Accepted Manuscript Title: Active and Retired Public Employees’ Health Insurance: Potential Data Sources Author: Melinda Sandler Morrill PII: DOI: Reference:
S0167-6296(14)00091-5 http://dx.doi.org/doi:10.1016/j.jhealeco.2014.06.011 JHE 1789
To appear in:
Journal of Health Economics
Received date: Accepted date:
23-6-2014 24-6-2014
Please cite this article as: Morrill, M.S.,Active and Retired Public Employees’ Health Insurance: Potential Data Sources, Journal of Health Economics (2014), http://dx.doi.org/10.1016/j.jhealeco.2014.06.011 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Active and Retired Public Employees’ Health Insurance: Potential Data Sources
This Version: June 16, 2014*
ip t
Melinda Sandler Morrill, North Carolina State University
cr
In July 2013, public workers, including state, local, and federal government employees,
accounted for about 15 percent of the U.S. non-farm labor force.1 Clemens and Cutler ()
us
estimated that state and local governments spent $117 billion on health insurance in 2010 alone. Health insurance represented about 12.2 percent of the total compensation cost for state and local government workers in June 2013.2 Underfunding and growing costs of benefits have implications for public policy
an
and the fiscal solvency of state and local governments. Furthermore, findings from the private sector may not be directly relevant in the public sector because many public sector workers are covered by union
M
contracts or fixed salary schedules and often benefit modifications require changes in legislation. Thus, from both an academic and a policy perspective, understanding health care costs and health insurance in the context of public sector workers is of paramount importance.
d
There are many important questions associated with employer-provided health insurance and
te
retiree health insurance for public sector workers. For example, we might wish to know what a given state or local employer spends on providing health insurance benefits to active and/or retired employees.
Ac ce p
Is the cost of health insurance and/or retiree health insurance passed on to workers? How do the benefits affect employee behavior? How will the Affordable Care Act affect state and local governments’ employee health insurance provision? While articles referenced herein attempt to address some of these research questions, much is still not known about this expensive and potentially important employee benefit in the public sector. One reason that these topics are relatively understudied is that it is difficult to obtain sufficiently large and representative data focusing on public sector employees. This article outlines some potential data sources that interested researchers might utilize to investigate topics concerning employer-provided health insurance for active and retired public sector employees. *
Melinda Morrill, Department of Economics, Box 8110, North Carolina State University, Raleigh, NC 27695-8110. Email: [email protected]. I would like to thank Robert Clark, Jeffrey Clemens, David Cutler, Maria Fitzpatrick, Joseph Newhouse, and Thayer Morrill for useful discussions and comments. Chrystelle Khalaf provided excellent research assistance. 1
See Bureau of Labor Statistics, Current Employment Statistics, http://www.bls.gov/web/empsit/ceseeb1b.htm, [accessed October 29, 2013]. 2
See Bureau of Labor Statistics, Economic News Release, http://www.bls.gov/news.release/ecec.t04.htm, [accessed October 29, 2013].
1 Page 1 of 15
I. Nationally-representative individual-level data Many commonly-used, nationally-representative, individual-level datasets allow for the identification of public sector employees. To be a useful data source for studying health benefits in the public sector, a dataset should contain some information on health insurance coverage, earnings/wages,
ip t
employment status, and sector of employment. While it is not uncommon to observe health insurance coverage, it is relatively rare for a dataset to include information on the parameters and generosity of an employee’s health plan. However, if we observe that an individual is an employee of a state government
cr
and we know that employee’s state of residence, we can match publicly available information about health insurance offerings to the survey or administrative data. Thus, it is particularly useful if a
us
nationally-representative dataset includes state of residence or other geographic identifiers.3 In this section, we outline several data sources that separately identify federal, state, and local
an
government employees.4 This is not an exhaustive list but attempts to cover the largest and most prominent datasets. As a basis of comparison, Table 1 presents a description of data availability for health insurance, health status, and geography. The table then presents sample sizes for a sample of full-
M
time, currently employed federal, state, and local government workers.5 First, counts are presented for workers ages 18 to 64; then, counts are presented separately for federal, state, and local workers between ages 50 and 64.
d
I.A. Health and Retirement Study
te
By far the most common survey for studying older workers and retirees is the Health and Retirement Study (HRS).6 The HRS is a panel dataset that began in 1992. The 2010 data release includes
Ac ce p
ten waves of data and five entry cohorts.7 The most important advantage of the HRS is the wealth of detail available both from the respondent and from merged employer data in some years. In addition, the panel nature allows one to study past employment (e.g., Shoven and Slavov, ). The more 3
For example, one might use the classifications of retiree health insurance percent of the premium paid available in the GAO (2007) report, data found in actuarial reports, or data available on state websites. While theoretically one might do a similar matching on county for local government employees, no systematic data exist on county and city health plans. Moreover, as state and local health plans may differ, it may be difficult to assign a survey respondent to the correct health plan information once collected. For example, in North Carolina teachers are state government employees and part of the State Health Plan, but might erroneously report on a survey being a local employee. 4
The U.S. Census Bureau provides several data sources that contain health information, as summarized at: http://www.census.gov/sdc/healthstats_JCD.pdf, [accessed October 29, 2013]. Census Bureau data typically includes a variable on “class of worker” which identifies the sector of employment. 5
When possible, we define “full time” as working at least 40 weeks per year, at least 35 hours per week, and currently working for pay. 6
For information on accessing these data see: http://hrsonline.isr.umich.edu/, [accessed October 29, 2013].
7
The entry cohorts are 1992 HRS cohort; the 1993 Study of Assets and Health Dynamics (AHEAD) cohort; the 1998 Children of Depression cohort; the 1998 War Babies cohort; and the 2004 Early Baby Boomer cohort.
2 Page 2 of 15
recent years of data separately identify federal, state, and local government workers.8 In these data, health insurance coverage, health costs, and health status are all observed over time. However, geographic information in more detail than Census region is only available in a restricted-access version of the dataset.
ip t
As illustrated in Table 1, the most significant limitation of the HRS is the small sample size. Using the 2010 data release, which includes all respondents and their spouses from all cohorts that were still alive at the time of the survey, only 1,049 observations out of the 4,863 full-time, currently employed
cr
individuals are public sector employees. When we further restrict the sample to those between the ages of 50-64, the sample size is only 954 workers.
us
The HRS is particularly useful when a researcher requires detailed information on wealth/assets or health and health behaviors. For example, Clark and Mitchell () consider how access to
an
and generosity of health insurance affects retirement savings. In another example of how the HRS might be used to study health insurance and public sector employees, Slavov and Shoven () consider how the availability of employer-provided health insurance and retiree health insurance affects retirement
M
behavior.
I.B. Survey of Income and Program Participation The Survey of Income and Program Participation (SIPP) is a large, nationally-representative
d
dataset collected by the Census Bureau since 1984 with extensive information on sources of income.9
te
The sample design includes short panels (between two and four years), with new samples drawn every four years (prior to 1996 new panels began every year). The panels range in size from 14,000 to 36,700
Ac ce p
and are each nationally-representative. The SIPP data identify separately individuals employed in the private sector (for profit versus non-profit) and local, state, or federal government employees. The SIPP includes topical modules that are fielded in various waves for some panels. The Medical Expenses/Utilization of Health Care module includes health status and some measures of medical services utilization. Another topical module includes information on Employer Provided Health Benefits. Also particularly relevant may be the Retirement and Pension Plan Coverage module, which even includes data on work interruptions for family care. The panel structure allows for the data included in the topical modules to be merged across waves, so that one can construct a dataset that looks at
8
The 2006, 2008, and 2010 waves separately identify state or local workers, while previous waves grouped them together. See the data appendix of Clark and Mitchell (2013) for a methodology for separately identifying state and local workers using the RAND version of the HRS. 9
For more information and to access the SIPP, see: http://www.census.gov/sipp/, [accessed October 29, 2013].
