Consumer Response to a National Marketplace for Individual Insurance. Appendix 1

06/28/2008

Full Description of Simulation Methods

This analysis was completed in three steps.  First, an inventory of available literature was completed to identify parameters for the simulation.  Second, we reviewed the literature and used empirical data to develop premium estimates for the simulation that reflect case-mix as well as state-specific differences.  Third, we used a revised version of the 2005 Medical Expenditure Panel Survey (MEPS) to complete a set of simulations to identify the impact of three different scenarios for a proposal to develop a national insurance market.

Characterize the state-specific individual insurance markets

(a) The first step in this simulation is to describe the regulatory environment of the individual insurance market in each state.  We used several secondary sources for this description, including Blue Cross/Blue Shield for state mandates; the Georgetown University Health Policy Institute for guaranteed issue and community rating; and Thomson-West’s Netscan/Health Policy Tracking Service (“Major Health Care Policies, 50 State Profiles, 2003/2004”) for any willing provider laws.  We attempted to be as consistent as possible by using the same sources of regulatory information used in the empirical work from which we take our cost estimates.  This was challenging because some of the studies failed to provide reference information.  This information was coded into a spreadsheet for use in subsequent steps of the analysis and is presented as Appendix 3.

(b) The second step is to identify the marginal cost of particular regulations, including mandates, guaranteed issue, community rating, and any willing provider laws.

  • Mandates are state regulations that require insurers to cover particular services or providers.  We opted to use the count of mandates in a state rather than trying to identify the separate cost of each mandate.  This decision follows the empirical work, which typically uses a count of state mandates.
  • Guaranteed issue laws require insurers to sell insurance to all potential customers regardless of health or pre-existing conditions.  However, this doesn’t necessarily mean that insurers can’t put riders on pre-existing conditions or incorporate premium adjustments for them.  Guaranteed issue provisions can be broad (e.g. applying to all products, all consumers, at all times) or narrow (e.g. applying to very specific populations or during specific open enrollment periods).  Our coding rules are biased towards those states that had fairly broad guaranteed issue provisions. 
  • Community rating requires insurers to limit premium differences across individuals.  We coded a state as having community rating if it had ‘pure’ (no premium differences are allowed) or ‘adjusted’ community rating.  We did not consider rating bands as part of this definition.
  • Any willing provider (AWP) laws restrict insurers’ ability to exclude providers from their networks.  There is a lot of variability here as well.  Many states apply AWP laws narrowly (e.g. to pharmacies only).  We coded a state as having an AWP law if it applied broadly to providers.

We conducted a literature review to identify estimates of the impact of these state laws and regulations on health insurance premiums.4 We used only studies of the individual insurance market, since this is the market of primary interest.  This ruled out using studies that focus on the relationship between regulations and premiums in the small-group market (e.g. Simon, 2005). 

States may adopt regulations for reasons that are also related to the effect of those regulations on premiums.  For example, a state may be ‘pro-regulation’ in all areas and that pro-regulation sentiment may enhance the effects of the regulations.  However, we could not find any study that controlled for states’ strong preferences for regulation.  This may be due to the fact that many regulations were adopted in the 1990s or before and there is no premium data that can be matched to ‘before’ and ‘after’ the regulations were implemented.  Because none of the studies controlled for self-selection, the results must be interpreted with caution.

Two studies (LaPierre, et al., 2005; Hadley and Reschovsky, 2003) analyzed the regulation-premium relationship using data on individuals who held health insurance policies.  People who hold insurance may have characteristics that differ from those who shopped and didn’t buy.  For example, those who hold insurance may be low-risk.  If these characteristics are not observed or controlled by the researcher, his or her estimates of the effects of regulations on premiums held by the insured will be biased.  We eliminated the LaPierre, et al. (2005) study because they did not attempt to control for this bias.  We retained the estimates from Hadley and Reschovsky (2003) since they used a selection-correction approach to control for unmeasured personal attributes related to both insurance and premiums. 

