Methodology Report #14: Estimation of Expenditures and Enrollments for Employer-Sponsored Health Insurance
by John Paul Sommers, Ph. D., Agency for Healthcare Research and Quality
Select for more information
on Health Care
Information and Electronic Ordering through the AHRQ Web Site and
on the
Medical Expenditure Panel Survey (MEPS).
Abstract
The Agency for Healthcare Research and Quality (AHRQ) conducts the Medical Expenditure Panel Survey (MEPS), a survey of health care use and spending. The MEPS Insurance Component (IC) is a survey of business establishments and governments in the United States. It is focused on employer-sponsored health insurance - by far the largest source of health insurance in the United States. Because of the size of expenditures for health insurance in the Nation and their high rate of increase, they are of great interest. This report gives details of the enrollment and expenditure estimation process in the IC. Parts of this process use very standard statistical estimates. Where the estimation process deviates from standard methods, more detail is provided. The report also discusses changes in the data collected and in the estimators used that have taken place since the first IC survey year.
Background
The Medical Expenditure Panel Survey (MEPS) Insurance Component (IC) is a survey of business establishments and governments in the United States. In this report, business establishments (specific places of business) are differentiated from firms, which are legal entities that can own one or many establishments. The survey has two samples, the list sample and household sample. The list sample is a random sample of business establishments and governments in the United States that employ at least one person besides the owner. The household sample is a sample of the employers of respondents from the MEPS Household Component (HC).
The focus of data collected for the IC is information on employer-sponsored health insurance. Employment sponsorship is by far the largest source of health insurance in the Nation. Among the data collected is information on enrollments, premiums, contributions, plan types, plan coverages, retiree coverage, and employer characteristics. The nature of selection of the two samples means that they have different purposes. The household sample data are linked back to other data for the HC respondent and can be used to analyze persons and their medical expenditure choices. The list sample is a nationally representative sample of a large percent of employers and can be used to make national, industry, State, and other levels of estimates of the characteristics of employer-sponsored health insurance (Sommers, 1999a).
A large number of tables of estimates are produced from the list sample (http://www.meps.ahrq.gov/data_stats/quick_tables.jsp). Most estimates are straightforward and require no explanation concerning how they are made. Weights that denote the number of establishments in the universe represented by the sampled establishment are created for each sample member in the list sample (Sommers 1999b). Most estimates are made using weighted sample sums and ratios of these sums using techniques described in Kish (1965). For instance, an estimate of the percent of private-sector establishments that offer health insurance is the ratio of the sum of the weights of private-sector establishments that offer health insurance and the sum of the weights of all private-sector establishments: the estimate of the total number of establishments that offer health insurance divided by the estimate of the total number of establishments.
Estimates of total expenditures, enrollments, and contributions for employer-sponsored health insurance and their breakdown into groups (such as by industry, employer versus employee, and government versus private sector) constitute a set of key estimates made with IC data (http://www.meps.ahrq.gov/data_stats/quick_tables.jsp). Because of the size of expenditures for health insurance in the Nation and the large percent represented by employer-sponsored health insurance, estimates of these expenditures are of great interest. In addition to their overall size, these estimates are of interest because of their rate of increase. Government agencies such as the Bureau of Economic Analysis need good data to produce such estimates as part of the production of important overall Gross Domestic Product estimates, used to monitor the economy of the United States (Interdepartmental Committee on Employment Related Health Insurance Surveys, 2000a).
This report gives details of the enrollment and expenditure estimation process. The parts of this process that use standard statistical estimation procedures such as those described above do not receive much discussion in this report. Where the estimation process deviates from standard methods, more detail is provided.
The report also discusses changes in the data collected and the estimators that have been used since the first IC survey year, 1996. Future changes will be reported in later reports in this series. Changes are made as a result of learning what data used to make enrollment and expenditure estimates are available from respondents and what resulting changes and compromises are required to produce estimates using the data that can be collected.
At the end of the report, selected estimates made using MEPS data are compared to estimates from other survey or administrative sources and conclusions are given.
Return to Top
Original Estimation Plan
The original plan for estimation of total employer-sponsored health insurance costs was intended to be simple. Employers were asked to provide total expenditures for health insurance for establishments in the sample. They were also asked to provide total enrollment related to each establishment: active, COBRA (optional extended coverage for former employees who have left the employer but are not retired), and retirees. National totals and subtotals for various populations were to be estimated using the weighted sum of establishment costs for health insurance or enrollees of the particular type for which an estimate was needed.