3 Page 3 of 15
employment transitions within one panel or information from multiple waves of one panel may be collapsed and used as a detailed cross-section.10 Besides the wealth of detail available in the topical modules, a real benefit of the SIPP is the large sample size combined with a panel structure (albeit for only 2.5 years in more recent panels). For Table
ip t
1, we include statistics only for the 2008 Panel, Wave 1. Thus, if one wanted to include all available panels the sample size would be substantially larger. The total number of full-time federal, state, or local government employees in the 2008 Panel is approximately 23,325. When restricting to only workers
cr
nearing retirement, ages 50-64, the total sample of public sector workers is 8,623. Thus, in only one panel the sample size is nearly ten times that of the more commonly used HRS.
us
I.C. Current Population Survey
The Current Population Survey (CPS) is sponsored jointly by the U.S. Census Bureau and the
an
U.S. Bureau of Labor Statistics.11 Each sample is about 60,000 households who are surveyed each month for 4 months, then out of the sample for 8 months, and then back in the survey for 4 months. The monthly data are used by the BLS to provide labor force statistics, such as the unemployment rate. There
M
are supplements in several of the months.12 The most commonly used supplement for labor economics research is the Annual Social and Economic Supplement conducted in March. One particular advantage of the March CPS is the inclusion of a measure of the employer’s contribution towards the employee’s
d
health insurance. Since 1996 the data also include a five point health status scale. In addition, the 2010
te
CPS includes a new set of questions on medical out-of-pocket expenses. Table 1 presents statistics for the March CPS pooling data from 2008 through 2012. Using just these five years of data, we see the total
Ac ce p
sample of public employees is 57,237, including 19,930 public employees ages 50-64.13 I.D. American Community Survey The American Community Survey (ACS), reported annually since 2006, replaces the long-form of the decennial Census.14 The ACS is released as 1-year, 3-year, or 5-year pooled estimates with more
10
Perhaps more so than the other datasets considered here, the SIPP panels have suffered from attrition problems. Although I know of no studies examining this issue in particular, it is possible that attrition problems are particularly problematic when studying life transitions between waves, such as retirements.
11
For information and access to the CPS, see: http://www.census.gov/cps/, [accessed October 29, 2013].
12
For information on available supplements, see: http://www.census.gov/cps/about/supplemental.html, [accessed October 29, 2013]. 13
Qin and Chernew () use the Merged Outgoing Rotation Group (often referred to as the MORG) CPS from 1992-2011 for 25-54 year old males to measure the compensating wage differentials from the rising costs of employer-provided health insurance in the public sector.
14
For more information and access to the ACS, see: https://www.census.gov/acs/www/, [accessed October 29, 2013].
4 Page 4 of 15
geographic detail and data for smaller geographies available for the pooled samples.15 Health insurance information has been collected annually since 2008. The information on health insurance concerns only coverage and source, and there are no other measures of health behavior or health status beyond measures of work-limiting disabilities. For Table 1, we present the most recently released 3-year estimates, 2009-
ip t
2011. There are approximately 540,258 full-time federal, state, or local government employees, with 208,586 between the ages of 50-64 included in the 2009-2011 3-year ACS. These data would be
particularly useful for researchers with supplemental information on health insurance at the state or local
cr
government employer level, since the pooled data releases allow for the identification of small geographic areas.
us
I.E. Medical Expenditure Panel Survey
The Medical Expenditure Panel Survey (MEPS) began in 1996 and is conducted by the Agency for Healthcare Research and Quality.16,17 The MEPS has several components including a Household
an
Component (HC), which has a publicly available version.18 The MEPS-HC is an overlapping panel design. A new sample is drawn annually from the prior year’s NHIS, and each sample is followed for
M
two calendar years collected in 5 rounds.19
The MEPS-HC collects detailed information on health expenditures. Unlike other health-focused datasets, the MEPS includes a “job file” supplement on the respondents’ employment history and current
d
status, including the “employee type” of private, federal, state, or local employee. Because of its focus on
te
health care expenditure, this dataset provides the most detail on health insurance, payments, and health behaviors and outcomes. However, the public use version of the data does not identify the state of
15
Ac ce p
residence. Table 1 provides sample sizes for the 2011 MEPS Household Component survey.20 Note that The Census Bureau provides the following guidance for selecting which sample to use: https://www.census.gov/acs/www/guidance_for_data_users/estimates/, [accessed October 29, 2013].
16
For more information and access to the household component data, see: http://meps.ahrq.gov/mepsweb/, [accessed October 29, 2013].
17
The sample for the MEPS is drawn from the much larger National Health Interview Survey (NHIS), conducted by the Center for Disease Control and Prevention, National Center for Health Statistics. The NHIS is described in more detail in Section I.F. For researchers interested in data from earlier time periods, the National Medical Care Expenditure Survey (NMCES), National Medical Care Utilization and Expenditure Survey (NMCUES) and the National Medical Expenditure Survey (NMES) are predecessor surveys to the MEPS. Information on these sources can be found at: http://www.icpsr.umich.edu/icpsrweb/ICPSR/series/45, [accessed December 13, 2013].
18
The MEPS also includes an Insurance Component (IC), which is a restricted-access only dataset that includes extensive detail gathered from both public and private sector employers on their health insurance plans. Summary statistics from these data are available at the state and metropolitan area for private sector employers, but only at the national-level or by Census division for public sector employers.
19
For an illustration of the data collection process, see: http://meps.ahrq.gov/mepsweb/survey_comp/hc_data_collection.jsp [accessed October 29, 2013].
20
The 2011 data include panel 15, rounds 3-5, and panel 16 rounds 1-3.
5 Page 5 of 15
this is just one year of data, and new panels are drawn every year. So, one could create a substantially larger dataset by pooling across years/panels. The total numbers of full-time public employees in the 2011 MEPS are 1,510 ages 18-64 and 543 ages 50-64. I.F. Other Datasets
ip t
Although not included in Table 1, three other datasets warrant some mention. First, the National Health Interview Survey (NHIS) is a large, nationally-representative dataset that includes a wealth of health-related information.21 The NHIS is collected annually by the CDC’s National Center for Health
cr
Statistics for the purpose of producing statistics on disease incidence and prevalence. The data are
nationally-representative, but state identifiers are only available in the restricted-access version. There
us
are over 50,000 households surveyed annually. For one adult in the household, information on sector of employment (separately identifying federal, state, and local government workers) is available. There is
an
health insurance information, including out-of-pocket premium payments, for all household persons. The survey asks about employer/union paid contributions to health insurance, but over two-thirds of the sample report not knowing. While the NHIS includes substantial detail on health status and medical care
M
utilization, there is less detail on employment and health expenditures than is available in the MEPS, as described above.
The National Longitudinal Survey of Youth, 1979 cohort (NLSY-79) is an ongoing panel of a
d
nationally-representative sample of 12,686 individuals who were ages 14-22 in 1979.22 Again, these data
te
identify federal, state, and local employees. In addition to information on health insurance coverage, there is a supplemental health module for individuals ages 40 and older and one for 50 and older. While
Ac ce p
the panel structure and detail available make these data attractive, the sample size is limiting and the data represent only one cohort.
Finally, the Panel Study of Income Dynamics (PSID) is the longest running longitudinal household survey having begun in 1968.23 The data have grown over time because spouses and children of initial survey respondents are incorporated into the sample, in addition to immigrant refresher samples. The PSID has detailed information on employment, wealth/assets, health, and health insurance. However, the sample size is relatively small compared with the other data sources listed here. In the 2011 wave, there were a total of 14,607 individuals and 5,495 families. 21
For information and access to the public-use version of the NHIS, see: http://www.cdc.gov/nchs/nhis.htm, [accessed November 3, 2013].
22
For general information and access to the NLSY, see: http://www.bls.gov/nls/nlsy79.htm. For more detailed information about the health insurance variables, see: https://nlsinfo.org/content/cohorts/nlsy79/topical-guide/health; and for more information about the employment variables, see: https://nlsinfo.org/content/cohorts/nlsy79/topicalguide/employment, [accessed November 3, 2013]. 23
For information and access to the PSID, see: http://psidonline.isr.umich.edu/, [accessed November 3, 2013].
