Methodological Description of Health Care Reform Premium and Discount Estimates Contributions to this paper were made by: * Department of Health and Human Services' Office of the Assistant Secretary for Planning and Evaluation * The Office of Management and Budget * Health Care Financing Administration's Office of the Actuary * Agency for Health Care Policy and Research * The Urban Institute's Center for Income and Benefits Policy Contents I. Background II. Description of the Major Models A. The Urban Institute's Transfer Income Model (TRIM2) B. The Health Care Financing Administration's Special Policy Analysis Model (SPAM) C. The Agency for Health Care Policy and Research's Simulation Model (AHSIM) III. Premium Estimation Under Reform A. Health Care Financing Administration B. Agency for Health Care Policy and Research C. Choice of Premium Estimates for Budgeting Purposes IV. Discount Estimates A. Employer Discounts 1. TRIM2: Employer Discounts 2. HCFA: Employer Discounts 3. AHSIM: Employer Discounts B. Discounts for the Self-Employed C. Discounts to Low Income Families D. Retiree Discounts E. Choice of Discount Estimates for Budgeting Purposes and Distrbutional Analyses V. National Spending Impacts Appendix A: Example of the Application of an Induction Factor to a Change in Insurance Status I. Background Estimates of premium costs, national health spending, and government program costs under health care reform have been necessary in the decisions leading to a health care reform bill. During the basic policy development process, exploration of alternative policies required estimates of the cost impacts of each possible variation. Specific areas included analyses of premium caps, the impacts on businesses of reqiured employer payments, the effects on households of required purchase of coverage, and the budgetary effects of the discount schedules. The development of estimates of this type is obviously a complex task. Given that no single authoritative data set exists which captures all spending for all services through all sources of funding, numerous data sources were used in this process. Federal surveys, especially the National Medical Expenditure Survey (1987), offer the best characterizations of national spending. The National Health Accounts generated by the Health Care Financing Administration (HCFA) summarize the best available data on total national spending by type of service and source of fund. Producing estimated spending under health reform, however, requires developing a comprehensive baseline summary for literally hundreds of affected sub-populations, and then estimating the future spending patterns associated with the reform. Estimates of future costs of reform are primarily derived through modeling transfers of current spending among the various channels of payment. Estimating the impacts of changing primary payers is relatively straightforward, given a baseline of national health spending. More difficult is estimating the net impacts of fee upgrades and paying for uncompensated care, since reimbursement levels will be set to achieve some amount of recapture of these increased outlays for current services. Also difficult is estimating the induced spending attributable to new or enriched insurance coverage. Estimates have been based upon experiences of government and private insurers as well as the results of academic studies of the demand for medical care. Multiple data sources, methodologies, and models were needed to produce estimates of premiums, discounts, and the overall effects of reform options. Major contributors included the Health Care Financing Administration's (HCFA) Office of the Actuary (OAct), the Agency for Health Care Policy and Research (AHCPR), the Treasury Department, and other government agencies. Numerous consultants assisted in the process, with major modeling contributions provided by the Urban Institute. This paper provides an overview of the major models and their ways of estimating premiums, discounts, and overall health spending under health care reform. II. Description of the Major Models A. The Urban Institute's Transfer Income Model (TRIM2): The Urban Institute has developed a microsimulation model called the Transfer Income Model (TRIM2). This model has been used to analyze the financing of national health care reform plans, and has particularly focused on the distributional effects of such proposals. TRIM2 is based upon the March 1992 Current Population Survey (CPS) and combines data from a number of other sources in order to provide a complete basis for assessing acute care health spending by the non-elderly in the U.S. population. [Historically, TRIM2 has been used to analyze current and alternative tax and transfer programs. See National Research Council (1991) and Lewis and Michel (1990) for a more complete description of the properties of this model and its recent applications.] The complete model has been aged to 1994, and all results are presented in 1994 dollars. The following description of TRIM2 and its capabilities has been adapted from Zedlewski, Holahan, Blumberg, and Winterbottom (1993). The TRIM2 model simulates the employer-based group health insurance system, nongroup or individually purchased health insurance, out-of-pocket spending, and the Medicaid program. The model assigns spending under these programs/systems at the individual and family levels and adjusts for regional variation in premium levels. It is then possible to assess the distributional effects of the financing of the current health care system. Detailed tax calculations allow the analysts to examine health spending on an after-tax basis and to calculate the after-tax value of employment-based health benefits. TRIM2 can also be used to simulate the distribution of health spending and health care financing burdens under alternative assumptions about how insurance would be provided and financed. Each component of the basic model is presented below. [Giannarelli (1992) describes the full TRIM2 model.] 1. Employment-Based Group Health Insurance. TRIM2 examines reported insurance coverage for individuals and families on the CPS to determine the number of family members who share coverage under an employment-based policy. [Zedlewski (1991) describes this model in more detail.] The model assigns an employment-based health insurance premium, including the shares paid by the employer and the employee, to each covered worker based on two 1989 private, employer-based surveys (from the Health Insurance Association of America and Foster-Higgins) and federal health insurance plan documents. These private surveys represent firms of different sizes, in all major industries (including state and local government), and in all regions of the country. Federal health insurance plan documents include information about premiums for single and family coverage and how these premiums are distributed between the employer and worker. TRIM2 statistically matches workers to health plans based on variables that the employer insurance plans and workers on the CPS have in common. These include the type of coverage (single or family), location, industry, the size of firm, and whether or not the worker has to pay part of the insurance premium. 