It was soon apparent that this original estimation plan was problematic. Among the reasons were the following.
- Large multi-site employers, which pay a large percentage of the costs, could not supply cost information at the establishment level because plans were offered over many establishments and data were not broken out by establishment. They could provide data only at the level that the bill was paid regional national, etc.
- Large multi-site employers do not keep retiree and COBRA information at the establishment level. Furthermore, there may be no establishment that corresponds to an individual retiree. Establishments move and close, so this link does not exist for all retirees. Thus, if a company did relate individual retirees to establishments, many retirees would be missed and national estimates would be low.
- The estimates made were very different from estimates that could be made from other sources. Although the other estimates were not of the highest quality, the IC estimates using the original method were too markedly different to be credible.
Because of these problems, other methods were developed to produce the weighted survey estimates. Questionnaires were changed between 1996 and 1998 to collect information that could support these estimates. (Appendixes A and B show the 1996 and 1998 questionnaires.) In 1996 and 1997, interim estimation methods were used. This report describes the 1998 estimation methods and also discusses the interim methods used in 1996 and 1997.
Return to Top
Current Estimation for Active Employees
The estimation process is broken into four parts:
- Hospital/ physician coverage for active employees.
- Hospital/ physician coverage for retirees.
- Hospital/ physician coverage for former employees with COBRA coverage.
- Optional coverage, such as separate dental insurance.
This division occurred because it provides breakouts that are of concern to users and because the data available from respondents require it.
Aside from breaking estimation into parts, a new approach was developed that does not use totals of expenditures for each establishment, but instead builds these establishment totals when necessary.
The new process estimates hospital and physician coverage expenditures and enrollments for active employees at each business establishment or government as follows. Annual expenditure estimates for four items are made for each establishment: total family and single contributions for the year for employees at the establishment and total family and single contributions for the year for the employer at the establishment. Each estimate is made by taking the sum over all plans within the establishment or government of the enrollment for the type of coverage for each plan times the employer or employee annual contribution to the premium. Thus, if two plans are offered at an establishment or government, the total single employee contribution for that establishment or government is the sum across plans of the product of the single enrollment and single contribution reported for the individual plans.
These estimates assume a constant enrollment within the year at each establishment or government. The enrollment for the year is set at the time of collection, which is late summer. Contributions per enrollee are set based on the contribution in effect at that time. These assumptions tend to work rather well. Bureau of Labor Statistics monthly employment estimates for the late summer tend to be close to the annual average (www.bls.gov/ces/). Furthermore, total employment for the IC is also post-stratified, so enrollments are estimated using these values and the percent enrolled (Sommers, 1999b). Also, contributions are fixed by plan year, and most establishments' plan years run from January 1 to December 31, the estimation period. Establishment and government values are weighted and summed to make national and subnational estimates. This was the original plan for all expenditure estimates. Here the method is applied to active employee expenditures and enrollments only. This method creates annual estimates that benchmark well to other available national totals. (Comparisons with other estimates are shown later in this report.) This type of estimate was used for active employees, retirees, and those with COBRA coverage in 1996. However, because total single enrollment collected included all types of enrollees - active, COBRA, and retirees - which was not the case for later years (Appendixes A and B), no single enrollment for active employment was collected. Thus, for 1996 this value was calculated by taking the percentage of all enrollees that were single enrollees and applying it to total active enrollees to obtain active single enrollees.
Return to Top
Estimates for Retirees
The primary reason for changing the original estimation methods was that there were problems with collection of data on retirees. Data for retirees are not associated with establishments (specific locations) but instead are available at the level of the firm (a legal entity that can own multiple establishments). This is a common problem in business surveys. The sampling unit must be chosen carefully to accurately reflect how data are collected and to allow the production of the estimates required. Sometimes not all data can be collected for the same unit, so special estimation methods are needed (Sommers, 2000).
In the case of the IC and retirees, the sampling unit is the establishment but the reporting unit now used for retiree information is the firm. Data on the total number of single and married retirees enrolled with the entire firm, along with average contributions for the most common plan, are currently collected. (See questionnaire in Appendix B.) This change was made for data collected for calendar year 1998 and beyond.