We utilized estimates from the following four studies: Congdon, et al. (2005); Henderson, et al. (2007); New (2006); and Hadley and Reschovsky (2003).5 It should be noted that only the Hadley and Reschovsky (2003) paper has been published in a peer-reviewed journal.  The other three are working papers.  In Table A1, we summarize the key findings:

Table A1 Summary of Studies of the Effects of Sate Regulations on Premiums in the Individual Health Insurance Market

Regulation/Law Congdon, et al. Henderson, et al. New Hadley & Reschovsky
Guaranteed Issue 94-114% increase in premium in one state (NJ) No effect NA (not assessed) No effect
Community Rating 20-27% increase in premium No effect  NA 15-34.6% increase in premium
Any Willing Provider 1.5-9% increase in premium 5-12% increase NA NA
Mandates Each additional mandate increases premium .4-.9%. Used indicator variables for a very comprehensive set of mandates.  Some increase and some decrease premium. Each additional mandate raises the monthly premium by 75 cents, approximately .5%. NA

To make our analysis comprehensive, we used three summary measures of the regulatory effects: (1) the midpoint of the range of the estimated effect of each regulation/mandate – our moderate estimate; (2) the minimum estimated effect; and (3) the maximum estimated effect.  These effects are summarized in Table A2.\

Table A2 Minimum, Maximum, and Midpoint Estimates of the Effects of Regulations

Regulation Minimum Increase Midpoint Increase Maximum Increase
Guaranteed Issue 0 57% 114%
Community Rating 0 17.3% 34.6%
Any Willing Provider 1.5% 6.75% 12%
Mandates .4% per mandate .65% per mandate .9% per mandate

 

Regulations and mandates represent important differences across state-specific individual insurance markets, but there may be other factors as well.  Here are a few issues:

(a) Regulations regarding look-back periods and pre-existing conditions: A lot of variation exists across states with respect to mandates regarding coverage of pre-existing conditions.  This will impact people with chronic/acute illnesses differently than those who are healthy, both in terms of coverage value, prices (potentially), and take-up.  Although we have information on state regulations for look-back periods and pre-existing conditions, we know of no studies that model the effect of these regulations on premiums.

(b) Premium taxes: We have not attempted to determine the effects of premium taxes on premiums in the non-group market.

(c) Provider networks and provider prices: Premium variation may also reflect differences across states (and plans within states) regarding the size of the provider network and plan types.  AWP laws may capture some of this variation, but the extent of provider market power and local variation in prices is also likely to drive premiums.

Calculate simulation premiums

The second step in the analysis requires the calculation of premiums adjusted for the effects of state regulations.  The basic idea behind a national market is that a person living in State A will be able to buy insurance licensed in State B.  Suppose I live in State A where the premium is $100 per month.  This reflects the influence of my state’s medical practice style and provider prices (which would not change if I bought insurance in State B) and the effects of regulations (which would change).  If I bought insurance in State B, the premium would be $100 minus the effects of fewer regulations in State B.   
 
To implement this step, we relied on the premiums reported by Congdon, Kowalski, and Showalter (2005).  These premiums were first adjusted by age and sex to reflect standard actuarial differences in health care costs, and then adjusted by the effects of regulations as summarized in Appendix 3.  The adjusted premiums will be used as inputs into the insurance take-up simulation model.

Simulation

In the third step we simulate the effect of a national market on take-up of individual health insurance.  This step requires that we know the state of residence for people in the MEPS-Household Component, (MEPS-HC), but the MEPS will not release person-specific state IDs.  Therefore, we had to devise a method for imputing each person’s state of residence.

State-Specific Imputation of MEPS

Below, we summarize the process of imputation which resulted in the creation of 51 synthetic state populations from the 2005 MEPS-HC.

(a) We used the 2005 American Community Survey (ACS) to define the strata that would be used to generate the sample.6 The final strata include four variables:  Age (18-34, 35-44, 45-54, and 55-64); Income (1 if household income is in the lowest quartile, 0 if not); Male (1 if male, 0 if not); White (1 if white, non-Hispanic, 0 if not).  Creating all possible combinations resulted in 32 cells per state.  The unit of analysis for data construction is the person, not the household.  Using person weights in the ACS, we tabulated the population frequencies for each of these strata by state.  