2. Private, Nongroup Health Insurance. TRIM2 estimates premiums for the families and individuals on the CPS who report insurance coverage through private, nongroup, health insurance policies. It does this by matching these people with plan data collected from Blue Cross and Blue Shield Plan offices across the U.S. Plan documents included premiums for typical health insurance plans covering single individuals, families, and dual (adult and child) insurance units in each state. The method does not, however, capture differences due to families' insurance preferences or income levels. For example, if low (or high) income families reporting private, nongroup health insurance are more likely to purchase catastrophic policies, the model will overstate their premiums. Conversely, the model will understate premiums for families who prefer broader coverage than that included in the prototypical plan. 3. Medicaid. TRIM2 uses detailed sets of algorithms to replicate the rules of state Medicaid programs. [See Holahan and Zedlewski (1989) for a full description of this model.] These algorithms identify Medicaid eligibles as all persons who meet the states' categorical, asset, and income criteria in effect July 1991. The model has procedures for selecting Medicaid enrollees from those who are eligible. The second part of the model imputes the insurance value of Medicaid. Separate estimates are made for adults, children, and the disabled; estimates also vary by age, sex, race, urban or rural residence, reason for enrollment, and number of months in the program. The model uses Medicaid state expenditure data to adjust for differences among states in program generosity and the cost of health services. 4. Out-of-Pocket Spending. The model uses data from the Consumer Expenditures Survey to predict out-of-pocket spending (other than health insurance premiums) for families on the CPS. [See Wade (1991) for a full description of this model.] Separate equations were estimated for persons with private insurance coverage, Medicaid, and for those uninsured. The equations predict the incidence and levels of spending as a function of families' socioeconomic characteristics including region of residence, income, the age-sex distribution of family members, and the family head's marital status, education, race, and work status. 5. Income and Payroll Taxes. The model calculates family disposable income and estimates the amount of income and payroll taxes required to finance health care spending by the federal government through the Medicaid and Medicare programs. Other federal taxes (such as corporate, estate, and excise taxes) are not included. [However, the TRIM2 model does have the capacity to simulate excise tax payments on alcohol and cigarettes.] The portion of Medicaid and Medicare spending that is financed through the federal personal income and payroll tax systems is calculated and can be allocated to families. The model can also calculate income and payroll taxes under the assumption that employer-paid health insurance premiums are taxable to estimate the tax value of this employee benefit. 6. Total Health Care Spending for the Nonelderly. The TRIM2 baseline distribution of direct spending from various sources accounts for most health spending for the nonelderly. The TRIM2 model excludes the institutionalized population. In addition, Medicare benefits for the nonelderly and military health benefits are excluded. Nevertheless, the model accounts for nearly all of the spending under systems that would be most affected by health care reform alternatives currently under debate. 7. Adjustments to TRIM2 Baseline Output. Two significant adjustments were made to the TRIM2 baseline health spending simulations at the request of the office of the Assistant Secretary for Planning and Evaluation (ASPE), Department of Health and Human Services (DHHS). Both employer health insurance premiums and private nongroup health insurance premiums in TRIM2 were downwardly adjusted to reflect health care reform premiums estimated by the Office of the Actuary (OACT). The TRIM2 employer plan data include employer spending for dental care and a few other benefits that not fully covered by the reform package. Without this adjustment for differences in coverage, employer and family spending under reform as calculated in TRIM2 (using HCFA OACT premiums) would not be comparable to employer and family spending under the current system according to TRIM2. Thus, estimated changes in spending under reform compared to current spending would be distorted without reconciling the spending levels. B. The Health Care Financing Administration's Special Policy Analysis Model (SPAM): The Health Care Financing Administration's (HCFA) Special Policy Analysis Model (SPAM) database is also based upon the March 1992 Current Population Survey (CPS). The March 1992 CPS acts as the host file, with each person on it being statistically matched to a person on the 1987 National Medical Expenditure Survey (NMES). Health expenditures and utilization from the NMES person record are then linked to the CPS record, and the entire data set is controlled to be consistent with 1994 National Health Account data. The parameters used in the linking the NMES file to the CPS were disability status (disabled or not), age and gender (male adult < 19, male adult 19-44, male adult 45-64, male adult 65+, female adult < 19, female adult 19-44, female adult 45-64, female adult 65+, dependent child < 19, and dependent child 19+), family income (family under 100% of poverty, family 100% to 185% of poverty, family at or above 185% of poverty), and insurance class of the person (employer sponsored insurance and Medicare, employer sponsored insurance and Medicaid, other employer sponsored insurance, Medicare and Medicaid and other private insurance, Medicare and Medicaid, Medicare only, Medicaid and other private insurance, Medicaid only, other insurance, and uninsured). For a CPS person to be considered disabled, one of the following situations had to be true: (a) person was a veteran, collecting veteran's disability, (b) person collected over $10,000 in workers compensation, (c) person had disability income, or (d) person was under 65 and had SSI income. For a NMES person to be considered disabled, one of the following had to be true: (a) person was a disabled veteran, (b) person didn't work due to disability/illness, or (c) person was without a job due to disability/illness. The determination of adult vs. child was made from "insurance families," which were appended to both the CPS and NMES files. These families were created using standard insurance industry definitions. Within an insurance family there can be one or two adults (single or married couple), and any number of dependent children (or none). The insurance classifications were hierarchical, and made on the person level using the appropriate variables on both the CPS and NMES. NMES insurance classifications were converted from round data [NMES-2, described more fully in the following section, involves 4 surveys of each family over a period of 16 months. Each of the 4 interview sessions is referred to here as a "round."] to "ever insured" data (for example, if a person had Medicare in one of the rounds of the NMES survey, they were coded as having