Tools for Monitoring the Health Care Safety Net

Estimating the Size of the Uninsured and Other Vulnerable Populations in a Local Area

By Lynn A. Blewett, Ph.D. and Timothy Beebe, Ph.D.


Contents

Introduction
Federally Sponsored National Surveys
Privately Sponsored National Surveys of Note
State Survey Efforts
Approaches for Local Estimation of Safety Net Demand
Comments on Estimating State Health Care Program Coverage
Summary of Issues and Recommendations
Conclusions
References

Introduction

The purpose of this paper is to focus on the methods used to measure the demand or need for safety net services. The amount of demand or potential changes in demand for safety net services is dependent on several factors, but one key component is the number of persons without health insurance coverage. This estimate of demand at the national level shows that demand continues to grow. The most recent estimates show an increase in the number of uninsured from 2000 to 2001 of 1.4 million people. There are now 41.2 million uninsured in the United States representing 14.6 pecent of the population (Mills, 2002). But drilling down to the State level shows significant variation in these rates. The number of uninsured individuals varies dramatically by State from a low of 7.2 percent in Rhode Island and 7.8 percent in Minnesota to a high of 23.2 percent in New Mexico and 23.0 percent in Texas.1

The number of uninsured in each State varies based on geographic area, such as urban versus rural or by region or county. Unfortunately, there are limited data sources to draw on to provide comparable estimates at the local and/or regional levels across the United States. It is important, however, to understand the data that are available at the State and national levels.

First, many State and national data sources are the only sources of information on the uninsured that are produced annually and consistently. State and national data can be used to provide benchmarks for localities in understanding the dynamics of health insurance coverage and in testing their own direct survey results of local health insurance coverage against State estimates.

Second, many localities may develop either proxy or model-based estimates that will rely on existing State and national survey data. Given that collecting survey data is costly and time-consuming, using existing data to develop proxy estimates may be the only available alternative. Finally, several States and Federal agencies are using their data and considerable expertise to develop better State and local estimation of coverage rates. It is important to understand the data that currently exist, the mechanisms used to collect this data, and the potential use of existing State and national data for local estimates of demand for safety net services.

In this regard, we begin first with an overview of the current State and national data available to estimate health insurance coverage and options available to localities for estimating local safety net demand using these and other sources of data. While acknowledging a broader set of populations who use safety net services, our paper focuses on:

We discuss three ways in which localities can estimate safety net demand by:

  1. Directly measuring uninsurance through the development and implementation of a local household survey.
  2. Using proxy measures from other data sources as an indirect means to estimate levels of demand.
  3. Developing statistical models using data from the local area combined with State and national data sources to develop what we refer to as model-based approaches to estimating demand.

Each approach is reviewed in turn with a discussion of the advantages and disadvantages and recommendations for analysts interested in pursuing local estimates of safety net demand.


1 State estimates are based on 3-year average 1999-2001 CPS data for percent of population without health insurance coverage for the entire year.


Return to Contents

Federally Sponsored National Surveys

The best way to obtain information on health insurance coverage is to ask people directly, commonly through a household survey. Another method is to assess the extent of employer-based health insurance coverage through a survey of employers. In this section, we describe the key surveys used to measure and monitor rates of health insurance coverage. It is difficult and often impossible to obtain local estimates of health insurance coverage using national survey data. The sample sizes for individual areas covered by aggregate surveys are small. As a result of these small sample sizes, the standard errors of local estimates are large, thus limiting the value of these estimates at the local level. However, it is important for analysts to understand where the State and national estimates come from and how these estimates are derived and used. Again, this survey data can be used to understand the dynamics of health insurance coverage and as benchmarks to assess current estimates at the local level.

Two Federal agencies, the U.S. Census Bureau and the Agency for Healthcare Research and Quality (AHRQ), are developing model-based estimates of uninsurance at the region and county level based on:

Each is discussed in turn.

Data from the Current Population Survey are used at the national level to compare State uninsurance rates, as well as in the Federal funding formula used to distribute State funds under the State Children's Health Insurance Program (SCHIP). Also, if State or local analysts pursue a model-based approach to local estimates of health insurance coverage, it is quite likely that one of these data sources will be used as part of the modeling exercise.