Since only establishment weights are created for the IC, it is necessary to convert firm-level information on retiree enrollments to establishment-level information. If this is done, the contributions reported at the firm level can be used to build establishment-level expenditures and produce estimates in the same manner as for active enrollees.
To carry out this estimation procedure, a simple proration method is used. Each firm on the frame and establishment linked to that firm has a frame value for total establishment and firm employment. This means that if the proportion of firm employment for each establishment on the frame is calculated, the sum of these proportions across the establishments in the firm adds to 1. Under these conditions, the weighted sum of the prorated enrollments is an unbiased estimate of total retiree enrollments (Sommers, 2000). Under the same assumptions used for active expenditures, this prorated enrollment can be multiplied by an average plan contribution to create a prorated total contribution for the establishment, and the weighted sum over the entire sample of establishments of these prorated contributions gives a national estimate of total retiree contributions. For example, if a firm has 1,000 retirees and 10,000 employees, then an establishment within the firm with 500 employees would be allocated 50 = 1,000 (500/ 10,000) retirees for estimation purposes.
Estimates for subtotals across a subset of establishments are obtained using the weighted sum of a similar subset from the sample. An example would be estimates made for each industry. There is an exception: currently, no information is collected about whether retiree plans are purchased or self-insured (where the employer pays medical costs directly). To make estimates that break expenditures and enrollees into these two types of plans, retiree enrollees are prorated using the percentage of active enrollees at the establishment in each type of plan.
As mentioned previously, the new method of estimation for retirees was adopted beginning in 1998. Before 1998, respondents were asked to report retirees at the establishment level. Retirees were grouped into family or single coverage using the percent of all enrollees who had single coverage. Estimates were made using weighted sums of these establishment-level values. There was no requirement for an adjustment firm-level information to the establishment level, as was done for the 1998 results. The change in methods had a significant effect on the estimated number of retirees with health insurance and the total expenditures for retiree health insurance. In 1996 and 1997, using the direct estimator, the total national estimate of private-sector retirees with health insurance through their former job was about 3.1 million. With the new estimator, the number for 1998 jumped to 5.7 million, with a corresponding increase in expenditures. Although the reporting problems associated with the 1996-97 method were well known, efforts were made to verify that the increase was caused by a change of estimators. This proved to be a somewhat difficult task, as no independent reliable estimates of numbers of retirees enrolled in employer-sponsored health insurance plans could be found. However, analysis of certain weighted and unweighted numbers provided evidence that the results for 1996 and 1997 were low.
The estimates of total enrollees for the private sector changed dramatically, from about 3.1 million for 1996-97 to 5.7 million for 1998. Analysis of the total reported number of retirees for the largest 500 firms in the sample shows that 3.1 million is likely too low a result. In 1998, the first year this information was collected, these firms reported approximately 2.5 million retired enrollees. Less than half of the employees who worked in establishments that offered health insurance were represented by this group of firms. Assuming that other establishments that offer health insurance to retirees have a similar ratio of retirees with health insurance to active employees gives an estimate of over 5 million retirees with employer-sponsored health insurance. Moreover, the fact that the top companies in the sample, which represent less than a quarter of the total employment, report retiree enrollment almost as large as the MEPS estimates for 1996 and 1997 makes a compelling argument that the higher estimate for 1998 is a better estimate.
The effect of this change on total annual expenditures for 1996 and 1997 is $8-$ 10 billion. This change will be discussed in terms of the overall national estimates of expenditures for employer-sponsored health insurance later in this report when IC estimates are compared with other results.
In comparison to private-sector enrollment, the estimates of retiree enrollment for State and local governments are similar for all three years (1996-98), which seems reasonable since in MEPS, governments are asked to report for the entire government, so the problem of site (establishment) and controlling entity (firm) does not exist. Government-wide reporting is similar to private-sector reporting at the firm level. Collection at this level was possible because State and local governments fall entirely within single States; thus, there was no need to gather data by location in order to make State estimates.