Medicare). Poverty classifications were calculated by adding income of all insurance family members together and comparing it to the appropriate poverty standard for the year in question (1991 for the CPS and 1987 for NMES), for the appropriate family size. Once each CPS person was linked to a NMES record, expenditure data by service (hospital inpatient, hospital outpatient, etc.) and source of payment (out-of-pocket, private insurance, Medicare, Medicaid, etc.) were attached. This file was then aged to 1994 through two steps. First, the 1992 CPS population was weighted to sum to the 1994 Social Security Administration (SSA) non-institutionalized population (about 20 million more than Census estimates). This was done by age (20 age groups), gender and marital status (single, married, divorced, and widowed). Second, the total national health expenditures by this SSA-weighted CPS population (the SPAM population) were then "benchmarked" by service category, channel of payment, and age category to the aggregate totals in the projected 1994 National Health Accounts. There are 13 service categories: hospital inpatient, hospital outpatient, hospital emergency room, physician inpatient, physician outpatient, physician emergency room, physician office visit, other professionals, prescription drugs, home health care, dental, vision, and other durable medical equipment. There are eight channels of payment: private health insurance, out of pocket, Medicare, Medicaid, other federal, other state and local, workers compensation, and other private. There are three age categories: under 19, 19 to 64, 65 years and over. An example of how this benchmarking worked is as follows. Suppose the ratio of current SPAM out-of-pocket inpatient hospital spending for persons under age 19 to NHA-consistent spending for the same cell is 0.9. Then each SPAM person's inpatient hospital spending total is multiplied by 1.11. The only divergence from this logic was that out-of-pocket spending for the uninsured was controlled separately from out-of-pocket spending for the insured population (which was thought to have risen at a rate closer to the rate of inflation in insurance). C. The Agency for Health Care Policy and Research's Simulation Model (AHSIM): AHSIM is based on AHCPR's 1987 National Medical Expenditure Survey (NMES-2), which is the most recent national effort to collect comprehensive, person-level profiles of health care use, spending, and insurance coverage. AHSIM currently is designed only to analyze the nonelderly (under 65), noninstitutionalized civilian population residing in the United States. Although the NMES-2 data were collected in 1987, demographic variables have been aged forward by reweighting individual records. New weights take into account changes in the distribution of the population by age, race, sex, insurance status, and poverty status observed between the November 1987 and March 1992 Current Population Surveys. Additional demographic aging is based on Census projections of the population by age, race, and sex beyond 1992. Real growth in service-specific health expenditures and insurance premiums have been incorporated through adjustments based on the appropriate rates of changes in HCFA's National Health Accounts and its projections. AHSIM draws primarily on the NMES-2 Household Survey and its two derivative components, the Health Insurance Plan Survey (HIPS) and the Medical Provider Survey. The Household Survey sample is representative of the civilian noninstitutionalized population of the United States in 1987. Each family in the Household Survey was interviewed four times ("rounds") over a period of 16 months to obtain information about the family's health and health care during calendar year 1987. Roughly 35,000 individuals and 14,000 households completed all rounds of data collection. The Medical Provider Survey obtained information directly from the physicians, hospitals, and other providers used by a portion of the household sample. These data were used to edit and supplement household survey data describing use of and spending on health services. HIPS data were collected from employers, unions, and insurers and include premiums paid by all sources and specific provisions of baseline private insurance coverage. They also provide information about the organizations offering insurance coverage and include in the case of employers, firm and establishment size, industry, and location. Other data sources were incorporated when needed for specific purposes. For example, survey data from the Health Insurance Association of America were used to project market shares for fee-for-service, HMO, and preferred provider health plans by region. Annual survey data from the American Hospital Association were used to determine the allocation of hospital spending between inpatient and outpatient services and to identify local areas in which at least one HMO is operating. County Business Patterns data were used to impute average payroll for employers, using a statistical match based on industry, location, and firm size. The Internal Revenue Service (IRS) Statistics of Income (SOI) data were used to expand NMES-2 income data and to calibrate the AHSIM tax module. Further details of the basic AHSIM Model are presented below. 1. Employment-Based Group Health Insurance. AHSIM does not model employers, only households and individuals. HIPS and Household Survey data measured the scope of employment-based insurance in 1987, as well as the allocation of premiums among employers, employees, and other sources (primarily unions). AHSIM assumes that any changes in the pattern of availability, benefits, premiums, or other plan provisions between 1987 and 1994 are controlled for in the aging process, using CPS and HCFA aggregate data. 2. Private, Nongroup Health Insurance. AHSIM handles people with private, nongroup insurance in the same way as it does those with employment-based insurance. This means, for example, that a tendency for people with systematically higher medical expenditures to purchase private, nongroup health insurance will be captured in the model. 3. Medicaid. NMES-2 measured Medicaid program participation directly. Since 1987, however, Medicaid eligibility has been expanded to include low-income pregnant women and children who are not otherwise categorically eligible. This Medicaid expansion was incorporated into the ASHIM model by identifying household survey respondents who would have become eligible for Medicaid benefits by 1994 and modifying their baseline insurance status accordingly. Baseline health spending for new Medicaid recipients was also modified to reflect the effect of a changes insurance status, using the same methods that were used to project estimates for the uninsured after reform. 