, and size category of employer). Guaranteed issue laws were classified according to the number of plans that were affected. Pre-existing conditions laws were coded according to the waiting period (number of months during which insurance would not pay for the preexisting condition) and look back period (number of months during which care was sought in the past for an existing illness for it to count as pre-existing). Renewability laws were classified according to the allowable increase in premiums. All laws were further classified by the size of the groups affected. While most laws applied to firms with 50 or fewer employees, some provisions applied only to groups with 25 or fewer employees 7 Individuals are asked about firm size and establishment size separately, if the firm operates in more than one location. No further differentiation in firm size is permitted other than 1-25, 26-99, and 100+ workers.
and a handful applied to other firm sizes. Since other factors about a local market and other state policies may influence employer decisions to offer health insurance, we also merge in data at the state level on the unemployment rate (monthly data by state) from the Bureau of Labor Statistics and an index of Medicaid/SCHIP generosity (by state by month, measured as the fraction of a nationally representative group of 14 year old children who would qualify for public health insurance in a particular state in a particular month). Method of Analysis The aim of the analysis, estimating the effect of HIPAA on small firm insurance rates, can be accomplished in several ways. There are three methods used in this paper: simple descriptive differences; two versions of differences in differences analysis; and a triple difference strategy. As argued above, the combination of laws that is expected to bind the most on insurer behavior is a combination of rating laws together with guaranteed issue. HIPAA caused exogenous variation in this because a number of states had rating reforms without guaranteed issue laws prior to 1997. On the other hand, a set of states contained guaranteed issue laws prior to 1997 as stringent as those imposed by HIPAA, forming a control group who should not be causally affected by the change in laws. A comparison of these two groups of states should yield the strongest test of the hypothesis that these laws had an effect on insurance in small firms. 8 Because of the time variation, observations in the same state before and after the law change can be used to control for time invariant differences in insurance rates between states. As a further
check, large firms in the same state can be used to control for secular time trends in insurance rates that vary across states in ways that may correlate with the change in laws. Thus, this gives rise to a series of differences, differences in differences and triple difference estimation strategies that one could choose between. The simplest method would be to compare the insurance rate in small firms in the group of states identified as the treatment group with the insurance rate in the states identified as the control group. The states that fall into the treatment group (those who has no plans guaranteed issue, but had some rating restrictions) are the following: AR; AZ; GA; IL; IN; LA; MS; NH; NV; UT; WI; and WV. The states in the control group (those that has guaranteed issue of all plans prior to HIPAA) include the following: FL; KY; MA; MD; MN; NJ; NM; NY; OR; VT; and WA. 9 Tables 1a and b show the results of this descriptive exercise. The pre period includes 1996 and 1997 data, and the post period includes the years 1998-2003. The outcomes are uninsurance and employer health insurance for the workers, as well as for the families of the workers, by whether they are connected to small or large firms, in these two sets of states. All descriptive statistics presented are weighted by sample weights. Table 1a: Before HIPAA Treatment Control Small emphi_worker.306 (.46).339 (.473) unins_worker.251 (.43).241 (.428) Emphi_family.54 (.50).57 (.50) unins_family.26 (.44).24 (.43) Large Emphi_worker.658 (.47).686 (.46) unins_worker.099 (.299).08(.273) 8 Another possible exercise would be to see whether guaranteed issue by itself (in the states that contained no reforms of any kind prior to HIPAA) affected insurance coverage. This is less theoretically interesting as insurers should be able to use pricing to circumvent these laws, and is thus not studied in this paper. 9 Note that of these states, only Vermont is not included in the SIPP.
Emphi_family.83 (.38).84 (.36) unins_family.096 (.29).08 (.27) Table 1b: After HIPAA Treatment Control Small emphi_worker.324 (.467).36 (.48) unins_worker.252 (.434).236 (.42) emphi_family.55 (.50).58 (.49) unins_family.26 (.44).23 (.42) Large emphi_worker.674 (.468).68 (.46) unins_worker.090 (.2866).079 (.27) emphi_family.83 (.38).84 (.37) unins_family.09 (.29).08 (.27) Note: Weighted means fractions are shown, such that.306 means 30.6 percentage points, with standard deviations in parentheses. The abbreviations used are as follows: emphi_worker: Employer health insurance in own name among the workers of this firm size unins_worker: Uninsurance rate of the workers of this firm size emphi_family: Employer based health insurance in the families attached to workers of this firm size unins_family: Uninsurance in the families attached to workers of this firm size In a simple difference exercise, comparing the own employer insurance rate in small firms in the treatment states before and after HIPAA shows that the rate increased from.306 to.324. But a difference in difference version that also took into account changes in the control states would show that the employer insurance rates increased there as well, from.34 to.36, yielding a net change of close to zero. A different type of difference in difference method would consider the change in small firms over time in treatment states to the change in large firms over time in treatment states. This second change is also an increase, from.658 to.674. A triple difference strategy would take all these factors into account and look at the change in small firms in treatment states, relative to the change in small firms in the control states, relative to changes in large firms in both types of states. The next section embeds these double and triple difference estimations in regressions that control for various other factors that may influence insurance rates across firm sizes, states and time.