Return to Top
COBRA Estimates
COBRA is coverage under laws for continuation of benefits. It is a small portion of the total employer-sponsored health insurance market. Like retirees, COBRA enrollees were part of total reported enrollments in 1996. This enrollment was divided into family and single using the percent of single coverage for all enrollees, the same method used for active and retiree enrollments. In 1997, COBRA enrollment was separate from the totals. Currently, the weighted sums method used for values for active employees is also used for COBRA enrollees. Estimates for family premiums are done in a similar manner. Since no data on breaks in single enrollment are collected for COBRA coverage, the percent of people with COBRA single coverage is assumed to be the same as the percent of active enrollment for the same plan at the establishment. Also, it is assumed that all contributions for COBRA coverage are made by the enrollee.
Return to Top
Optional Coverage
Some employers offer coverage for single items, such as vision or dental care, as an optional coverage. An employee can enroll in these plans separately from the more standard hospital/ physician coverage normally offered. Expenditures for this coverage are estimated by obtaining the total employer costs for this coverage at the establishment and taking a weighted sum of these establishment totals over the sample. This method parallels the method that was originally intended for costs of hospital/ physician coverage. However, this original estimation method was not abandoned for estimates of costs of optional coverage. Several factors prevented the change for optional coverage:
- Very little is known about the levels of premiums, so it is difficult to evaluate results.
- Optional coverage is only a very small portion of employer health insurance expenditures; thus, a major change in estimation method may have little effect on total expenditure estimates.
- A single employer may offer many plan types. To request enrollment and premiums for each plan type may increase respondent burden without a major payback in the quality of estimates.
Steps have been taken to improve the estimates through changes in the questionnaire. Many hospital/ physician plans include coverage for items, such as dental care, that are obtained through a special third-party plan. For instance, a health maintenance organization such as Kaiser Permanente runs its own medical clinics but provides dental coverage by subcontract to a specialized dental provider. This can also occur with self-insured plans, where the employer pays for medical costs directly. Such self-insured plans can choose to obtain dental coverage by buying a separate dental plan. In 1996, some employers reported optional plans, which could only be obtained as part of their medical plan, as optional coverage. However, they also included the plan premiums in the premiums for the main hospitalization coverage. Thus, some expenditures were double-counted in the national estimates.
Because of this problem, the questionnaire
was changed in 1997 and better instructions given to respondents
(Appendixes A and B). The estimate for optional coverage expenditures
dropped from approximately $12 billion in 1996 to $8 billion in 1997.
The estimate for 1998 was $10 billion. At the same time, the number
of employers reporting optional dental or prescription coverage in
1997 and 1998 was lower than the number reporting these types of
coverage in 1996. Since these are the two most common types of coverage
included in hospital/physician coverage to be subcontracted, it appears
that the change in the questionnaire had the desired effect and that
the estimates of employers reporting this optional coverage were
more accurate.
Return to Top
Benchmark Comparisons
Efforts were made to evaluate results produced by the IC (Interdepartmental Committee on Employment Related Health Insurance Surveys, 2000b). In this section, selected examples of these comparisons are given. Because the IC is unique, no single comparable survey or set of comparable results is available. Thus, individual estimates from the IC are compared to estimates from a variety of specific sources.
Table 1 compares the MEPS IC estimate of the percent contributed by individuals in the private sector for single coverage with estimates from the 1993 National Employer Health Insurance Survey, or NEHIS (National Center for Health Statistics, 2002; Westat, 1994) and the 1999 Employer Health Benefits Survey, funded by the Kaiser Family Foundation (Levitt, Lundy, Hoffman, et al., 1999). Values are generally within statistical error when considering error from the IC alone. (Errors are not available for any of the other estimates used for comparison in this report.) Although the sample years are different, under the assumption that percent contributions are relatively stable across years, the three surveys seem to agree among themselves.
Table 2 compares the percent of private-sector employees enrolled in employment-related health insurance by industry as estimated by the 1993 NEHIS; the 1997 MEPS IC; and the 1996-97 Employee Benefits Survey (EBS), conducted by the Bureau of Labor Statistics (Interdepartmental Committee on Employment Related Health Insurance Surveys, 2000b). As with percent contributions, most of the estimates fall into line with each other when the likely standard error is considered. Estimates for the construction industry group appear to be different. However, the sample size for construction in the EBS is small, so there could be a large standard error associated with that estimate.
Table 3 shows single premiums by industry for the 1993 NEHIS and the 1997 MEPS IC. Although one would expect some price change over this period of time, the general size and order of size of the premiums by industry are well correlated.