4. Out-of-Pocket Spending. NMES-2 directly measures baseline out-of-pocket (OOP) spending on cost-sharing and noncovered services for all household survey respondents. However, because the AHSIM model cannot use actual NMES-2 expenditure data directly, data on OOP spending reported in the household survey are used to develop a set of estimates that can be incorporated into the model. [Actual NMES-2 expenditure are not used as baseline spending in the AHSIM Model for two reasons. For some people, e.g., those affected by the expansion of Medicaid eligibility, baseline insurance status is different than what is was in 1987 when the NMES data were collected. In addition, it is important that reform expenditures only differ from baseline for reasons related to reform. Because reform expenditures must be predicted for people based on their "new" insurance status, it is helpful to predict baseline spending with the same methodology] In particular, a system of equations was estimated to predict the percent of expenses paid out-of-pocket in the baseline as a function of demographic characteristics, insurance coverage, health status, and other relevant explanatory variables. These equations were then used to impute baseline OOP spending as a percent of imputed total spending, by type of service. While baseline expenditures are imputed in the AHSIM Model, they are still internally consistent with the rest of the NMES-2 data because the estimation procedures preserve the pattern of spending observed in the original household survey data. 5. Income and Payroll Taxes. The AHSIM model distinguishes between households and tax filing units. The effect of wage changes induced by employer mandates on federal personal income and payroll taxes can be calculated with respect to the 1991 tax treatment of employer paid premiums. 6. Total Health Care Spending for the Nonelderly. The AHSIM baseline includes most health spending by the civilian, resident population of the United States under the age of 65. AHSIM excludes the institutionalized population, Medicare beneficiaries among the nonelderly, and spending by active-duty military personnel. 7. Adjustments to AHSIM Baseline Data. Because the aggregate health insurance premiums reported in the HIPS component of NMES-2 are not consistent with the benefits paid by private health insurance according to either the Household Survey or the National Health Accounts, the NMES-2 HIPS-based premiums are calibrated to these other data sources. Tax estimates and wage income are calibrated to SOI data, as described above. In general, according to extensive analysis by HCFA and AHCPR staff to account for differences in definition and coverage, the NMES-2 expenditure data and the National Health Accounts yield similar estimates. III. Premium Estimation Under Reform Two agencies, the Health Care Financing Administration and the Agency for Health Care Policy and Research, estimated the cost of health insurance premiums under reform. Their estimates are in 1994 dollars and reflect the benefits included in the standard benefit package. The premium estimation methodology used by each of these agencies is described below. A. Health Care Financing Administration: The first step in HCFA's simulation process was to determine each individual's insurance status. The modelers used CPS indicators for this, and considered a person to be insured if he/she was covered by employer-sponsored insurance, other private insurance, CHAMPUS, Medicare, or Medicaid. Insurance could be either in one's own name or through inclusion in a policy held by an adult in the insurance unit. Also, some dependents are covered by private insurance policies owned by people outside the family (for example, a child of divorced parents may be covered through insurance carried by the parent who does not live with the child). HCFA modelers then adjusted health expenditures to reflect the coverage offered through the regional alliance plan. That coverage is restricted to hospital care, physician and other professional services, prescription drugs, and durable medical equipment other than vision and hearing products. Therefore, the analysts excluded all other National Health Accounts expenditure categories. The cost of coverage for mental health, dental, and preventive care in the standard benefit package was estimated separately, from aggregate data, and added in at the end of the process. Once expenses were adjusted for coverage differences, the modelers applied the fee-for-service plan deductibles, coinsurance, and cost-sharing limits to each person covered through the regional alliance. An insurance-induced demand adjustment was applied to all those enrolled in the regional alliance. The basis for the induced demand was the difference between out-of-pocket spending under current law and that determined by the reform simulation described above. The induction factor varied by type of service. The application of the factors and the specific values used are described in appendix A. Post-induction spending is equal to the expenditures calculated previously plus (minus) the induced spending calculated as described. Following these steps, HCFA analysts imputed expenses to currently uninsured people. Existing patterns of use for the uninsured person were discarded, because those patterns are influenced by the absence of insurance. An imputation file was created for each service covered under the regional alliance. To create the file, insured people (excluding people who received SSI cash payments) were divided into groups according to gender, four age classes, and three poverty status classes. Expenditures were tabulated for each group to determine: (a) the proportion that had no expenditure and (b) mean expenditures and use for each decile of the user distribution. Expenses were imputed for an uninsured person using these imputation files. For each type of service, the person was assigned a random number ranging from 0 to 1. If that random number fell within the nonuser proportion for the service, the person was given no expenditure for the service. Otherwise, the person was given the mean expenditure and use for the decile of users into which the random number placed them. Analysts assumed that facility and physician use was correlated for hospital services, and used the same random number for hospital inpatient and physician inpatient use. They did the same for hospital outpatient and physician outpatient, and for hospital emergency room and physician emergency room use. Analysts performed a final simulation to determine which people were covered by the alliances. Typically, they excluded people who received AFDC or SSI cash payments. Similarly, most Medicare enrollees were excluded; only those who worked or whose spouse worked were included in the premium calculations. The remaining people were divided between the corporate alliance and the regional alliance according to the worker status of the adults in the insurance family, and were assigned to one of three policies: individuals (and couples with no dependents), one adult plus dependents, and two adults plus dependents. In a final pass through the family's health expenditures, analysts applied the family limits on out-of-pocket spending to determine the plan benefits and copayments. In order to generate an upper-bound discount estimate, whenever a two-earner couple had one worker in a large firm (5,000 or more workers) and one in a firm that would be covered through a regional alliance, the couple was assumed to choose coverage in the regional alliance. This maximizes the potential cost of the discounts costs given that no government discounts are available through the corporate alliances. After plan benefits had been determined, premiums were