6 Page 6 of 15
II. Nationally-representative proprietary data and surveys The following resources are based on micro-level data, either at the employee or employer level, but only aggregate statistics are publicly available. It may be that interested researchers could access the underlying microdata through a special agreement with individual agencies, but for the purposes of this
ip t
article we will only discuss publicly available versions. The level of aggregation of publicly available data varies, but the data are typically broken out by state, type of employer (public versus private sector), and/or number of workers at the firm. State-level aggregation may be as detailed as necessary when
cr
considering variation between state government employers. In addition, these data are useful for
observing trends and for benchmarking smaller datasets that are not necessarily nationally-representative.
relevant websites and a basic summary of the data available.
us
Table II presents four prominent surveys as examples of this type of data. The table includes links to the
an
(1) National Conference of State Legislators: State Employee Health Benefits
The National Conference of State Legislators (NCSL) releases reports on state employee health benefits.24 These data include state-level information on the lowest cost plan and a standard policy
M
insurance option. Clemens and Cutler (2013) and Qin and Chernew () provide examples of how these data might be used in practice. These papers utilize data on plan costs and state versus employee contributions collected by the NCSL to determine the percent of the premium that state
d
governments pay for active worker’s health insurance.
te
(2) Bureau of Labor Statistics: National Compensation Survey The Bureau of Labor Statistics (BLS) conducts the National Compensation Survey (NCS)
Ac ce p
annually, which includes both private and non-federal public sector employers.25 The data include information from establishments on wages and benefits. The BLS produces several series using these data including the Employer Costs of Employee Compensation (ECEC) and Employment Cost Index (ECI). NCS Databases provide national-level estimates broken out by sector (state, local, or state/local
24
For more information on these data, see: http://www.ncsl.org/research/health/state-employee-health-benefitsncsl.aspx, [accessed November 4, 2013]. Monthly premium costs for individual and family coverage can be found in the following reports, [accessed June 12, 2014; report links sent by Richard Cauchi, Program Director, Health Program, National Conference of State Legislatures]: 2009 (with information on 2006 and 2009): http://www.ncsl.org/portals/1/documents/health/FamilyPrem09.pdf and http://www.ncsl.org/portals/1/documents/health/IndivPrem09.pdf. 2011: http://www.ncsl.org/Portals/1/documents/health/StateEHBenefits2011.pdf 2012: http://www.ncsl.org/documents/health/2012NCSLStateEmployeeHealthBenefits.pdf 2013: http://www.ncsl.org/documents/health/2013NCSLStateEmployeeHealthBenefits.pdf 25
For more information on these data, see: http://www.bls.gov/ncs/, [accessed November 4, 2013].
7 Page 7 of 15
combined) of measures such as health insurance spending, coverage of health insurance, and provision of other benefits.26 BLS provides on-site access to the NCS data.27 (3) The Segal Company: Study of State Employee Health Benefits The Segal Company conducts annual survey titled the “Study of State Employee Health Benefits.”28
ip t
The Segal Group publishes an annual report summarizing the findings. On the last page Segal states, “Segal can be retained to provide custom data reports, including comparisons of coverage costs among
describe the average state offerings of health insurance plans in 2012.
cr
plan types, regions, and/or population size.” Clemens and Cutler (2013) use Segal’s 2013 report to
(4) Kaiser Family Foundation/Health Research and Educational Trust: Employer Health Benefits
us
Survey
The Kaiser Family Foundation/Health Research and Educational Trust (KFF-HRET) conducts an
an
annual survey of private sector and non-federal public sector employers.29 These data include not only premiums and health benefit information for active workers, but they also separately analyze retiree health benefits. The annual reports have been used by researchers to understand health insurance trends
M
in the private and public sectors (e.g., Clemens and Cutler 2013). Micro-data are available for interested researchers by application only, but identifying information (including state identifiers) is removed in the
III. Teachers and schools data
d
public use data file.
te
Teachers and school employees represent almost half of all state and local government workers nationally.30 While these workers may not be representative of public sector workers in general,
Ac ce p
researchers have found focusing on schools and teachers to be a fruitful exercise, particularly because the data for this group are more readily available. The U.S. Department of Education’s National Center for Education Statistics annually collects the Common Core of Data (CCD).31 These data provide detail on individual schools, school districts, and state-level education agencies. The data include aggregated information on demographics, student outcomes, and both revenues and expenditures. The CCD includes several component surveys. Clemens 26
To create customized tables, see: http://www.bls.gov/ncs/data.htm, [accessed October 29, 2013].
27
For more information about accessing the NCS, see: http://www.bls.gov/bls/blsresda.htm.
28
A report describing findings from the 2012 survey can be found at: http://www.segalco.com/publications/surveysandstudies/2012statestudy.pdf, [accessed October 29, 2013].
29
A report describing findings from the 2013 survey can be found at: http://kff.org/private-insurance/report/2013employer-health-benefits/, [accessed November 4, 2013]. 30
See Bureau of Labor Statistics, Current Employment Statistics, http://www.bls.gov/web/empsit/ceseeb1b.htm, [accessed October 29, 2013].
31
For more information on these data, see: http://nces.ed.gov/ccd/, [accessed November 4, 2013].
8 Page 8 of 15
and Cutler () use CCD data from the School District Finance Survey, which has detail on school district revenues and expenditures that can be matched to employment data. Unfortunately, health benefit and pension costs are grouped together in these data.32 In addition to these nationallyrepresentative datasets, individual states often collect and track data on teachers and schools, as described
IV. Administrative records for states or large local governments
ip t
in Section IV below.
Almost all of the data sources described thus far are nationally-representative samples, and almost
cr
all are collected by national agencies or organizations for the purposes of documenting trends and/or
conducting research. An alternative source of data is the administrative records a public employer or
us
health insurance company maintains. Focusing on only one (albeit large) employer reduces the amount of potential policy variation to exploit. But, detailed information for one large employer does allow for the
an
more precise measurement of the benefits provided and often yields a large dataset. The structure of administrative records generally allows for the creation of a panel, so that one can track how individuals who remain with their employer behave over time when exposed to different policies. The most
M
important limitations of this type of data are the lack of detail on family structure and the difficulty in tracking behavior after health insurance coverage and/or employment is terminated. Public employers with defined benefit pension plans must maintain an up-to-date data file with
d
information on active and retired employees to submit to plan actuaries for the calculation of pension
te
liabilities annually. Nearly identical data, along with information on health plan offerings, is provided to actuaries to calculate liabilities associated with retiree health insurance, following Government
Ac ce p
Accounting Standards Board Rule 45 (GASB, 2004). Similarly, any health insurance plan maintains records of its members and must process medical claims. Although these data are maintained for the purposes of administering and valuing a plan, they may be a useful resource for research. Table 3 illustrates some variables that would commonly be found in these types of data. Each state or local employer likely collects slightly different data fields or may treat certain variables as particularly sensitive.33 Researchers interested in utilizing these types of data must apply to a state administrator directly.34 Ideally, a researcher could merge data from a health plan and pension plan to create a rich
32
The National Center for Education Statistics has begun a pilot project, the Teacher Compensation Survey, to collect better detail on benefits. For more information on the early efforts, see: http://nces.ed.gov/ccd/tcssurv.asp [accessed November 18, 2013].
33
These data sources likely have personal identifiers such as names, addresses, and Social Security numbers. Health information beyond simple plan choice is protected under HIPAA and will typically involve a much more thorough application and review process than information about employment or salary.
34
Because pension plans are paid through public dollars, some may even consider these data part of the public record and subject to a Freedom of Information Act request (see, e.g., Leiserson, 2013).
9 Page 9 of 15
dataset. However, typically these data are collected by separate agencies, presenting significant legal and administrative hurdles. Recent research highlights how state-level administrative records can be used in practice. Using data maintained by a state health plan administrator, BlueCross BlueShield of North Carolina, Clark,
ip t
Morrill, and Vanderweide () consider how changing health plan characteristics over time affect members’ choice of health plan. Fitzpatrick () uses administrative files from the Illinois Public Schools, referred to as the Teacher Service Record, which contains information on
cr
experience, salary, position, assignment, and school. She combines the administrative records with statelevel policy changes on retiree health insurance availability, exploiting the detail on tenure to measure
us
eligibility for the program. Leiserson (2013) uses data on public sector workers included in the state health plan from the Pennsylvania State Employees’ Retirement System combined with information on
an
retiree health insurance policies over time. These studies demonstrate that administrative records from just one state, although not detailed, combined with information on policies and policy changes can be a powerful tool in understanding how health insurance relates to labor market outcomes such as
M
employment, retirement, and wages.