Other available individual comparisons of IC results with data from other sources show similar results. The total expenditure values for health insurance for the entire United States are of particular interest. Table 4 shows estimates based on the MEPS IC along with official estimates produced by the Centers for Medicare & Medicaid Services, or CMS (formerly the Health Care Financing Administration). The CMS results are produced using data from industry, administrative records, and surveys (Centers for Medicare & Medicaid 13 Services, 2002). Three estimates are produced, and then a final value is selected after judging their results. No error interval is given with these results, nor is any evaluation of past results given. The three results can vary among themselves by 5 percent or more within a single year. Considering the IC sampling error and the variability of the other three results in past years, the results seem to track rather well.
Return to Top
References
Interdepartmental Committee on Employment Related Health Insurance Surveys. Report: Data Needs Subcommittee. Washington; May 2000a.
Interdepartmental Committee on Employment Related Health Insurance Surveys. Draft report: Data Evaluation Subcommittee [unpublished report]. Washington; Oct 2000b.
Kish L. Survey sampling. New York: John Wiley and Sons; 1965.
Levitt L, Lundy L, Hoffman C, et al. Employer health benefits: 1999 Annual Survey. Chicago: Health Research and Educational Trust; 1999. Educational Trust Cat. Number 097501.
National Center for Health Statistics. Web site: http://www.cdc.gov/nchs/about/major/nehis/nehis.htm. Accessed Nov 8, 2002.
Sommers JP. List sample design of the 1996 Medical Expenditure Panel Survey Insurance Component. Rockville (MD): Agency for Health Care Policy and Research; 1999a. MEPS Methodology Report No. 6. AHCPR Pub. No. 99-0037.
Sommers JP. Construction of weights for the 1996 Medical Expenditure Panel Survey Insurance Component. Rockville (MD): Agency for Health Care Policy and Research; 1999b. MEPS Methodology Report No. 8. AHCPR Pub. No. 00-0005.
Sommers JP. Methods to produce establishment and firm level estimates for an economic survey. In: Proceedings of the International Conference on Establishment Surveys II. Buffalo (NY); June 2000. Alexandria (VA): American Statistical Association.
Westat, Inc. National Employer Health Insurance Survey (NEHIS): Survey design. Rockville (MD): Agency for Health Care Policy and Research; May 1994. Contract No. 200-94-7003.
Return to Top
Table 1
Percent of total premium for single health insurance coverage paid by employee for employer-sponsored insurance among private-sector employees: United States
| Industry |
1993
National
Employer
Health
Insurance
Survey
estimate |
1999
Employer
Health
Benefits
Survey
estimate |
1997
Medical
Expenditure
Panel
Survey
Insurance
Component
Estimate |
1997
Medical
Expenditure
Panel
Survey
Insurance
Component
Standard
Error |
| Total |
15.1 |
16 |
15.6 |
0.52 |
| Construction |
17.0 |
13 |
15.6 |
1.71 |
| Manufacturing |
13.6 |
12 |
14.1 |
0.62 |
| Transportation |
10.4 |
11 |
11.0 |
0.92 |
| Wholesale trade |
15.4 |
13 |
15.0 |
1.48 |
| Retail trade |
23.0 |
26 |
22.6 |
1.39 |
| Finance |
15.4 |
16 |
16.6 |
1.15 |
| Services |
14.7 |
21 |
14.6 |
0.80 |
Sources: National Employer Health Insurance Survey - http://www.cdc.gov/nchs/about/major/nehis/meps_ic.htm#national%20data (accessed Nov. 8, 2002); Employer Health Benefits Survey - Levitt L, Lundy L, Hoffman C, et al. Employer health benefits: 1999 Annual Survey. Chicago: Health Research and Educational Trust Cat. Number 097591; Medical Expenditure Panel Survey - www.meps.ahrq.gov (accessed Nov. 8, 2002).