calculated for each of the policy types and alliance types. An offset was applied to expenses to reflect current-law cost-shifting attributable to uncompensated care. Under the current system, private sector premiums are higher than they would be if there were no uncompensated care in the system since providers pass these unpaid costs on to insured, paying patients. Under reform, all persons will be insured; consequently, baseline premiums should be reduced to reflect the elimination of non-payers from the system. A load factor was applied to the (reduced) benefit cost per policy. The load factor was 15 percent for the regional alliance. B. Agency for Health Care Policy and Research: AHCPR's method of generating premium estimates has seven steps. First, following conventions in health economics, AHSIM estimates a two-part model of expenditures for each service. The unit of observation is the person. The first equation in each service's set of two equations estimates the probability of using the service at all as a function of demographic, income, insurance, employment, and health status measures from the 1987 NMES-2. The second equation estimates annual expenditures on the service for all users of the service, as a function of the same explanatory variables. Combining the result of these equations (i.e., multiplying the probability of use times the coefficients in the second equation) yields an equation that predicts expenditures for each type of person. Predicted expenditures are aged to 1994. Health expenditures for each person are then predicted for each of the ten services included in the AHSIM Model using this system of equations. Predictions for both the probability and the level (given any use) of an expense were made for each person based on these regressions. The procedure assigns the same expected values to people with private insurance and similar personal characteristics, based on a hypothetical "average" insurance policy. Expected values are modified to take into account specific plan provisions using information from the RAND National Health Insurance Experiment about the effects of such provisions. Reform expenditures are imputed to all people in the model using a stochastic process that maintains observed correlations in expenditures across service types while controlling for the demographic characteristics and health status of individual NMES-2 respondents. Every individual included in the AHSIM Model actually had three types of reform expenditures assigned to them, indicating their (assumed) behavior under fee-for-service (FFS), managed care (HMO), and preferred provider (PPO) insurance arrangements. Expenses for benefits paid, cost-sharing and noncovered services were calculated separately for each type of plan by applying claims-processing logic to the appropriate estimated expenditure. Premiums for each type of insurance plan were computed on the basis of average benefits paid per insurance policy plus an administrative load set at a percent of benefits paid. In this way, each person was taken into account in computing initial premium levels. Premiums were adjusted for current regional variations in prices. Individual choice of health plans under reform was modelled by randomly assigning health insurance units to one of the three types of plans (FFS, HMO, PPO) described above. The assumed probabilities of selecting particular plans were based primarily upon market shares observed by HIAA in their annual surveys, trended forward to 1994. These estimates were modified by assuming a 10 percent reduction in FFS under reform as a result of managed competition. Market shares were allowed to vary on the basis of region, urban/rural location, and the availability of discounts for out-of-pocket (OOP) expenses and premiums. Two passes through the data are made to compute the final set of premiums. The first pass implements decision rules regarding the distribution of premium payments under reform. It also computes the cost of noncovered services and cost-sharing requirements borne by individual households. Based on these calculations, the model determines the extent to which a household's direct costs will be offset by supplemental insurance and OOP discounts. In the second pass through the data, expenditures are increased to reflect additional spending induced by supplemental insurance and OOP discounts. Insurance premiums are then adjusted to reflect these higher expenditures. C. Choice of Premium Estimates for Budgeting Purposes: One set of premium estimates had to be chosen for final budgeting purposes. Although AHCPR's premiums were used by that agency in their estimation of discounts to employers and households and those discount estimates were used as a check on estimates done by HCFA and the Urban Institute, the Administration opted to use the HCFA premiums for purposes of final federal budgeting and distributional effects analyses. This choice was made for two reasons. First, the premiums estimated by HCFA were higher than those estimated by AHCPR, and it was viewed as desirable to have an official estimate that was more conservative (i.e., that would lead to higher costs associated with the program -- see Table 1 below). Second, the HCFA estimates are benchmarked to the National Health Accounts, the most reliable measure of aggregate spending in the current health care system. Given that the National Health Accounts are considered to be the "gold standard" in measuring total health expenditures, it seemed most appropriate to keep the official premium estimates consistent with that standard. Table 1 Alliance Premium Estimates Policy Type: HCFA AHCPR Single $1933 $1735 Couple $3865 $3471 One Adult Family $3894 $3647 Two Adult Family $4361 $4262 IV. Discount Estimates The national health care reform plan includes a number of different discounts, targeted at different payers. There are two employer discounts: one directed at all firms in the regional alliance, and one directed at small firms with less than 75 employees. There is a discount for the family share (20 percent of the actuarial value) of premiums and for out-of-pocket payments for both working and nonworking low income families. There is also a discount for the 80 percent premium share for those families who do not have at least one full time worker (or equivalent), including early retirees. The major models are similar in how they estimate most components. A. Employer Discounts: The general firm discount consists of a 7.9 percent of payroll cap on all firm premiums, regardless of firm size, provided the employer is in the regional alliance. If the cost of providing 80 percent of the adjusted premium per worker exceeds 7.9 percent of firm payroll, the share paid by the federal government is equal to the difference between the two amounts, or: N sub S (.8P sub S)+N sub C (.8P sub C)+N sub {SP}(.8P sub {SP}) +N sub {DP} (.8P sub {DP})] - (.079*firm payroll) where N is the number of workers of each contract type (S=singles, C=couples without children, SP=single parent families, and DP=dual parent families) and P is the adjusted per worker premium for each contract type. The small firm discount schedule provides lower payroll caps (less than 7.9 percent) for firms with less than 75 employees and average pay below $24,000 per year. The small firm schedule is shown in Table 2. Table 2 Small Firm Discounts Average Firm Payroll Size of Firm (Number of Employees) Less Than 25 25 to 50 50 to 75 Less $12,000 3.5% 4.4% 5.3% $12,000-15,000 4.4% 5.3% 6.2% $15,000-18,000 5.3% 6.2% 7.1% $18,000-21,000 6.2% 7.1% 7.9% $21,000-24,000 7.1% 7.9% 7.9% Greater Than $24,000 7.9% 7.9% 7.9% [Because 75 workers was not a firm size break included in the data sets being used, modelers were asked to use a firm size of 100 for this subsidy calculation. Given that the subsidies will apply only to firms up to size 75, the results overestimate the subsidy costs.] 