Measuring the demand for safety net services may benefit from the use of data from national surveys that focus on measuring health insurance coverage rates. We focus here on national surveys that include a sample that is State representative. This means that States were used as a stratifier in selecting which households to interview and that a sufficient sample was drawn within each State to make a direct estimate of coverage rates at the State level. While there are many national surveys with nationally representative samples that assess health insurance coverage, there are relatively few national surveys that are State representative. The most often cited is the Current Population Survey (CPS), but other surveys include:

These and other national surveys have been described in detail in other places. Consequently, we highlight only key points related to estimating health insurance coverage rates and implications for State and local population estimates of uninsurance rates. (For more information, refer to: Bindman and Gold, 1998; Eden 1998; Fronstin, 2000; Gold 1998; Lewis, 1998; Liska, Brennan, and Bruen, 1998; SHADAC, 2001. Also, links to the respective Web sites and to other national surveys are provided in Appendix A.)

Other national surveys that emphasize health insurance information but are not State representative include the Medical Expenditure Panel Survey-Household Component (MEPS-HC)2, the National Health Interview Survey (NHIS), and the Survey of Income and Program Participation (SIPP). Although important surveys in their own right, they are less applicable to State and local estimation of the uninsured.


2 The MEPS-HC has included selected large States in its sample redesign to attempt to obtain some direct State estimates, but published State estimates are not available. Currently, adjusted State NHIS data files are available from only 1990 through 1994 (Madans et al., 2001).


Current Population Survey

The CPS is the most commonly cited source of data for estimating health insurance coverage. This survey is conducted monthly by the U.S. Census Bureau for the Bureau of Labor Statistics. Its primary objective is to measure labor force participation and unemployment, and the data are used to produce monthly unemployment figures for the United States as well as for each State individually. The annual demographic supplement to the CPS which is conducted each March, includes questions on income, (and the income-derived measure of poverty), household composition, and health insurance status. Due to the inclusion of the latter questions, coupled with its State and national representation, the CPS has become one of the most commonly used data sources for estimating rates of health insurance coverage at both the Federal and State levels, even though such measurement is not its principal purpose.

Technical Issues with Using CPS To Measure Health Insurance Coverage

While the CPS was originally intended to provide national estimates and trends over time, policy analysts began using the CPS to derive State estimates of health insurance coverage. Noted limitations to the usefulness of the CPS for State-level estimates are that the State sample size is relatively small and the sampling frame does not include all counties within the State. An advantage of the CPS is that it does include a State representative sample. It also includes in-person as well as telephone surveys, increasing the potential response of low-income and uninsured participants and reducing the bias associated with surveys that rely primarily on a telephone mode of data collection.

The Census Bureau has employed several methodological innovations over the years designed to enhance the health insurance coverage information the survey provides (Liska, Brennan, and Bruen, 1998; Swartz 1997; SHADAC, 2001). Included in these improvements was the addition of a verification question, which confirms a person's response on what type of health insurance coverage he or she has. The addition of this question in March 2000 reduced the estimated uninsurance rate across States by an average of 7 percent (Nelson and Mills, 2001). Other recent improvements to the CPS will also improve its ability to report State estimates, including the inclusion of State-specific program names in the survey (e.g., Medicaid is referred to as Medi-Cal in California, Medical Assistance in Minnesota) as well as the addition of a catch-all question asking about participation in some State-run programs (e.g., TennCare and MinnesotaCare).

Another improvement was the sample size expansion undertaken to improve State estimates of poor uninsured children, as mandated by the Balanced Budget Refinement Act of 1999. Although the CPS has been used to produce a variety of State health insurance coverage estimates for policy analysis, the large standard errors associated with the estimates of low-income uninsured children in SCHIP funding formulas were a concern. The best way to improve the precision of the estimates was to increase the sample size on which the estimates are based. In an attempt to help stabilize the funding formula, Congress approved a $10 million annual increase in funding to expand the size of the CPS. Specifically, the sample size expansion was aimed at producing more precise State-specific data on the SCHIP target population: the number of low-income children who do not have health insurance coverage (Balanced Budget Refinement Act of 1999).