Regression Analysis The first outcome studied is whether employees receive health insurance from their employers. The second is whether employees are uninsured. The third and fourth outcomes are employer health insurance and uninsurance among families of workers attached to small and large firms. The regression model for all outcomes takes the following form, and is estimated using OLS for simplicity (in the future, logit or probit models will be used). Y ( X S A _ HIPAA T A _ HIPAA S T S T A _ HIPAA i 1 2 i 3 i 4 t 5 i 6 i i 7 t i 8 t i A _ HIPAA S T ) 9 t i i Y stands for an insurance outcome at the individual (i) level, X stands for a vector of other factors that influence insurance status, S is a dummy variable for a small firm worker, A_HIPAA is an indicator for the post HIPAA time period, T is a dummy variable for the treatment group of states. The coefficient on the last term has a DDD interpretation, and the model can be simplified to yield various DD estimates, for example, by limiting the sample to only small firms or to only the treatment states. In the estimation, instead of T (dividing states into treatment and control groups), state fixed effects are used for more flexibility. Instead of A_HIPAA (dividing time periods into before and after HIPAA), year fixed effects are used for similar reasons. Likewise, second level interactions are expanded, except in the case of T*A_HIPAA in which case state specific linear time trends are used instead of state by year fixed effects. Standard errors are clustered at the state level. For all four outcomes, the coefficients and standard errors reported are the
relevant DD or DDD terms for brevity of presentation; coefficients and standard errors for other regressors are available upon request. In regressions studying the first two outcomes, the insurance status of workers, the vector X contains the following variables: gender, race (White, Black, Hispanic, Asian and Other), age in years and its square, hours worked a week and its square, marital status (married or not), number of kids in the family, education (high school drop out, completed just high school, has some college, completed college, has attended graduate school), state monthly unemployment rates, Medicaid generosity index, SIPP panel fixed effects, SIPP wave fixed effects, month of year fixed effects), in addition to state fixed effects, year fixed effects, fixed effects for state by small firm, fixed effects for year by small firm, and state specific time trends. In the family health insurance regressions, we use the education level of the main worker instead of the individual s own education level, and drop the hours of work variable. All other variables are included as above. Results Table 2 shows results from the regressions for all four outcome variables (defined as in Table 1a and b), for the DDD model as well as for two DD models. The first column shows the coefficient and the standard error of the estimates. The next two columns show the sample size and the adjusted R 2 of the regression. All models indicate no statistically significant effects of reform on insurance outcomes for workers or for families, in terms of employer provided health insurance in one s own name, in a family member s name, or lack of any insurance. Furthermore, point estimates are close to zero. For example, the DDD model shows that the prevalence of employer insurance in ones own name decreased by.4 percentage points, but is not statistically significantly different from zero
at conventional levels of confidence. The estimates that comes closest to reaching statistical significance at the 10% level (it is only statistically significant at p=.11 when calculated to the fourth decimal place) is the emphi_family coefficient of.024. These estimates do not show any discernible impact of reform on insurance status in small firms. Table 2: Effect of Reforms on Insurance Outcomes. Regression Coefficients and Standard Errors Outcomes DDD Estimates DD (only small firms) DD (only treatment group) N R 2 N R 2 N R 2 emphi_worker -.004 (.01) 254,264 0.26.0016 (.01) 61,404 0.14.008(.007) 115,747 0.26 Unins_worker.004 (.01) 254,264 0.14.007 (.01) 61,404 0.15.002(.008) 115,747 0.14 emphi_family.009 (.018) 417,140 0.18.002 (.01) 93,684 0.13.024(.014) 192,278 0.18 Unins_family.006 (.009) 417,140 0.13 -.004 (.01) 93,684 0.13 -.011(.008) 192,278 0.14 There are various specification checks that can be performed to investigate these results further. Two such checks are undertaken here. The first is to take advantage of the panel aspect of the SIPP and see if using person fixed effects produces the same results. This is done in the first column of Table 3, where the 1 st column of Table 2 is re-estimated with person fixed effects added (with all other features of the model, including state level clustering, still present). This shows that estimates are not appreciably changed except in one instance. For example, the coefficient of interest in the model for workers own employer health insurance status is -.005 with a standard error of.01, whereas it was -