1. TRIM2: Employer Discounts. In the TRIM2 model, employer obligations (either 80 percent of the adjusted premium for each worker or a percent of total payroll) are calculated for each worker; there are no firms per se on the CPS, although each worker has employer information associated with them. TRIM2 assigns firm average payroll information from the County Business Patterns (CBP) data to each worker, using a statistical matching procedure that relies on industry (the 3-digit SIC codes), state of residence, and establishment size. In addition, an average firm premium is imputed to each worker. Take, for example, retail firms with 100-500 workers. Assume that according to the CPS, of the workers who report being employed by that type of firm, 40 percent are singles, 20 percent are married but have no children, 30 percent are married with children, and 10 percent are single parents. The weighted average firm premium that an employer of that type faces is equal to (.4P sub S) + (.2P sub C) + (.1P sub {SP}) + (.3P sub {DP}) where P_s, P_c, PSP, and P_dp are as described earlier. The employer's payment is proxied by the comparison of average pay times the appropriate percentage cap (3.5 percent to 7.9 percent) to 80 percent of the average firm premium. If the 80 percent of the average firm premium is less than capped average pay, the employer would pay 80 percent of the correct adjusted premium for each worker. If, on the other hand, capped average pay is less than 80 percent of the average firm premium, the employer would contribute 7.9 percent (or the appropriate percentage less than 7.9 percent) of total payroll to the alliance. If the firm cap is the less expensive option, the worker's record is appended with an employer payment equal to the appropriate cap times average pay in the firm. The amount paid by the federal government on behalf of the employer is also added to the record in the amount of: (.80P sub i)-(C AP*(Avg. Firm Pay)) where P is the adjusted per worker premium for the worker's health insurance unit type i (single, couple, single parent, dual parent), CAP is equal to the appropriate percentage cap (ranging from 3.5 percent to 7.9 percent) and Avg. Firm Pay is equal to the firm's average payroll as imputed from the CBP data. If, conversely, 80 percent of the adjusted per worker premium is the less expensive option, the worker's record is appended with an employer payment equal to .80*Pi, where i is equal to the appropriate health insurance unit type for that worker, and there is no government employer discount. 2. HCFA: Employer Discounts. In the SPAM model, the basic calculations of employer discounts are similar to those in TRIM2, other than the development of firm-level average payrolls. While TRIM2 imputes payroll data from the County Business Patterns data set, SPAM uses payrolls created by synthesizing firms from employees on the CPS. For each record of an employee on the CPS, one of the firms created using that employee is linked back to serve as the firm description for that employee. The resulting payroll distribution is similar to that implied by the CBP. 3. AHSIM: Employer Discounts. In the AHSIM model, the calculations are also similar to those in TRIM2. AHCPR uses County Business Pattern data for estimating average payroll. The links to NMES-2 make use of the Household Survey detailed responses by firm-size, industry, and other variables, confirmed by the NMES-2 Health Insurance Plan Survey. B. Discounts for the Self-Employed. Those individuals who are self-employed are obligated to make a contribution to the alliances based upon the same schedule used to determine small business payments. Those with self-employment income between $0 and $12,000 per year, for example, pay the lesser of 3.5 percent of self-employment income and: (.80*P sub i) - EC where P_i is as before and EC is the credit received by the self-employed person due to employer contributions made on their behalf while doing wage work. So, for example, a self-employed person who is also employed by a firm and who is working a full-time, full- year job for wages/salaries has no further obligation with regard to the 80 percent/employer share. An individual who works full time for wages for 8 months and then quits that job and becomes self-employed is only obligated up to a maximum of 4 months of the 80 percent of the adjusted per worker premium for his/her health insurance unit type. C. Discounts to Low Income Families. Low income workers and non-workers (those with family income less than 150 percent of poverty) [The family size specific poverty guidelines used are as follows: single -- family size is 1 couple -- family size is 2 single parent family -- family size is 3 dual parent family -- family size is 4.] are eligible for government discounts to assist in the payment of the family share of the premium and to assist with family out-of-pocket payments (co-insurance and deductibles). The family premium share discounts work as follows: 1. Those with family incomes at or above 150 percent of poverty are responsible for paying the full 20 percent share, subject to a maximum of 3.9 percent of family income. 2. Those with family incomes below 150 percent of poverty have their premium obligation calculated as: MARG sub 1 (INC sub 1 - 1000)+MARG sub 2 (INC sub 2 - INC sub 1) where INC_1 is equal to the family income up to the appropriate poverty guideline, INC_2 is equal to family income if it exceeds 100 percent but is less than 150 percent of the appropriate poverty guideline, MARG_1 is the contribution rate applied to family income below poverty, and MARG_2 is the contribution rate applied to family income between 100 and 150 percent of poverty. The two contribution rates are such that families below poverty do not pay more than 3 percent of income for their family premium share contribution, those with income below $1000 [In 1994 dollars. The income "disregard" is indexed by the CPI in future years.] have no premium contribution. Families at 150 percent of poverty pay the full 20 percent share, or 3.9 percent of family income, whichever is less. The government payment is equal to 20 percent of the actuarial premium for the health insurance unit type, less the family contribution calculated above. For purposes of this calculation, family income is equal to adjusted gross income less unemployment compensation plus non-taxable interest income. For each marginal rate (MARG_1 and MARG_2), there are two sets of rates to be used. The first set (MARG_1single, MARG_2single) is applicable for single health insurance units and uses the poverty guidelines for a