The general strategy to increase the CPS sample was aimed disproportionately at those States with the largest standard errors associated with the estimate of low-income uninsured children (i.e., typically smaller States with small initial sample sizes). In addition, there was a specific strategy to increase the representation of minorities and children in the sample. Specifically, households not in the CPS sample for March but in the sample for February or April, or retired from the CPS sample in November, were asked the CPS questions if:

Table 1 presents the sample size increases by showing the change in the sample size by State for 2001, comparing the old to the new "expanded sample." The sample sizes across all States increased from 49,596 to 78,000 households, a 57-percent increase in the number of interviewed households. As seen in Table 1, the sample size expansions varied by States from New Hampshire's 171-percent increase in sample size to New Mexico's 23-percent increase. Overall, the expansion will have a positive impact on the precision of estimates of health insurance coverage at the national and State levels.

Using the CPS To Estimate Coverage Rates (Uninsurance) at Local Levels

The key issue with using the CPS to estimate local-levels of coverage concerns the way in which the sample is drawn. In many cases, the CPS cannot be used to make precise sub-State estimates of insurance coverage due to small sample sizes and the fact that the CPS sample does not include all counties within the State. While it is representative of the State population, it is not representative of counties or even regions within States. Additionally, for most States, it is not possible to make precise estimates for specific population groups defined by age, race, or country of origin because of small sample sizes.

Even at the State level, the Census Bureau recommends using 3-year averages for reporting State health insurance coverage estimates to help with the precision of the estimate by reducing the standard error. Pooling 3 years of sample from a specific State reduces the amount of sampling error associated with a specific State coverage estimate by approximately 30 percent. The reduction in sampling error allows an analyst to make more precise estimates of health insurance coverage for a specific State.

However, the CPS can be used to estimate coverage levels for some larger communities across the United States. There are some counties in the CPS sample that are large enough to obtain health insurance coverage estimates. These counties are depicted in Figure 1. More specific information on these counties can be obtained from the State Health Access Data Assistance Center (SHADAC - www.shadac.org).

Behavioral Risk Factor Surveillance System

The BRFSS was established based on the perceptions that:

States conduct the rolling monthly telephone surveys based on a common sampling methodology and a list of core questions to allow comparisons across States. More than 150,000 interviews are routinely conducted annually across all States. An advantage of the BRFSS for State analysts is that the States conduct the surveys themselves, have control over questions included in the State-specific modules, and have access to the person-level survey data for ongoing State analysis. Some States have also pursued additional samples and developed a stratification that allows them to estimate prevalence for regions within their respective States.

Technical Issues with Using the BRFSS To Estimate Health Insurance Coverage

For State health coverage policy, the central drawback of the BRFSS is the survey's public health focus on working-aged adults and its limited focus on children and health insurance coverage. Although some States have added a child component, the focus of BRFSS has not been on children. In addition, the BRFSS' focus on health risks has limited the amount of information collected on health insurance coverage and associated factors. Concerns have also been raised about the potential for undersampling low-income households by using a telephone survey. This concern has led to recent criticism of the BRFSS for its lack of data on special populations, including racial/ethnic minority populations and its lack of the city- or county-specific data needed for State health policy initiatives (Figgs, 2000). In addition, the survey relies on telephone administration with no uniform mechanism to adjust for the bias associated with not contacting households without telephones. A more common concern relates to the lack of standardization in data collection methods and the difficulty in overseeing 50 separate State data collection processes and assessing the impact of these varying methods on population estimates.

Using the BRFSS To Estimate Coverage Rates (Uninsurance) at Local Levels

The BRFSS sample design was established to be State representative and to provide some State stratification methods to allow regional estimates. While there is generally not enough sample size to allow for county-level estimates, there is some ability to make estimates at regional levels. Several States have developed their own model-based methods to develop county-level estimates of variables of interest, but these methods are State-specific and not universally adopted. Given its focus on working adults and on health risks, coupled with sampling methodology issues, the BRFSS has limited applicability for local area coverage estimates at the present time. For some States that have added a child component and increased the health insurance coverage question module, there may be an opportunity for model-based estimates (discussed in a later section). Statisticians and analysts at the CDC are currently working on using the BRFSS to develop model-based estimates of local uninsurance and other rates. However, it is unclear when those estimates will be available. Appendix A provides a link to the CDC's Web site so that the current status of this activity can be assessed.