(b) We divided the 2005 MEPS into four regions – Northeast, Midwest, South, and West.  The District of Columbia is in the South region.  We selected only 18-64 year-olds to match the ACS selection criteria.  The regional MEPS samples had the following sizes:

Table A3 – 2005 Regional MEPS Sample Size by Region

Region Sample Size
Northeast 2,874
Midwest 3,734
South 7,520
West 5,132

 

Within each of these regions, the strata were defined.  We then wrote a STATA computer program to draw a random sample with replacement of 1,000 (approximately, given rounding) observations from the region containing a particular state.7 The frequency of observations by strata was matched to represent the population (e.g. if 10% of the state is age 18-34, low-income, male, and non-white, then 100 of the 1,000 observations would be drawn from MEPS individuals of this type).  After all of the random samples were drawn, the data were appended to form a national data set. 

(c) While we know that the state samples match the socio-demographic criteria with respect to the strata, additionally we wanted to check to see how our samples looked with respect to insurance holding.  To do this, we computed state-specific estimates of uninsurance from the 2006 Current Population Survey (CPS).  We compared the uninsurance estimates generated for our synthetic state populations with the CPS estimates.  This comparison fares pretty well.  There are only two notable issues: (1) we tend to underestimate the amount of uninsurance in synthetic Northeast states due to the small MEPS sample and the population heterogeneity in the Northeast; and (2) uninsurance was overestimated in Washington, DC, because the sample is drawn from the entire South region and there is no easy way to account for the concentration of federal government workers in DC. 
 
(d)  After completing this exercise, we merged several other variables into the file and selected the sample to mimic the one we have used previously in simulations.  In particular, we deleted cases of adult dependents who did not have an ESI offer but had a spousal offer (n = 8,609), those who reported having public insurance at any point during round 1 of MEPS (n = 4,725), and full-time students (n = 892).  Also, we constructed the number of plans offered to each person by using an ordered probit model to predict whether those with an offer of ESI were offered 1, 2, 3, or 4+ plans.  We computed predicted probabilities for each category and identified the category with the maximum probability as the number of offered plans. 

Application of State-Specific MEPS to National Simulation Model

Using a simulation model developed from previous analyses (Feldman, Parente, Abraham, et al, 2005; Parente, Feldman and Abraham, 2007), we applied the Synthetic State MEPS (SS-MEPS) described above to develop a set of national estimates.  The simulation model is capable of generating estimates of national health plan take-up for both the individual and the ESI markets.  The estimates are based on predictions from a set of parameter estimates from a conditional logistic regression model of health plan choice.  The conditional logistic regression model requires information on wage income, single or family status, presence of chronic illness, age, gender, and health plan premiums.  The data used to generate the parameter estimates come from an aggregate database of large employers’ human resources and claims data from 2003.

One of the distinguishing attributes of the simulation model is the presence of consumer driven health plans (CDHPs).  Specifically, there are two types of CDHPs: a low-option Health Reimbursement Arrangement (HRA) and a high-option HRA.  The low-option HRA is very similar in deductible, coinsurance and premium structure to a Health Savings Account (HSA) plan.  This enabled us to model both HRA and HSA choices in the simulation as well as high, moderate and low-option Preferred Provider Organizations (PPOs), and a Health Maintenance Organization (HMO).

In the simulation, consumers in the individual market have five choices: high, moderate and low-option PPO, HSA, and the choice to be uninsured.  The uninsurance parameter is calibrated based on the national rate of the uninsured in the individual market by income quartiles as determined from the 2005 MEPS sample.  Consumers with employer-sponsored coverage are given up to eight choices including HMO, three PPO options, an HRA, an HSA where the employee opts out of employer sponsored coverage, an HSA where the employer picks up most of the cost of the HSA/high deductible insurance policy, and finally a choice to turn down coverage for any reason (e.g. already had coverage from spouse).