family of size one. The second set (MARG_1other, MARG_2other) is applicable to all other health insurance units and is based on the poverty guidelines for a family of size four. Set one is calculated as follows: MARG_1_single = (0.03 * POVG_1) / (POVG_1 - 1000) MARG_2_single = ((0.2 * PREM_s) - (0.03 * POVG_1))/(0.5*POVG_1) where POVG_1 is based on the poverty guidelines for a family of size one, and PREM_s is the premium for a single individual. In 1994, these rates are estimated to be 3.5 percent and 4.8 percent, respectively. Set two is calculated as follows: MARG_1_other = (0.03*POVG_4) / (POVG_4 - 1000) MARG_2_other = ((0.2*PREM_DP) - (0.03*POVG_4)) / (0.5*POVG_4) where POVG_4 is the poverty guideline for a family of size four, and PREM_f2 is the premium for a dual parent family. In 1994, these rates are estimated to be 3.2 percent and 5.8 percent, respectively. These rates result in singles and dual parent families paying their full 20 percent premium share at 150 percent of poverty. At 150 percent of poverty, singles pay 3.6% of their income and dual parents families pay 3.9 percent. When the second set of marginal rates are applied to single parents and couples, these families are paying approximately 3.9 percent of their income at 150 percent of their appropriate poverty guideline (family size set at three for single parents and two for couples); however, that amount is less than 20 percent of their respective premiums. Consequently, couples and single parents with incomes in excess of 150 percent of poverty will be required to pay the lesser of 20 percent of their premium and 3.9 percent of income. An out-of-pocket spending discount is available for those families below 150 percent of poverty who live in an area that does not provide access to a low cost sharing (HMO) plan. In such cases, the family is only obligated to pay the cost sharing that would be required if the family had actually enrolled in an HMO (i.e., $10 copayment for outpatient services); and discounts will be available for the remainder. Families without at least one full time worker or equivalent [Two examples of families with a "full time worker equivalent" are: 1. each spouse works half time for the full year; 2. one spouse works full time for 8 months and the other works full time for 4 months.] may be required to pay at least some portion of the 80 percent adjusted premium share that is covered for workers through their employers. Families with non-wage income below 250 percent of poverty are eligible for some subsidization of this obligation. If eligible, a family's payment for this portion of the premium is equal to: MARG sub3 (NWINC sub 1-1000)+MARG sub4 (NWINC sub2 - NWINC sub1) where NWINC_1 is equal to the family non-wage income up to the appropriate poverty guideline, NWINC_2 is equal to family non-wage income if it exceeds 100 percent but is below 250 percent of the appropriate poverty guideline, MARG_3 is the contribution rate applied to family non-wage income below poverty, and MARG_4 is the contribution rate applied to family non-wage income between 100 and 250 percent of poverty. MARG_3 is set such that families below poverty do not pay more than 5.5 percent of their non-wage income for this portion of the premium, and families with less than $1000 in non-wage income have no required contribution towards this portion of the premium. The federal payment is equal to 80 percent of the appropriate adjusted per worker premium less employer payment credits, less self-employment contributions, and less family contributions as defined above. Non-wage income is calculated as Adjusted Gross Income (AGI) less wages and salaries less unemployment compensation and less self-employed income. [The actual legislation excludes wages and salaries up to $60,000 per year. Wages and salaries in excess of this amount count towards this calculation. The $60,000 exclusion cap was not modelled, making the subsidy estimates somewhat over-stated.] Income in this category includes: rents and royalties, interest (including non-taxable interest income), dividends, alimony, capital gains/losses, the taxable portion of social security, partnerships, and trusts. Aside from items mentioned above, other categories of excluded income are: welfare payments, VA benefits, worker's compensation, child support income, inherited money, and proceeds from life insurance. There are four sets of contribution rates which can be applied to non-wage income: one each for single, couples, single parent, and dual parent families. They are calculated using the formulas shown below, using the family sizes of 1, 2, 3, and 4 respectively to determine poverty guidelines. For family type 'i': MARG_3_i = (0.055 * POVG_i) / (POVG_i - 1000) MARG_4_i = ((0.8 * P_i) - (0.055 * POVG_i))/(1.5 * POVG_i) where POVG_i is the poverty guideline for the appropriate family size, and P_i is the appropriate adjusted per worker premium. In 1994, these rates are estimated to be as follows: for singles, 6.4 percent and 10.7 percent; for couples, 6.1 percent and 10.9 percent; for single parent units, 6.0 percent and 9.8 percent; for dual parent units, 5.9 percent and 7.5 percent. These rates were calculated so that families pay their full employer share of the premium at 250 percent of poverty. D. Retiree Discounts. Families with retirees [The policy defines retirees as those nonworkers who have fulfilled a requirement of a minimum number of working quarters and who are between the ages of 55 and 64, inclusive. However, the models being used to simulate the cost of the plan do not have data on quarters worked. Consequently, all individuals 55 to 64, who are not working or work part time or part year, are modelled as being eligible for the special retiree subsidy.] are eligible for a special discount. When fully phased in, government discounts cover the full 80 percent/employer share for non-working retirees. Government discounts are offset to some extent by the employers of retirees who work part time and the employers of working spouses. For example, a 58 year old man who is working half time will have half of his employer contributions made by his employer and half of his contributions will be made by the federal government. No government discount is necessary when a retiree has a full time working spouse, as the spouse's employer's contributions will fulfill the coverage responsibility. However, if a retiree is married to a non-worker, the government contribution will cover the couple (or family). E. Choice of Discount Estimates for Budgeting Purposes and Distributional Analyses: For reasons noted above in the section on premium estimation, the HCFA premiums were selected as the official Administration estimates. This choice necessitated that a model using the HCFA premiums be used as the official Administration discount estimates. For this reason, the discount estimates used for budgeting purposes are from the HCFA simulation model. It should be noted however, that all three discount estimates were within 7 percent of each other. Consequently, all estimates are well within the discount "cushion." [The HCFA discount estimates were increased by 15 percent in an effort to budget a more conservative level of discounts. ] For purposes of distributional analyses, the