To understand the cost of and liabilities associated with providing health benefits, particularly to retirees, researchers can access reports prepared by plans to be in compliance with the GASB 45 ruling
d
(GASB 2004). The Comprehensive Annual Financial Reports (CAFRs) typically have extensive
te
information about the financial status of a public sector employer’s pension and retiree health plans. A researcher might also access the actuarial reports directly, which provide much more detail on the
Ac ce p
assumptions underlying the calculations of liabilities associated with retiree health insurance (typically these reports are referred to simply as “GASB reports”). Clark and Morrill (2010) provide more information on the data included in the actuarial reports. Lutz and Sheiner () illustrate the value of gathering the CAFRs and GASB reports to provide data on plan funding and actuarially accrued liabilities. These reports can be obtained from individual states and local governments, but are not systematically available in one place. V. Discussion and Conclusions
To date, researchers interested in studying issues surrounding health insurance and/or retiree health insurance in the public sector have had to confront limitations in data availability. This article highlights some potentially useful data sources, noting some studies using these data. Many of these data sources, of course, have been used to study similar questions with respect to private sector employees. Nationally-representative, individual-level data often includes indicators for sector of employment and at least some information on health insurance coverage. For state government employees or employees of large cities or counties, often supplemental data on benefits can be merged on using geographic 10 Page 10 of 15
identifiers. Similarly, aggregated data at the state-level may be useful for cross-state analyses of state government workers. Data on public schools and teachers may be more detailed and readily available than other segments of the public sector labor market. And, finally, state-specific administrative records from pension plans or health plans may be a useful resource for researchers. These administrative
ip t
datasets include large samples and high-quality data. Thus, as seen in articles published in this issue, there are sufficient data to conduct insightful, policy-relevant research on topics related to health
insurance provision by public sector employers. Moreover, this is an under-researched area with the
cr
potential of uncovering findings that would improve the fiscal sustainability of state and local
governments. This article is not meant to be a comprehensive list of available data, but it should provide
us
interested researchers a starting point for exploration.
an
References
Clark, Robert and Olivia Mitchell. (2013) “How Does Retiree Health Insurance Influence Public Sector Employee Saving?” NBER Working Paper #19511.
M
Clark, Robert and Olivia Mitchell. () “How Does Retiree Health Insurance Influence Public Sector Employee Saving?” Journal of Health Economics.
d
Clark, Robert L. and Melinda Sandler Morrill. 2010. Retiree Health Plans in the Public Sector: Is There a Funding Crisis? Northhampton, MA: Edward Elgar Publishing.
te
Clark, Robert, Melinda Morrill, and David Vanderweide. () “The Effects of Retiree Health Insurance Plan Characteristics on Retirees’ Choice and Employers’ Costs.” Journal of Health Economics.
Ac ce p
Clemens, Jeffrey and David Cutler. (2013) “Who Pays for Public Employee Health Costs?” NBER Working Paper #19574. Clemens, Jeffrey and David Cutler. () “Who Pays for Public Employee Health Costs?” Journal of Health Economics. Fitzpatrick, Maria. () “Retiree Health Insurance for Public School Employees: Does It Affect Retirement and Mobility?” Journal of Health Economics. Government Accountability Office (GAO). (2007). “State and Local Government Retiree Health Benefits: Current Status of Benefit Structures, Projections, and Fiscal Outlook for Funding Future Costs.” Report to the Committee on Finance, US Senate. September GAO-07-1156. http://www.gao.gov/new.items/d071156.pdf. Government Accounting Standards Board. 2004. Statement No. 45. Accounting and Financial Reporting by Employers for Post-employment Benefits Other Than Pensions. Leiserson, Gregory, (2013), “Essays on the Economics of Public Sector Retirement Programs,” MIT Dissertation, June 6, 2013. Lutz, Byron and Louise Sheiner. () “Examining the Whole Picture: Retire Health Obligations and the Long-Term Budget Outlook for the State and Local Government Sector.” Journal of Health Economics.
11 Page 11 of 15
Qin, Paige, and Michael Chernew. (). “Compensating Wage Differentials and the Impact of Health Insurance in the Public Sector on Wages and Hours,” Journal of Health Economics.
Ac ce p
te
d
M
an
us
cr
ip t
Shoven, John and Sita Slavov. () “The Role of Retiree Health Insurance in the Early Retirement of Public Sector Employees.” Journal of Health Economics.
12 Page 12 of 15
i cr Variables Included in the Data
Example Dataset: Full-Time Public Sector Employees
an
Dataset
us
Table 1: Nationally-representative individual-level data
No health behaviors Health insurance coverage and source only.
ce
March Current Population Survey (CPS)
ed
Survey of Income and Program Participation (SIPP)
Includes: detailed health insurance; health status; medical service utilization Includes: detailed health insurance; health status; medical service utilization (some years) Includes: detailed health insurance; health status
pt
Health and Retirement Study (HRS)
Sample & Years
Restrictedaccess only
2010 data, uses all cohorts
Full-Time, Currently Employed Ages 18-64 Total: 4,863 Public: 1,049
Yes
2008 Panel, Wave 1 only
Total: 124,298 Public: 23,325
Yes
2008-2012
Total: 338,926 Public: 57,237
Yes, detailed geography available Restrictedaccess only
2009-2011 3year estimates
Total: 2,889,703 Public: 540,258
Public Sector Employees Ages 50-64 Federal: 206 State: 383 Local: 365 Federal: 1,879 State: 2,735 Local: 4,009 Federal: 4,359 State: 6,285 Local: 9,286
Federal: 42,957 State: 64,801 Local: 100,828 Medical Expenditure Includes: detailed health 2011 Total: 56,907 Federal: 97 Panel Survey (MEPS) insurance; health status; Public: 1,510 State: 236 medical service utilization Local: 210 Notes: The example dataset includes only individuals that are currently employed full-time and are age 18-64. The final column further restricts this sample to include only federal, state, or local government employees ages 50-64. All datasets, excepting the HRS, have additional years of data (or panels) available.
Ac
American Community Survey (ACS)
State of Residence
M
Health & Health Insurance
13 Page 13 of 15
i cr us an
Table 2: Aggregated Employee Benefit Studies
Survey Title and Years
National Conference of State Legislatures
State Employee Health Benefits Annually 2001-2006, 2009, 2011-2012
Bureau of Labor Statistics, National Compensation Survey
Employee Benefits Survey (1985-2006)
http://www.bls.gov/ncs/ebs/
National Compensation Survey – Benefits (2010 – 2013)
Series produced as: “Employer Costs for Employee Compensation” http://www.bls.gov/news.release/ecec.htm
Kaiser Family Foundation
ce
pt
ed
M
Agency
Health Research and Education Trust
Segal
Ac
Annual Surveys 1999-present Annual Study of State Employee Health Benefits
Website http://www.ncsl.org/research/health/stateemployee-health-benefits-ncsl.aspx
http://kff.org/privateinsurance/report/2013-employer-healthbenefits/ http://www.segalco.com/publications/ surveysandstudies/2012statestudy.pdf
Health Insurance Information State-level information on lowest cost and standard policy insurance options. Only national-level estimates, broken out by sector (state, local, or state and local).
Survey of private and nonfederal public firms; includes separate industry category for state/local government. Reports statistics on state employee health benefits.
14 Page 14 of 15
i cr us an
Table 3: Typical Administrative Data Elements
Pension or Employment Database
Basic Demographics
• Date of birth • Gender
• Date of birth • Gender • (Teachers data) highest degree attained
Family Structure
• Coverage of dependents • Age and gender for covered dependents
ed
M
Health Insurer Database
• Date of first coverage • Date of last coverage
• • • •
Date of hire Date of separation Date of retirement Years of service (contributory, non-contributory, purchased, calculated) • Salary (may have panel of earnings or information for final average salary calculation) • Job classification • Agency of employment
Ac
ce
pt
Labor Force Information
Health Insurance Information
• Marital status (unlikely to be complete) • May have beneficiary information for current retirees.