Chronic illness is modeled at the contract level in the simulations.  That is, either the person choosing insurance, or someone covered by their insurance contract, has a chronic illness.  This assumption was made because the data used to estimate the health plan choice model could only be attributed to contract holders, not the person receiving care under a contract.  As a result, the chronic illness metric reflects a household’s illness burden, more than that of one individual, unless the person is only buying a single-coverage contract.

The simulation model adjusts premiums for the tax treatment of health insurance offered by employers in the ESI market.  Specifically, premiums are adjusted to take into consideration the federal marginal tax rate as well as the social security tax burden.  The capability to adjust for state tax effects is also possible, but not considered in this model in order to identify the pure effects of differences in insurance regulation by state.

We use premium estimates for each of the plan choices based on our earlier work (Feldman, Parente, Abraham, et al., 2005).  These premium estimates are derived from a combination of ehealthinsurance.com and Kaiser/Commonwealth estimates of premium prices.  These premium estimates are adjusted to 2008 dollars. 

We develop state-specific premium inflators/deflators from the AHIP individual market single and family coverage report.  Individual market premiums were experience rated for age and gender (with the exception of community rated states).  For this analysis, we define the small group market as one where an employer has less than 250 employees.  At this level, employers generally do not self-insure.  Premiums for employers with less than 250 employees were adjusted by state-specific regulatory effects.    Finally, HSA premiums include a $1,000/$2,000 investment in accounts depending upon whether the person was choosing a single or family insurance product, respectively.

The simulation is based only on choices made by adults aged 19-64 who are not students, not covered by public insurance, and not eligible for coverage under someone else’s ESI policy.  As a result, our baseline uninsured and turned down population represents 32.3 million people (excluding military, students, persons under age 18 or 65 and older, and those without an ESI offer who could be covered by a spouse).  However, we present results for our selected sample as well as a national approximation that would yield 47 million people uninsured.

Scenarios for Policy Simulation

We developed three different scenarios for policy simulation.  Each of these simulations was run on a set of minimum, moderate and maximum impacts of state-specific regulations as derived from the literature.  The impact of each scenario was calculated by multiplying a given person’s original premium by a state min/mod/max specific multiplier.  These multipliers are described in Appendix 4 by state.  For each scenario, if the consumer faces a lower premium as a result of the proposed policy change, the consumer will choose the better price.  If the new possible premium is not a better deal than that in the consumer’s home state, they will stick with their home state in the simulation.  The three scenarios are:

Scenario 1: Competition among 5 largest states

In this scenario, only the five largest states are permitted to be available for the national market along with the consumer’s own state.  This scenario was considered in a previous legislative proposal with the rationale that large states would have the critical skills in their insurance departments to take on additional regulatory responsibilities for new consumers from out-of-state.  The five largest states in the United States, based for population size, are (in order of descending population size): California, Texas, New York, Florida, and Illinois.  Of these, Texas has the least regulated health insurance environment and is the comparison state in the simulations.

Scenario 2: Competition among all 50 states

For this scenario, the state with the least regulation is identified as Alabama.  In this simulation, all consumers are assumed to find AL the state to which they would switch policies unless they were already residents of Alabama.  This could be the most extreme outcome of the legislation similar to that proposed by Rep. John Shadegg (R-AZ) for the last few years.

Scenario 3: Competition within regions

Under this scenario, the United States’ health insurance market is broken into four regions: Northeast, South, Midwest, and West.  Residents in each region buy insurance from a state within their region with the most favorable premium due to decreased regulation.  This scenario was based on the regional Part D and TriCare contract models for insurance carriers.  For the Northeast, the state with least-cost regulatory impact was New Hampshire.  In the Midwest, Nebraska was the favored state.  In the West, the state of choice was Arizona and in the South, the state of choice was Alabama.

View full report

Preview
Download

"report.pdf" (pdf, 144.12Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®