Administration's official estimates come from the Urban Institute's TRIM2 model, which was benchmarked to the National Health Accounts and which used the HCFA estimated premiums. The Urban Institute is the most experienced of the three groups in doing the type of complex distributional analyses needed for the reform process. V. National Spending Impacts The change in spending produced by health reform can be summarized in terms of the impacts on businesses, households, and governments. Present business spending is here limited to employer contributions for employer-sponsored health insurance and for active workers and retirees. Under reform, employers are required to pay 80 percent of the average worker premium in their area (net of discounts) for most workers. Beyond the required outlays, it is expected that there will be supplementation of the required coverage. Those employers currently paying more than the required employer contribution percentage, or buying richer coverage (e.g., lower cost-sharing) are assumed to continue to pay more than the required minimum. The calculations of changes in business outlays are similar in TRIM2 and SPAM. If an employer currently pays more than 80 percent of premiums, TRIM2 increases employer spending under reform to match the proportion contributed by the employer currently, as long as this does not exceed current spending. If maintenance of the current proportion would exceed current spending, it is assumed that employers increase their spending only to the point of current spending. Worker contributions are reduced accordingly. This first part of supplementation is then increased to add the cost for enhancing the richness of coverage up to the current level of plan richness associated with each currently insured worker. The cost of matching the current richness of benefits is paid by the employer and the worker in proportion to current premium contributions. In the SPAM model, additional coverage is assumed wherever current payments are better for the family than modeled future payments under the mandated benefit package. Supplementation amounts are accumulated equal to the difference between current and required benefits. Employer contributions are assumed to cover the supplement, although employer payments for the required coverage are held to the mandated minimum. The AHSIM Model assumes that both employers and households attempt to hold their spending on health insurance constant from baseline to reform. To the extent that baseline spending on employer-sponsored insurance exceeds expenditures required under reform, employers are first assumed to buy down their employees' required contributions. If baseline spending exceeds reform requirements for either households or employers after taking this transfer into account, the AHSIM model then allows both households and employers to buy supplemental insurance. For each health insurance unit in AHSIM, the actuarial value of supplemental insurance purchased under reform cannot exceed baseline levels. The total amount of supplemental insurance is also limited by the level of potential out-of-pocket expenses (cost-sharing plus noncovered services) under reform. Supplemental insurance is also assumed to carry a higher administrative load than basic health plans, 25 percent in most recent simulations. Any employer excess that remains after buying supplemental insurance is assumed to increase other tax-preferred fringe benefits. Household spending is defined to be the employee contributions for employer-sponsored health insurance, direct premiums for non-group coverage (under the current system) or direct purchase of alliance coverage (under reform), and cost-sharing payments. In the baseline, the employee contributions are defined to include employee payments irrespective of tax status; pre-tax employee contributions are counted as employee payments despite IRS treatment of such sums as employer contributions. To the extent supplementation implies higher business payments, household spending is reduced by like amounts. Total changes in cost-sharing are calculated as the net of reduced payments due to new and enriched coverage, against increased cost-sharing attributable to required purchase of insurance leading to increased utilization and some personal payments (rather than reliance on uncompensated care mechanisms). Government spending changes reflect transfers between the Federal government and other levels of government, as well as increased Federal responsibilities (particularly in arranging discounts for low-wage firms). Baseline Federal spending is primarily Medicaid and Medicare. Under reform, Medicaid non-cash populations move into alliance plans, with some direct business payments. Similarly, more Medicare recipients fall under working aged rules, with direct employer contributions reducing Medicare responsibilities. State and local baseline spending is primarily Medicaid, although significant sums are currently spent on other programs, most notably direct payments to hospitals. Under reform, Medicaid savings will be redirected under maintenance of effort requirements for use in paying for discounts for low-income populations in the alliances. References Burner, S. and D. Waldo. "Using HCFA's Special Policy Analysis Model to Simulate Benefits in a Health Reform Scenario". Health Care Financing Administration Internal Memorandum, March 1993. Doyle, P. "Microsimulation and Health Care Reform." Proceedings of the American Statistical Association 1993 Annual Conference. Alexandria, VA: American Statistical Association, 1993. Giannarelli, L. (1992). An Analysts' Guide to TRIM2. Washington, D.C.: The Urban Institute Press. Holahan, J. and S. Zedlewski. (1989). Insuring Low-Income Americans Through Medicaid Expansion. Washington, D.C.: The Urban Institute. Lewis, G. and R. Michel. (1990). Microsimulation Techniques for Tax and Transfer Analysis. Washington, D.C.: The Urban Institute Press. National Research Council. (1991). Improving Information for Social Policy Decisions. Washington, D.C.: National Academy Press. Wade, M. (1991). "Out-of-Pocket Expenditure Model (revised). Urban Institute Research Memorandum. (October). Zedlewski, S., J. Holahan, L. Blumberg, and C. Winterbottom (1993). "The Distributional Effects of Alternative Health Care Financing Options," Washington, D.C.: The Urban Institute. Zedlewski, S.R. (1991). Expanding the employer-Provided Health Insurance System: Effects on Workers and Their Employers. Urban Institute Report 91-3. Washington, D.C.: The Urban Institute. APPENDIX A Example of the Application of an Induction Factor To a Change in Insurance Status Before After Change Change Current law spending $200 $200 Multiplied by copayment rate X 50% X 20% Out-of-pocket spending $100 $ 40 Initial change in out-of-pocket $ 60 Multiplied by induction factor X .7 Equals change in total spending $ 42 Current law spending $200 Plus induced demand 42 Equals new total spending 242 Less new out-of-pocket (20%) -48 Equals new benefits 194 Induction Factors Used in SPAM Simulations Hospital inpatient (facility and physician) 0.3 Prescription drugs 1.0 Emergency room services and DME 0.0 All other services 0.7 SOURCE: Office of the Actuary Health Care Financing Administration