• Plan choice/type • Utilization (claims information)
• Use tenure and job code measures to impute benefit information
15 Page 15 of 15
S0167-6296(14)00091-5 http://dx.doi.org/doi:10.1016/j.jhealeco.2014.06.011 JHE 1789
To appear in:
Journal of Health Economics
Received date: Accepted date:
23-6-2014 24-6-2014
Please cite this article as: Morrill, M.S.,Active and Retired Public Employees’ Health Insurance: Potential Data Sources, Journal of Health Economics (2014), http://dx.doi.org/10.1016/j.jhealeco.2014.06.011 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Active and Retired Public Employees’ Health Insurance: Potential Data Sources
This Version: June 16, 2014*
ip t
Melinda Sandler Morrill, North Carolina State University
cr
In July 2013, public workers, including state, local, and federal government employees,
accounted for about 15 percent of the U.S. non-farm labor force.1 Clemens and Cutler ()
us
estimated that state and local governments spent $117 billion on health insurance in 2010 alone. Health insurance represented about 12.2 percent of the total compensation cost for state and local government workers in June 2013.2 Underfunding and growing costs of benefits have implications for public policy
an
and the fiscal solvency of state and local governments. Furthermore, findings from the private sector may not be directly relevant in the public sector because many public sector workers are covered by union
M
contracts or fixed salary schedules and often benefit modifications require changes in legislation. Thus, from both an academic and a policy perspective, understanding health care costs and health insurance in the context of public sector workers is of paramount importance.
d
There are many important questions associated with employer-provided health insurance and
te
retiree health insurance for public sector workers. For example, we might wish to know what a given state or local employer spends on providing health insurance benefits to active and/or retired employees.
Ac ce p
Is the cost of health insurance and/or retiree health insurance passed on to workers? How do the benefits affect employee behavior? How will the Affordable Care Act affect state and local governments’ employee health insurance provision? While articles referenced herein attempt to address some of these research questions, much is still not known about this expensive and potentially important employee benefit in the public sector. One reason that these topics are relatively understudied is that it is difficult to obtain sufficiently large and representative data focusing on public sector employees. This article outlines some potential data sources that interested researchers might utilize to investigate topics concerning employer-provided health insurance for active and retired public sector employees. *
Melinda Morrill, Department of Economics, Box 8110, North Carolina State University, Raleigh, NC 27695-8110. Email: [email protected]. I would like to thank Robert Clark, Jeffrey Clemens, David Cutler, Maria Fitzpatrick, Joseph Newhouse, and Thayer Morrill for useful discussions and comments. Chrystelle Khalaf provided excellent research assistance. 1
See Bureau of Labor Statistics, Current Employment Statistics, http://www.bls.gov/web/empsit/ceseeb1b.htm, [accessed October 29, 2013]. 2
See Bureau of Labor Statistics, Economic News Release, http://www.bls.gov/news.release/ecec.t04.htm, [accessed October 29, 2013].
1 Page 1 of 15
I. Nationally-representative individual-level data Many commonly-used, nationally-representative, individual-level datasets allow for the identification of public sector employees. To be a useful data source for studying health benefits in the public sector, a dataset should contain some information on health insurance coverage, earnings/wages,
ip t
employment status, and sector of employment. While it is not uncommon to observe health insurance coverage, it is relatively rare for a dataset to include information on the parameters and generosity of an employee’s health plan. However, if we observe that an individual is an employee of a state government
cr
and we know that employee’s state of residence, we can match publicly available information about health insurance offerings to the survey or administrative data. Thus, it is particularly useful if a
us
nationally-representative dataset includes state of residence or other geographic identifiers.3 In this section, we outline several data sources that separately identify federal, state, and local
an
government employees.4 This is not an exhaustive list but attempts to cover the largest and most prominent datasets. As a basis of comparison, Table 1 presents a description of data availability for health insurance, health status, and geography. The table then presents sample sizes for a sample of full-
M
time, currently employed federal, state, and local government workers.5 First, counts are presented for workers ages 18 to 64; then, counts are presented separately for federal, state, and local workers between ages 50 and 64.
d
I.A. Health and Retirement Study
te
By far the most common survey for studying older workers and retirees is the Health and Retirement Study (HRS).6 The HRS is a panel dataset that began in 1992. The 2010 data release includes
Ac ce p
ten waves of data and five entry cohorts.7 The most important advantage of the HRS is the wealth of detail available both from the respondent and from merged employer data in some years. In addition, the panel nature allows one to study past employment (e.g., Shoven and Slavov, ). The more 3
For example, one might use the classifications of retiree health insurance percent of the premium paid available in the GAO (2007) report, data found in actuarial reports, or data available on state websites. While theoretically one might do a similar matching on county for local government employees, no systematic data exist on county and city health plans. Moreover, as state and local health plans may differ, it may be difficult to assign a survey respondent to the correct health plan information once collected. For example, in North Carolina teachers are state government employees and part of the State Health Plan, but might erroneously report on a survey being a local employee. 4
The U.S. Census Bureau provides several data sources that contain health information, as summarized at: http://www.census.gov/sdc/healthstats_JCD.pdf, [accessed October 29, 2013]. Census Bureau data typically includes a variable on “class of worker” which identifies the sector of employment. 5
When possible, we define “full time” as working at least 40 weeks per year, at least 35 hours per week, and currently working for pay. 6
For information on accessing these data see: http://hrsonline.isr.umich.edu/, [accessed October 29, 2013].
7
The entry cohorts are 1992 HRS cohort; the 1993 Study of Assets and Health Dynamics (AHEAD) cohort; the 1998 Children of Depression cohort; the 1998 War Babies cohort; and the 2004 Early Baby Boomer cohort.
2 Page 2 of 15
recent years of data separately identify federal, state, and local government workers.8 In these data, health insurance coverage, health costs, and health status are all observed over time. However, geographic information in more detail than Census region is only available in a restricted-access version of the dataset.
ip t
As illustrated in Table 1, the most significant limitation of the HRS is the small sample size. Using the 2010 data release, which includes all respondents and their spouses from all cohorts that were still alive at the time of the survey, only 1,049 observations out of the 4,863 full-time, currently employed
cr
individuals are public sector employees. When we further restrict the sample to those between the ages of 50-64, the sample size is only 954 workers.
us
The HRS is particularly useful when a researcher requires detailed information on wealth/assets or health and health behaviors. For example, Clark and Mitchell () consider how access to
an
and generosity of health insurance affects retirement savings. In another example of how the HRS might be used to study health insurance and public sector employees, Slavov and Shoven () consider how the availability of employer-provided health insurance and retiree health insurance affects retirement
M
behavior.
I.B. Survey of Income and Program Participation The Survey of Income and Program Participation (SIPP) is a large, nationally-representative
d
dataset collected by the Census Bureau since 1984 with extensive information on sources of income.9
te
The sample design includes short panels (between two and four years), with new samples drawn every four years (prior to 1996 new panels began every year). The panels range in size from 14,000 to 36,700
Ac ce p
and are each nationally-representative. The SIPP data identify separately individuals employed in the private sector (for profit versus non-profit) and local, state, or federal government employees. The SIPP includes topical modules that are fielded in various waves for some panels. The Medical Expenses/Utilization of Health Care module includes health status and some measures of medical services utilization. Another topical module includes information on Employer Provided Health Benefits. Also particularly relevant may be the Retirement and Pension Plan Coverage module, which even includes data on work interruptions for family care. The panel structure allows for the data included in the topical modules to be merged across waves, so that one can construct a dataset that looks at
8
The 2006, 2008, and 2010 waves separately identify state or local workers, while previous waves grouped them together. See the data appendix of Clark and Mitchell (2013) for a methodology for separately identifying state and local workers using the RAND version of the HRS. 9
For more information and to access the SIPP, see: http://www.census.gov/sipp/, [accessed October 29, 2013].