.004 with a standard error of.01 without person fixed effects. The only estimate that changes in a notable manner is the one for uninsurance of families attached to small firms (the last row). The coefficient now indicates that there was a statistically significant rise in uninsurance of 1.6 percentage points as a result of reform. The next point that is explored is whether there is any difference in effect by health status. Theoretically, it may not be surprising that there is no effect on average, but we expect those with worse health to benefit the most from the laws. In this specification check, I separate individuals that have self reported health status of Excellent from those who have reported Poor or Fair. Again, there is only one statistically significant result. The last row shows that the uninsurance rate among family members who are less healthy rose by a statistically significant 2.5 percentage points. This result runs against expectations. Table 3: Robustness Checks. Effect of Reforms on Insurance Outcomes. Regression Coefficients and Standard Errors DDD (with person fixed Outcomes effects) DDD (healthy) DDD (less healthy) emphi_worker -.005 (.01).011 (.018) -.002 (.03) unins_worker.01 (.01) -.004 (.016).008 (-.028) emphi_family -.002 (.01).02 (.02).003 (.03) unins_family.016 (.0087).003 (.02).025 (.013) Conclusion and Future Work Small employers consistently cite cost as the main reason why health benefits are not offered. For example, of those surveyed in the Employee Benefits Research Institute Small Employer Health Benefits Survey in 2000, 10 53% of employers said that affordability was a main reason they did not offer coverage. Only 10% cited as a major reason the fact that their workers were healthy and didn t require it, although 30% said a 10 Available at http://www.ebri.org/sehbs/
major reason that they do not offer health insurance is because their workers prefer wages over benefits. It also appears that employers who do not offer coverage are unaware of several features than act in their favor, such as the tax treatment of employer benefits (the majority of employers did not know the correct answers to questions on this subject) and the regulations that make health insurance guaranteed issue or rating reforms (the majority of employers did not know about these laws either). This suggests that policies to encourage take-up on the part of employees and offers of coverage on the part of employers through information and education campaigns may be promising. Other policies that encourage employer pooling to reduce administrative costs also may help. However, results from these preliminary investigations suggest that efforts of the form of guaranteed issue enacted through HIPAA have had no discernible impact on uninsurance rates of rates of employer health insurance in small firms. In future work, I plan to conduct several more empirical investigations into the robustness of these results using: 1) better measures of health than self reported health status ( the SIPP contains health care utilization data which is better suited for this task); 2) more detailed measures of reform (for example, the tightness of the rating bands will be interacted with the guaranteed issue laws since there is heterogeneity within the treatment group of states in the extent to which rates are regulated); 3) other data sets ( I plan to re-analyze this question using the March CPS, and to investigate employer behavior using the Medical Expenditure Panel Survey Insurance Component List Sample which spans the years 1996-2002). References
AHRQ 2004a,Agency for Healthcare Research and Quality. Percent of private-sector establishments that offer health insurance by firm size and selected characteristics (Table I.A.2), years 1996-2002: 1996 (Revised March 2000), 1997 (March 2000), 1998 (August 2000), 1999 (August 2001), 2000 (August 2002), 2001 (August 2003), 2002 (July 2004). Medical Expenditure Panel Survey Insurance Component Tables. Generated using MEPSnet/IC. <http://www.meps.ahrq.gov/mepsnet/ic/mepsnetic.asp> (December 09, 2004) AHRQ 2004b, Agency for Healthcare Research and Quality. Percent of private-sector employees eligible for health insurance that are enrolled in health insurance at establishments that offer health insurance by firm size and selected characteristics (Table I.B.2.a.1), years 1996-2002: 1996 (Revised March 2000), 1997 (March 2000), 1998 (August 2000), 1999 (August 2001), 2000 (August 2002), 2001 (August 2003), 2002 (July 2004). Medical Expenditure Panel Survey Insurance Component Tables. Generated using MEPSnet/IC. <http://www.meps.ahrq.gov/mepsnet/ic/mepsnetic.asp> (December 09, 2004) Chollet, D. 2004. What have we learned from research on individual market reform? in Alan C Monheit and Joel Cantor ed. State Health Insurance Market Reform. Routledge. London and New York. Fronstin, Paul. Employee Benefits Research Institute (EBRI), 2004. Notes. October 2004, vol 25, no, 10. available at http://www.ebri.org/ebri_notes_10-2004.pdf http://www.gao.gov/new.items/d028.pdf http://www.meps.ahrq.gov/papers/st46/stat46.pdf Simon, K. 2004. What have we learned from research on small-group insurance reforms? in Alan C Monheit and Joel Cantor ed. State Health Insurance Market Reform. Routledge. London and New York. US Census Bureau, 2004. Income Stable, Poverty Up, Numbers of Americans With and Without Health Insurance Rise, Census Bureau Reports Press Release. August 26 th.