Return to Top
Table 2
Percent of all private-sector employees enrolled in employer-sponsored health insurance: United States
| Industry |
1993
National
Employer
Health
Insurance
Survey
Estimate |
1996-97
Employer
Health
Benefits
Survey
Estimate |
1997
Medical
Expenditure
Panel
Survey
Insurance
Component
Estimate |
1997
Medical
Expenditure
Panel
Survey
Insurance
Component
Standard
Error |
| Total |
57.5 |
57 |
57.2 |
0.43 |
| Construction |
42.4 |
61 |
44.3 |
1.22 |
| Manufacturing |
78.3 |
79 |
78.6 |
0.35 |
| Transportation |
72.5 |
73 |
75.1 |
1.24 |
| Wholesale trade |
67.5 |
73 |
71.7 |
1.18 |
| Retail trade |
34.3 |
32 |
36.8 |
0.68 |
| Finance |
67.9 |
75 |
70.5 |
0.99 |
| Services |
53.7 |
51 |
52.5 |
0.68 |
Sources: National Employer Health Insurance Survey - http://www.cdc.gov/nchs/about/major/nehis/meps_ic.htm#national%20data (accessed Nov. 8, 2002); Employee Benefits Survey - Interdepartmental Committee on Employment Related Health Insurance Surveys. Draft report: Data Evaluation Subcommittee [unpublished report]. Washington; Oct. 2000; Medical Expenditure Panel Survey - www.meps.ahrq.gov (accessed Nov. 8, 2002).
Return to Top
Table 3
Average annual premium for single health insurance coverage for employer-sponsored insurance among private-sector employees: United States
| Industry |
1993 National Employer
Health Insurance Survey
Estimate |
1997 Medical Expenditure
Panel Survey
Insurance Component
Estimate |
1997 Medical Expenditure
Panel Survey
Insurance Component
Standard Error |
| Total |
$2,069 |
$2,051 |
$20 |
| Construction |
1,924 |
1,931 |
45 |
| Manufacturing |
2,017 |
1,919 |
28 |
| Transportation |
2,271 |
2,199 |
106 |
| Wholesale trade |
1,980 |
2,088 |
49 |
| Retail trade |
1,868 |
1,843 |
29 |
| Finance |
2,195 |
2,122 |
25 |
| Services |
2,118 |
2,134 |
32 |
Sources: National Employer Health Insurance Survey - http://www.cdc.gov/nchs/about/major/nehis/meps_ic.htm#national%20data(accessed Nov. 8, 2002); Medical Expenditure Panel Survey - www.meps.ahrq.gov (accessed Nov. 8, 2002).
Return to Top
Table 4
Total annual health insurance premiums for employer-sponsored insurance for private-sector employees: United States
| Survey |
Billions of dollars
1996 |
Billions of dollars
1997 |
Billions of dollars
1998 |
| Centers for Medicare & Medicaid Services |
$344.8 |
$359.4 |
$383.2 |
|
Medical Expenditure Panel Survey Insurance Component Estimate |
346.4 |
347.7 |
392.8 |
| Medical Expenditure Panel Survey Insurance Component Standard Error |
7.8 |
7.8 |
8.6 |
Sources: Center for Medicare & Medicaid Services - http://www.cms.hhs.gov/statistics/nhe/default.asp#business (accessed Nov. 8, 2002); Medical Expenditure Panel Survey - www.meps.ahrq.gov (accessed Nov. 8, 2002).
Return to Top
Appendix A
1996 Insurance Component Questionnaire
FORM (7-7-97) MEPS-10 (PDF File, 135 kb) Medical Expenditure Panel Survey (Insurance Component) ESTABLISHMENT QUESTIONNAIRE
FORM (7-7-97) MEPS-10(S) (PDF File, 84 kb) Medical Expenditure Panel Survey (Insurance Component) SUPPLEMENTAL SHEET ESTABLISHMENT QUESTIONNAIRE
Return to Top
Appendix B
1998 Insurance Component Questionnaire
FORM (4-27-99) MEPS-10 (PDF File, 69 kb) Medical Expenditure Panel Survey (Insurance Component) ESTABLISHMENT QUESTIONNAIRE
FORM (4-27-99) MEPS-10(S) (PDF File, 64 kb) Medical Expenditure Panel Survey (Insurance Component) HEALTH INSURANCE COST STUDY PLAN INFORMATION QUESTIONNAIRE
U. S. Department of Health and Human Services Public Health Service Agency for Healthcare Research and Quality
AHRQ Pub. No. 03-0009 March 2003
ISBN 1-58763-126-1 ISSN 1531-5673
Return to Top
|
Back
Email Updates