3 Page 3 of 15
employment transitions within one panel or information from multiple waves of one panel may be collapsed and used as a detailed cross-section.10 Besides the wealth of detail available in the topical modules, a real benefit of the SIPP is the large sample size combined with a panel structure (albeit for only 2.5 years in more recent panels). For Table
ip t
1, we include statistics only for the 2008 Panel, Wave 1. Thus, if one wanted to include all available panels the sample size would be substantially larger. The total number of full-time federal, state, or local government employees in the 2008 Panel is approximately 23,325. When restricting to only workers
cr
nearing retirement, ages 50-64, the total sample of public sector workers is 8,623. Thus, in only one panel the sample size is nearly ten times that of the more commonly used HRS.
us
I.C. Current Population Survey
The Current Population Survey (CPS) is sponsored jointly by the U.S. Census Bureau and the
an
U.S. Bureau of Labor Statistics.11 Each sample is about 60,000 households who are surveyed each month for 4 months, then out of the sample for 8 months, and then back in the survey for 4 months. The monthly data are used by the BLS to provide labor force statistics, such as the unemployment rate. There
M
are supplements in several of the months.12 The most commonly used supplement for labor economics research is the Annual Social and Economic Supplement conducted in March. One particular advantage of the March CPS is the inclusion of a measure of the employer’s contribution towards the employee’s
d
health insurance. Since 1996 the data also include a five point health status scale. In addition, the 2010
te
CPS includes a new set of questions on medical out-of-pocket expenses. Table 1 presents statistics for the March CPS pooling data from 2008 through 2012. Using just these five years of data, we see the total
Ac ce p
sample of public employees is 57,237, including 19,930 public employees ages 50-64.13 I.D. American Community Survey The American Community Survey (ACS), reported annually since 2006, replaces the long-form of the decennial Census.14 The ACS is released as 1-year, 3-year, or 5-year pooled estimates with more
10
Perhaps more so than the other datasets considered here, the SIPP panels have suffered from attrition problems. Although I know of no studies examining this issue in particular, it is possible that attrition problems are particularly problematic when studying life transitions between waves, such as retirements.
11
For information and access to the CPS, see: http://www.census.gov/cps/, [accessed October 29, 2013].
12
For information on available supplements, see: http://www.census.gov/cps/about/supplemental.html, [accessed October 29, 2013]. 13
Qin and Chernew () use the Merged Outgoing Rotation Group (often referred to as the MORG) CPS from 1992-2011 for 25-54 year old males to measure the compensating wage differentials from the rising costs of employer-provided health insurance in the public sector.
14
For more information and access to the ACS, see: https://www.census.gov/acs/www/, [accessed October 29, 2013].
4 Page 4 of 15
geographic detail and data for smaller geographies available for the pooled samples.15 Health insurance information has been collected annually since 2008. The information on health insurance concerns only coverage and source, and there are no other measures of health behavior or health status beyond measures of work-limiting disabilities. For Table 1, we present the most recently released 3-year estimates, 2009-
ip t
2011. There are approximately 540,258 full-time federal, state, or local government employees, with 208,586 between the ages of 50-64 included in the 2009-2011 3-year ACS. These data would be
particularly useful for researchers with supplemental information on health insurance at the state or local
cr
government employer level, since the pooled data releases allow for the identification of small geographic areas.
us
I.E. Medical Expenditure Panel Survey
The Medical Expenditure Panel Survey (MEPS) began in 1996 and is conducted by the Agency for Healthcare Research and Quality.16,17 The MEPS has several components including a Household
an
Component (HC), which has a publicly available version.18 The MEPS-HC is an overlapping panel design. A new sample is drawn annually from the prior year’s NHIS, and each sample is followed for
M
two calendar years collected in 5 rounds.19
The MEPS-HC collects detailed information on health expenditures. Unlike other health-focused datasets, the MEPS includes a “job file” supplement on the respondents’ employment history and current
d
status, including the “employee type” of private, federal, state, or local employee. Because of its focus on
te
health care expenditure, this dataset provides the most detail on health insurance, payments, and health behaviors and outcomes. However, the public use version of the data does not identify the state of
15
Ac ce p
residence. Table 1 provides sample sizes for the 2011 MEPS Household Component survey.20 Note that The Census Bureau provides the following guidance for selecting which sample to use: https://www.census.gov/acs/www/guidance_for_data_users/estimates/, [accessed October 29, 2013].
16
For more information and access to the household component data, see: http://meps.ahrq.gov/mepsweb/, [accessed October 29, 2013].
17
The sample for the MEPS is drawn from the much larger National Health Interview Survey (NHIS), conducted by the Center for Disease Control and Prevention, National Center for Health Statistics. The NHIS is described in more detail in Section I.F. For researchers interested in data from earlier time periods, the National Medical Care Expenditure Survey (NMCES), National Medical Care Utilization and Expenditure Survey (NMCUES) and the National Medical Expenditure Survey (NMES) are predecessor surveys to the MEPS. Information on these sources can be found at: http://www.icpsr.umich.edu/icpsrweb/ICPSR/series/45, [accessed December 13, 2013].
18
The MEPS also includes an Insurance Component (IC), which is a restricted-access only dataset that includes extensive detail gathered from both public and private sector employers on their health insurance plans. Summary statistics from these data are available at the state and metropolitan area for private sector employers, but only at the national-level or by Census division for public sector employers.
19
For an illustration of the data collection process, see: http://meps.ahrq.gov/mepsweb/survey_comp/hc_data_collection.jsp [accessed October 29, 2013].
20
The 2011 data include panel 15, rounds 3-5, and panel 16 rounds 1-3.
5 Page 5 of 15
this is just one year of data, and new panels are drawn every year. So, one could create a substantially larger dataset by pooling across years/panels. The total numbers of full-time public employees in the 2011 MEPS are 1,510 ages 18-64 and 543 ages 50-64. I.F. Other Datasets
ip t
Although not included in Table 1, three other datasets warrant some mention. First, the National Health Interview Survey (NHIS) is a large, nationally-representative dataset that includes a wealth of health-related information.21 The NHIS is collected annually by the CDC’s National Center for Health
cr
Statistics for the purpose of producing statistics on disease incidence and prevalence. The data are
nationally-representative, but state identifiers are only available in the restricted-access version. There
us
are over 50,000 households surveyed annually. For one adult in the household, information on sector of employment (separately identifying federal, state, and local government workers) is available. There is
an
health insurance information, including out-of-pocket premium payments, for all household persons. The survey asks about employer/union paid contributions to health insurance, but over two-thirds of the sample report not knowing. While the NHIS includes substantial detail on health status and medical care
M
utilization, there is less detail on employment and health expenditures than is available in the MEPS, as described above.
The National Longitudinal Survey of Youth, 1979 cohort (NLSY-79) is an ongoing panel of a
d
nationally-representative sample of 12,686 individuals who were ages 14-22 in 1979.22 Again, these data
te
identify federal, state, and local employees. In addition to information on health insurance coverage, there is a supplemental health module for individuals ages 40 and older and one for 50 and older. While
Ac ce p
the panel structure and detail available make these data attractive, the sample size is limiting and the data represent only one cohort.
Finally, the Panel Study of Income Dynamics (PSID) is the longest running longitudinal household survey having begun in 1968.23 The data have grown over time because spouses and children of initial survey respondents are incorporated into the sample, in addition to immigrant refresher samples. The PSID has detailed information on employment, wealth/assets, health, and health insurance. However, the sample size is relatively small compared with the other data sources listed here. In the 2011 wave, there were a total of 14,607 individuals and 5,495 families. 21
For information and access to the public-use version of the NHIS, see: http://www.cdc.gov/nchs/nhis.htm, [accessed November 3, 2013].
22
For general information and access to the NLSY, see: http://www.bls.gov/nls/nlsy79.htm. For more detailed information about the health insurance variables, see: https://nlsinfo.org/content/cohorts/nlsy79/topical-guide/health; and for more information about the employment variables, see: https://nlsinfo.org/content/cohorts/nlsy79/topicalguide/employment, [accessed November 3, 2013]. 23
For information and access to the PSID, see: http://psidonline.isr.umich.edu/, [accessed November 3, 2013].