State and Local Area Integrated Telephone Survey

The SLAITS allows researchers to collect data using customized questionnaires and the National Immunization Survey sampling frame of nearly 1 million households (National Center for Health Statistics, 2001a). The funding for SLAITS does not come from ongoing core National Center for Health Statistics (NCHS) Federal funds; rather it relies on other Federal agencies or outside sources for funding. The timing of the survey and opportunities for States to buy-in sample at the sub-State level are dependent on these funding sources, and sponsors may implement existing SLAITS survey modules or fund the development of new SLAITS modules at any time (National Center for Health Statistics, 2001b). There are presently four existing SLAITS survey modules including:

Technical Issues with Using SLAITS to Estimate Health Insurance Coverage

The major drawback of SLAITS is that its timing and funding are variable. Nonetheless, SLAITS has the flexibility to accommodate State-specific needs and has tremendous potential as a mechanism for State- and local-level comparisons. The large number of households screened in every State for the National Immunization Survey, which serves as the sampling frame for SLAITS, make the SLAITS survey particularly attractive. Moreover, selected modules include detailed information on the history of Medicaid and SCHIP coverage, awareness and attitudes toward Medicaid and SCHIP, and barriers to needed care.

One downside of SLAITS is that it is a telephone survey, which increases efficiencies but also introduces potential bias for non-telephone coverage. However, statistical adjustments are made to account for this non-coverage before estimates are released.

Using the SLAITS To Estimate Coverage Rates (Uninsurance) at Local Levels

SLAITS has the potential to be used as "...a broad-based ongoing surveillance system at the State and local levels to track and monitor..." health insurance coverage for both children and adults. (Ezzati-Rice et al., 2000) The survey mechanisms have the ability to increase sample size and over sample populations of interest. Inherent in SLAITS is the flexibility for States to sponsor additional data collection at the sub-State level (Madans et al., 2001). The SLAITS questionnaire items also mirror those used by the National Health Interview Survey to provide a national comparison benchmark for State and local estimates.

Medical Expenditure Panel Survey-Insurance Component

The MEPS-IC is a large State and nationally representative employer survey conducted by the Census Bureau for AHRQ. The survey collects detailed information from public and private employers on health insurance offer and take-up rates, costs of premiums (including employer contributions), and general employer characteristics. The information on coverage is presented from the employer perspective and does not include information on the distribution of coverage for a State's population nor an estimate of its uninsurance rate. AHRQ publishes the direct State estimates in tabular form annually. Researchers who want to conduct their own analyses of the MEPS-IC data must adhere to Census data access guidelines, which include the following requirements:

There are eight RDCs across the country: Washington, DC; Boston, MA; Pittsburgh, PA; Los Angeles, CA; Berkeley, CA; Durham, NC; Chicago, IL; and Ann Arbor, MI. There are strict confidentiality rules regarding what can be released from the data center. AHRQ will conduct special analyses for State-specific requests, but the resulting statistics must be aggregated to preclude any confidentiality issues posed by a small number of observations.

Using MEPS-IC To Estimate Coverage Rates (Uninsurance) at Local Levels

Again, MEPS-IC sampling frame does not allow analysts to examine subpopulations or geographic areas. While there are strict data confidentiality rules that prevent the release of data with county or even regional identifiers, MEPS-IC contains aggregate information that local analysts could use, including information on insurance offer and take-up rates by firm size, average wage, and type of industry. Analysts could use this information to develop a simple model-based approach to estimating employer coverage rates for a specific geographic area by using local information on employer characteristics. For example, one could examine the relationship between the likelihood of offering coverage within the MEPS-IC dataset at the State-wide level and employer-specific variables, such as wage or firm size. This ratio could theoretically then be used to estimate coverage rates as either a function of wage or firm size. (For further explanation, see the discussion of model-based estimation below.) Go to Appendix A for a list of Web links to the MEPS-IC State table files.

Return to Contents

Privately Sponsored National Surveys of Note

Two additional national surveys should be noted:

These privately funded surveys are focused on specific State and local areas. For analysts in these areas, these surveys provide access to important data. Unfortunately, the areas covered by these surveys are limited. For a complete list of State surveys and the State and local areas they cover, go to Table 2.