6 Page 6 of 15
II. Nationally-representative proprietary data and surveys The following resources are based on micro-level data, either at the employee or employer level, but only aggregate statistics are publicly available. It may be that interested researchers could access the underlying microdata through a special agreement with individual agencies, but for the purposes of this
ip t
article we will only discuss publicly available versions. The level of aggregation of publicly available data varies, but the data are typically broken out by state, type of employer (public versus private sector), and/or number of workers at the firm. State-level aggregation may be as detailed as necessary when
cr
considering variation between state government employers. In addition, these data are useful for
observing trends and for benchmarking smaller datasets that are not necessarily nationally-representative.
relevant websites and a basic summary of the data available.
us
Table II presents four prominent surveys as examples of this type of data. The table includes links to the
an
(1) National Conference of State Legislators: State Employee Health Benefits
The National Conference of State Legislators (NCSL) releases reports on state employee health benefits.24 These data include state-level information on the lowest cost plan and a standard policy
M
insurance option. Clemens and Cutler (2013) and Qin and Chernew () provide examples of how these data might be used in practice. These papers utilize data on plan costs and state versus employee contributions collected by the NCSL to determine the percent of the premium that state
d
governments pay for active worker’s health insurance.
te
(2) Bureau of Labor Statistics: National Compensation Survey The Bureau of Labor Statistics (BLS) conducts the National Compensation Survey (NCS)
Ac ce p
annually, which includes both private and non-federal public sector employers.25 The data include information from establishments on wages and benefits. The BLS produces several series using these data including the Employer Costs of Employee Compensation (ECEC) and Employment Cost Index (ECI). NCS Databases provide national-level estimates broken out by sector (state, local, or state/local
24
For more information on these data, see: http://www.ncsl.org/research/health/state-employee-health-benefitsncsl.aspx, [accessed November 4, 2013]. Monthly premium costs for individual and family coverage can be found in the following reports, [accessed June 12, 2014; report links sent by Richard Cauchi, Program Director, Health Program, National Conference of State Legislatures]: 2009 (with information on 2006 and 2009): http://www.ncsl.org/portals/1/documents/health/FamilyPrem09.pdf and http://www.ncsl.org/portals/1/documents/health/IndivPrem09.pdf. 2011: http://www.ncsl.org/Portals/1/documents/health/StateEHBenefits2011.pdf 2012: http://www.ncsl.org/documents/health/2012NCSLStateEmployeeHealthBenefits.pdf 2013: http://www.ncsl.org/documents/health/2013NCSLStateEmployeeHealthBenefits.pdf 25
For more information on these data, see: http://www.bls.gov/ncs/, [accessed November 4, 2013].
7 Page 7 of 15
combined) of measures such as health insurance spending, coverage of health insurance, and provision of other benefits.26 BLS provides on-site access to the NCS data.27 (3) The Segal Company: Study of State Employee Health Benefits The Segal Company conducts annual survey titled the “Study of State Employee Health Benefits.”28
ip t
The Segal Group publishes an annual report summarizing the findings. On the last page Segal states, “Segal can be retained to provide custom data reports, including comparisons of coverage costs among
describe the average state offerings of health insurance plans in 2012.
cr
plan types, regions, and/or population size.” Clemens and Cutler (2013) use Segal’s 2013 report to
(4) Kaiser Family Foundation/Health Research and Educational Trust: Employer Health Benefits
us
Survey
The Kaiser Family Foundation/Health Research and Educational Trust (KFF-HRET) conducts an
an
annual survey of private sector and non-federal public sector employers.29 These data include not only premiums and health benefit information for active workers, but they also separately analyze retiree health benefits. The annual reports have been used by researchers to understand health insurance trends
M
in the private and public sectors (e.g., Clemens and Cutler 2013). Micro-data are available for interested researchers by application only, but identifying information (including state identifiers) is removed in the
III. Teachers and schools data
d
public use data file.
te
Teachers and school employees represent almost half of all state and local government workers nationally.30 While these workers may not be representative of public sector workers in general,
Ac ce p
researchers have found focusing on schools and teachers to be a fruitful exercise, particularly because the data for this group are more readily available. The U.S. Department of Education’s National Center for Education Statistics annually collects the Common Core of Data (CCD).31 These data provide detail on individual schools, school districts, and state-level education agencies. The data include aggregated information on demographics, student outcomes, and both revenues and expenditures. The CCD includes several component surveys. Clemens 26
To create customized tables, see: http://www.bls.gov/ncs/data.htm, [accessed October 29, 2013].
27
For more information about accessing the NCS, see: http://www.bls.gov/bls/blsresda.htm.
28
A report describing findings from the 2012 survey can be found at: http://www.segalco.com/publications/surveysandstudies/2012statestudy.pdf, [accessed October 29, 2013].
29
A report describing findings from the 2013 survey can be found at: http://kff.org/private-insurance/report/2013employer-health-benefits/, [accessed November 4, 2013]. 30
See Bureau of Labor Statistics, Current Employment Statistics, http://www.bls.gov/web/empsit/ceseeb1b.htm, [accessed October 29, 2013].
31
For more information on these data, see: http://nces.ed.gov/ccd/, [accessed November 4, 2013].
8 Page 8 of 15
and Cutler () use CCD data from the School District Finance Survey, which has detail on school district revenues and expenditures that can be matched to employment data. Unfortunately, health benefit and pension costs are grouped together in these data.32 In addition to these nationallyrepresentative datasets, individual states often collect and track data on teachers and schools, as described
IV. Administrative records for states or large local governments
ip t
in Section IV below.
Almost all of the data sources described thus far are nationally-representative samples, and almost
cr
all are collected by national agencies or organizations for the purposes of documenting trends and/or
conducting research. An alternative source of data is the administrative records a public employer or
us
health insurance company maintains. Focusing on only one (albeit large) employer reduces the amount of potential policy variation to exploit. But, detailed information for one large employer does allow for the
an
more precise measurement of the benefits provided and often yields a large dataset. The structure of administrative records generally allows for the creation of a panel, so that one can track how individuals who remain with their employer behave over time when exposed to different policies. The most
M
important limitations of this type of data are the lack of detail on family structure and the difficulty in tracking behavior after health insurance coverage and/or employment is terminated. Public employers with defined benefit pension plans must maintain an up-to-date data file with
d
information on active and retired employees to submit to plan actuaries for the calculation of pension
te
liabilities annually. Nearly identical data, along with information on health plan offerings, is provided to actuaries to calculate liabilities associated with retiree health insurance, following Government
Ac ce p
Accounting Standards Board Rule 45 (GASB, 2004). Similarly, any health insurance plan maintains records of its members and must process medical claims. Although these data are maintained for the purposes of administering and valuing a plan, they may be a useful resource for research. Table 3 illustrates some variables that would commonly be found in these types of data. Each state or local employer likely collects slightly different data fields or may treat certain variables as particularly sensitive.33 Researchers interested in utilizing these types of data must apply to a state administrator directly.34 Ideally, a researcher could merge data from a health plan and pension plan to create a rich
32
The National Center for Education Statistics has begun a pilot project, the Teacher Compensation Survey, to collect better detail on benefits. For more information on the early efforts, see: http://nces.ed.gov/ccd/tcssurv.asp [accessed November 18, 2013].
33
These data sources likely have personal identifiers such as names, addresses, and Social Security numbers. Health information beyond simple plan choice is protected under HIPAA and will typically involve a much more thorough application and review process than information about employment or salary.
34
Because pension plans are paid through public dollars, some may even consider these data part of the public record and subject to a Freedom of Information Act request (see, e.g., Leiserson, 2013).
9 Page 9 of 15
dataset. However, typically these data are collected by separate agencies, presenting significant legal and administrative hurdles. Recent research highlights how state-level administrative records can be used in practice. Using data maintained by a state health plan administrator, BlueCross BlueShield of North Carolina, Clark,
ip t
Morrill, and Vanderweide () consider how changing health plan characteristics over time affect members’ choice of health plan. Fitzpatrick () uses administrative files from the Illinois Public Schools, referred to as the Teacher Service Record, which contains information on
cr
experience, salary, position, assignment, and school. She combines the administrative records with statelevel policy changes on retiree health insurance availability, exploiting the detail on tenure to measure
us
eligibility for the program. Leiserson (2013) uses data on public sector workers included in the state health plan from the Pennsylvania State Employees’ Retirement System combined with information on
an
retiree health insurance policies over time. These studies demonstrate that administrative records from just one state, although not detailed, combined with information on policies and policy changes can be a powerful tool in understanding how health insurance relates to labor market outcomes such as
M
employment, retirement, and wages.