Community Tracking Survey

The CTS is sponsored by a grant from The Robert Wood Johnson Foundation to the Center for Studying Health System Change. This national representative survey collects data from 60 communities and conducts in-depth case studies in 12 localities. The CTS supports a household survey that addresses health insurance coverage rates as well as companion surveys of employers and physicians. The household survey is primarily a telephone survey with an in-person component to represent households without telephones. The sample includes over 60,000 individuals. The local and regional estimates of uninsurance provided by the CTS are important to States in the context of broader State policy as well as in identifying pockets of concern or the need for local intervention. Like many surveys, it underreports Medicaid participation (Lewis et al. 1998).

Using CTS To Estimate Coverage Rates (Uninsurance) at Local Levels

Uninsurance estimates can be made for the 12 metropolitan communities in the CTS sample. Table 2 shows the States with CTS communities. While the data collected in CTS communities may be compared with other communities in the sample, they cannot be compared with other State or other local estimates. CTS is designed to be representative of the Nation as a whole and of several communities throughout the country. State estimates are neither possible nor valid since the communities surveyed within a specific State are not representative of that State.

National Survey of America's Families

The NSAF has a nationally representative sample with additional representation for 13 select States (Table 2 lists NSAF States). In addition to focusing on the well-being of children and families, the NSAF provides comprehensive information on health insurance coverage. The data are timely and the Urban Institute makes the micro-data available to States for additional analyses. NSAF includes an area-probability sample to conduct surveys in households without telephones. In-person interviews facilitate collection in this component of the sample. In addition, the oversampling of low-income households, coupled with the effort to include households without telephones, has generated enough sample to provide estimates of some racially/ethnically diverse populations as well as immigrants. In 1999, there were more than 109,000 individuals in the sample. Also, the Urban Institute recently released a software interface tool to provide access for States interested in additional analyses of the data.

Using NSAF To Estimate Coverage Rates (Uninsurance) at Local Levels

There have been discussions between the Urban Institute and several States to fund additional sample for the NSAF in place of conducting separate State surveys. While this would allow examination of greater geographic detail, to date no States have signed on to a buy-in proposal. There is currently limited information available at the local level and limited sample size per State to allow for direct estimation of health insurance coverage.

Return to Contents

State Survey Efforts

Despite the improvements to national surveys such as the CPS, States continue to conduct surveys. The State Planning Grant (SPG) program of the Health Resources and Services Administration (HRSA) has stimulated some of this activity, but some States have had ongoing household surveys of health care coverage for many years. States pursue their own surveys because State-initiated surveys typically yield many more cases (sample) than Federal surveys provide, which, in turn, provides better estimates of health insurance coverage both Statewide and for subpopulations of interest (e.g., geographic areas and race/ethnicity). Finally, State surveys allow hands-on work with the data that fosters State-specific policy development and the potential to use State-level data to simulate the effect of different policy options on coverage and costs.

Many States that have conducted their own household health insurance surveys have designed their sampling approach so that regional estimates can be obtained. (More detail on each survey such as sample size, dates, and cost is available in Blewett et al., 2002). A few States have designed their surveys so that county-level estimates can be produced. An overview of the State surveys that yield regional and county estimates through direct estimation is presented in Table 2. More detail on the activities of these States with regard to sub-State area estimation is provided in Table 3.

Again, local analysts will want to assess the information collected as part of their State survey initiative. Many States include enough sample to provide direct estimates of coverage at the county and regional level. In addition, several States are developing model-based estimates using data from their State survey. Some examples of specific State activity aimed at providing direct estimates of coverage at sub-State levels include the following:

In the current round of SPG States, Georgia will be able to generate estimates for some of its larger counties and Atlanta. Alabama is using a sample design that focuses on geographic regions representing contiguous counties that will make it possible to generate direct estimates for some of the larger counties. West Virginia will be able to provide estimates at the county level according to its SPG application. It is unclear, however, whether these estimates will be generated synthetically through modeling or directly in the sampling design.

Overall, we have documented 36 States that have recently conducted household surveys and 20 States that are conducting employer surveys. Figure 2 provides a graphic illustration of the States and their survey efforts.

Return to Contents
Proceed to Next Section