To understand the cost of and liabilities associated with providing health benefits, particularly to retirees, researchers can access reports prepared by plans to be in compliance with the GASB 45 ruling
d
(GASB 2004). The Comprehensive Annual Financial Reports (CAFRs) typically have extensive
te
information about the financial status of a public sector employer’s pension and retiree health plans. A researcher might also access the actuarial reports directly, which provide much more detail on the
Ac ce p
assumptions underlying the calculations of liabilities associated with retiree health insurance (typically these reports are referred to simply as “GASB reports”). Clark and Morrill (2010) provide more information on the data included in the actuarial reports. Lutz and Sheiner () illustrate the value of gathering the CAFRs and GASB reports to provide data on plan funding and actuarially accrued liabilities. These reports can be obtained from individual states and local governments, but are not systematically available in one place. V. Discussion and Conclusions
To date, researchers interested in studying issues surrounding health insurance and/or retiree health insurance in the public sector have had to confront limitations in data availability. This article highlights some potentially useful data sources, noting some studies using these data. Many of these data sources, of course, have been used to study similar questions with respect to private sector employees. Nationally-representative, individual-level data often includes indicators for sector of employment and at least some information on health insurance coverage. For state government employees or employees of large cities or counties, often supplemental data on benefits can be merged on using geographic 10 Page 10 of 15
identifiers. Similarly, aggregated data at the state-level may be useful for cross-state analyses of state government workers. Data on public schools and teachers may be more detailed and readily available than other segments of the public sector labor market. And, finally, state-specific administrative records from pension plans or health plans may be a useful resource for researchers. These administrative
ip t
datasets include large samples and high-quality data. Thus, as seen in articles published in this issue, there are sufficient data to conduct insightful, policy-relevant research on topics related to health
insurance provision by public sector employers. Moreover, this is an under-researched area with the
cr
potential of uncovering findings that would improve the fiscal sustainability of state and local
governments. This article is not meant to be a comprehensive list of available data, but it should provide
us
interested researchers a starting point for exploration.
an
References
Clark, Robert and Olivia Mitchell. (2013) “How Does Retiree Health Insurance Influence Public Sector Employee Saving?” NBER Working Paper #19511.
M
Clark, Robert and Olivia Mitchell. () “How Does Retiree Health Insurance Influence Public Sector Employee Saving?” Journal of Health Economics.
d
Clark, Robert L. and Melinda Sandler Morrill. 2010. Retiree Health Plans in the Public Sector: Is There a Funding Crisis? Northhampton, MA: Edward Elgar Publishing.
te
Clark, Robert, Melinda Morrill, and David Vanderweide. () “The Effects of Retiree Health Insurance Plan Characteristics on Retirees’ Choice and Employers’ Costs.” Journal of Health Economics.
Ac ce p
Clemens, Jeffrey and David Cutler. (2013) “Who Pays for Public Employee Health Costs?” NBER Working Paper #19574. Clemens, Jeffrey and David Cutler. () “Who Pays for Public Employee Health Costs?” Journal of Health Economics. Fitzpatrick, Maria. () “Retiree Health Insurance for Public School Employees: Does It Affect Retirement and Mobility?” Journal of Health Economics. Government Accountability Office (GAO). (2007). “State and Local Government Retiree Health Benefits: Current Status of Benefit Structures, Projections, and Fiscal Outlook for Funding Future Costs.” Report to the Committee on Finance, US Senate. September GAO-07-1156. http://www.gao.gov/new.items/d071156.pdf. Government Accounting Standards Board. 2004. Statement No. 45. Accounting and Financial Reporting by Employers for Post-employment Benefits Other Than Pensions. Leiserson, Gregory, (2013), “Essays on the Economics of Public Sector Retirement Programs,” MIT Dissertation, June 6, 2013. Lutz, Byron and Louise Sheiner. () “Examining the Whole Picture: Retire Health Obligations and the Long-Term Budget Outlook for the State and Local Government Sector.” Journal of Health Economics.
11 Page 11 of 15
Qin, Paige, and Michael Chernew. (). “Compensating Wage Differentials and the Impact of Health Insurance in the Public Sector on Wages and Hours,” Journal of Health Economics.
Ac ce p
te
d
M
an
us
cr
ip t
Shoven, John and Sita Slavov. () “The Role of Retiree Health Insurance in the Early Retirement of Public Sector Employees.” Journal of Health Economics.
12 Page 12 of 15
i cr Variables Included in the Data
Example Dataset: Full-Time Public Sector Employees
an
Dataset
us
Table 1: Nationally-representative individual-level data
No health behaviors Health insurance coverage and source only.
ce
March Current Population Survey (CPS)
ed
Survey of Income and Program Participation (SIPP)
Includes: detailed health insurance; health status; medical service utilization Includes: detailed health insurance; health status; medical service utilization (some years) Includes: detailed health insurance; health status
pt
Health and Retirement Study (HRS)
Sample & Years
Restrictedaccess only
2010 data, uses all cohorts
Full-Time, Currently Employed Ages 18-64 Total: 4,863 Public: 1,049
Yes
2008 Panel, Wave 1 only
Total: 124,298 Public: 23,325
Yes
2008-2012
Total: 338,926 Public: 57,237
Yes, detailed geography available Restrictedaccess only
2009-2011 3year estimates
Total: 2,889,703 Public: 540,258
Public Sector Employees Ages 50-64 Federal: 206 State: 383 Local: 365 Federal: 1,879 State: 2,735 Local: 4,009 Federal: 4,359 State: 6,285 Local: 9,286
Federal: 42,957 State: 64,801 Local: 100,828 Medical Expenditure Includes: detailed health 2011 Total: 56,907 Federal: 97 Panel Survey (MEPS) insurance; health status; Public: 1,510 State: 236 medical service utilization Local: 210 Notes: The example dataset includes only individuals that are currently employed full-time and are age 18-64. The final column further restricts this sample to include only federal, state, or local government employees ages 50-64. All datasets, excepting the HRS, have additional years of data (or panels) available.
Ac
American Community Survey (ACS)
State of Residence
M
Health & Health Insurance
13 Page 13 of 15
i cr us an
Table 2: Aggregated Employee Benefit Studies
Survey Title and Years
National Conference of State Legislatures
State Employee Health Benefits Annually 2001-2006, 2009, 2011-2012
Bureau of Labor Statistics, National Compensation Survey
Employee Benefits Survey (1985-2006)
http://www.bls.gov/ncs/ebs/
National Compensation Survey – Benefits (2010 – 2013)
Series produced as: “Employer Costs for Employee Compensation” http://www.bls.gov/news.release/ecec.htm
Kaiser Family Foundation
ce
pt
ed
M
Agency
Health Research and Education Trust
Segal
Ac
Annual Surveys 1999-present Annual Study of State Employee Health Benefits
Website http://www.ncsl.org/research/health/stateemployee-health-benefits-ncsl.aspx
http://kff.org/privateinsurance/report/2013-employer-healthbenefits/ http://www.segalco.com/publications/ surveysandstudies/2012statestudy.pdf
Health Insurance Information State-level information on lowest cost and standard policy insurance options. Only national-level estimates, broken out by sector (state, local, or state and local).
Survey of private and nonfederal public firms; includes separate industry category for state/local government. Reports statistics on state employee health benefits.
14 Page 14 of 15
i cr us an
Table 3: Typical Administrative Data Elements
Pension or Employment Database
Basic Demographics
• Date of birth • Gender
• Date of birth • Gender • (Teachers data) highest degree attained
Family Structure
• Coverage of dependents • Age and gender for covered dependents
ed
M
Health Insurer Database
• Date of first coverage • Date of last coverage
• • • •
Date of hire Date of separation Date of retirement Years of service (contributory, non-contributory, purchased, calculated) • Salary (may have panel of earnings or information for final average salary calculation) • Job classification • Agency of employment
Ac
ce
pt
Labor Force Information
Health Insurance Information
• Marital status (unlikely to be complete) • May have beneficiary information for current retirees.
• Plan choice/type • Utilization (claims information)
• Use tenure and job code measures to impute benefit information
15 Page 15 of 15
