Skip to main content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

State Estimates of Uninsured Children, January 1998. Final Report.

Publication Date
May 16, 2000

By:
Allen L. Schirm
John L. Czajka

May 17, 2000

Submitted to:
Department of Health and Human Services
Assistant Secretary for Planning and Evaluation HHH Building,
Room 442E 200 Independence Ave.,
SW Washington, DC 20202

Project Officer:
Robert Stewart
Sarah Goodell

Submitted by:
Mathematica Policy Research, Inc.
600 Maryland Ave., SW
Suite 550
Washington, DC 20024-2512
(202) 484-9220

Project Director:
John L. Czajka

"

Acknowledgments

The authors would like to thank several individuals who contributed to this work. We are especially grateful to Angela Schmitt for carrying out all of the extensive programming required to produce the database and the simulation model from which the state estimates reported herein were derived. We want to thank Alan Zaslavsky and Cara Olsen for their technical contributions and their assistance with numerous modeling issues that arose in the course of our work. We also want to recognize Mark Brinkley for producing the state tables and Margo Rosenbach for providing helpful comments on an earlier draft of the report.

Executive Summary

State estimates of the number and characteristics of uninsured children have attracted growing interest with the passage of legislation establishing the State Children’s Health Insurance Program (SCHIP).The Current Population Survey (CPS), the most widely cited source of data on the uninsured, has been criticized because its state samples are currently inadequate to provide state estimates of uninsured children with sufficient statistical precision to serve policy needs. To produce detailed breakdowns of uninsured children goes well beyond what most of the state samples can support. But while it may be inappropriate to rely on “direct sample” estimation to produce the tabulations that policymakers require, there exist innovative but well-grounded techniques for developing state estimates by the application of statistical procedures that “borrow strength” from other data sources.

This report employs statistical procedures that allow us to (1) make use of the entire national CPS sample in developing estimates for each state and (2) incorporate data from a wide variety of other sources. We apply these procedures to the March 1998 CPS to develop state estimates of the number of uninsured children in January 1998 by poverty level and age. The report also includes illustrative estimates of the number of children simulated to be eligible for Medicaid and the number who would be eligible for coverage under the SCHIP if the September 1999 rules had been in effect in 1997. The resulting tabulations are much more precise than those obtained from the CPS alone, and they can provide state and federal policymakers with baseline information that will be valuable in designing future expansions of SCHIP and in evaluating their progress in reducing the number of children who are without health insurance.

By simulating key features of the Medicaid and SCHIP eligibility provisions by state, we obtained estimates of the number of uninsured children in each state who were eligible for Medicaid and the additional number of uninsured children who would be made eligible for coverage under SCHIP. Nationally, we estimated that out of 11.5 million uninsured children, 4.2 million were eligible for Medicaid and an additional 2.4 million would become eligible for coverage under SCHIP, based on program provisions implemented by September 1999. Thus one-third of the uninsured children who were not already covered by Medicaid would become eligible for coverage under SCHIP. This proportion varied substantially across the states, depending in part on how many children under 200 percent of poverty were already covered under state Medicaid programs and how generously states elected to extend coverage. Six states appeared to be extending coverage to less than 10 percent of the remaining uninsured children in their states while 15 states were extending coverage to more than 40 percent of the remaining uninsured children.

Medicaid coverage is underreported in the CPS, but there are no widely accepted estimates of how many children whose Medicaid coverage is not reported are counted as uninsured.Our estimates include no adjustment for the underreporting of Medicaid, so they almost certainly overstate how many uninsured children are eligible for Medicaid and not enrolled. The estimates of additional uninsured children who would be covered by SCHIP are not affected by this undercount, but children who were actually enrolled in SCHIP in 1997 are excluded from the count of potential new enrollees to the extent that their coverage was reported in the CPS.

The report includes 10 tabulations for each of the 50 states and the District of Columbia. Each tabulation classifies children by six poverty levels and four age groups. Tables are presented for: all children under 19, the number and percent of children uninsured the previous year, total children and uninsured children simulated to be Medicaid-eligible, and total children and uninsured children simulated to be Medicaid- or SCHIP-eligible. Direct sample estimates of all children under 19 and the number and percent of children uninsured the previous year are included among the 10 tabulations to allow comparison with the model-based estimates.

I. Overview

A. Introduction

State estimates of not only the number of uninsured children but their characteristics have attracted growing interest with the passage of legislation establishing the State Children's Health Insurance Program (SCHIP). The Current Population Survey (CPS), the most widely cited source of data on the uninsured, has been criticized because its state samples are inadequate to provide state estimates of uninsured children with sufficient statistical precision to serve policy needs. To produce detailed breakdowns of uninsured children goes well beyond what most of the state samples can support. But while it may be inappropriate to rely on "direct sample" estimation to produce the tabulations that policymakers require, there exist innovative but well-grounded techniques for developing state and substate estimates by the application of statistical procedures that "borrow strength" from other data sources. This report employs statistical procedures that allow us to (1) make use of the entire CPS sample in developing estimates for each state and (2) incorporate data from a wide variety of other sources. We apply these procedures to the March 1998 CPS to develop state estimates of the number of uninsured children in January 1998 by poverty level and age. The report also includes illustrative estimates of the number of children simulated to be eligible for Medicaid and the number who would be eligible for coverage under the SCHIP if the September 1999 rules had been in effect in 1997.

Part I of this report is organized as follows. Section B discusses the purpose of the tables in this report, and Section C provides a brief description of the methodology. Section D presents estimates of uninsured rates among children for the 50 states and the District of Columbia (DC), and Section E presents illustrative estimates of the number of children who are eligible for Medicaid and SCHIP, by state. These estimates are based on a simulation model that incorporates state program detail. Finally, Section F discusses some caveats regarding these estimates.

Part II of the report consists primarily of the state tables--10 per state. Three of the tables are based entirely on the March 1998 CPS data without any of the enhancements that we have introduced. They are included to demonstrate, first, how limiting they are for most states and, second, to show how closely (or not) our enhanced estimates match in broad terms certain state characteristics that can be estimated directly from the CPS sample data. A Technical Appendix follows the state tables and outlines the procedures that we employed to develop the estimates.

B. Purpose of the Tables in This Report

While much attention has focused on the inadequacy of the state estimates of uninsured children that the CPS provides, state and federal needs go well beyond a simple count of the number of uninsured children (or adults) in each state. How many of the uninsured children fall into different poverty levels? What are their ages? The answers to these questions are important because they have a direct bearing on how many children are eligible for existing public health insurance programs or may be eligible for new programs. To states that are trying to incrementally expand their coverage of children, it is very important to know how many children may become eligible for coverage with a given incremental change in the eligibility criteria. States need to know what it will cost, potentially, to extend coverage beyond current levels. Uncertainty about these numbers and therefore the cost implications of expanding coverage usually translates into caution in the design of new programs or modifications to existing programs. With better information, many states may find that they can afford to be more generous in the coverage that they extend to the uninsured.

The tables in this report are intended to provide baseline information that states could find useful in planning future expansions of their SCHIP initiatives and in evaluating their progress in reducing the number of children who are without health insurance.1 Moreover, the database that was used to create these tables can be used to generate customized tables that better suit the needs of individual states or federal policymakers.

C. Methodology

The estimates presented in this report were derived from the March 1998 CPS public use file after making a number of enhancements discussed below and in the Technical Appendix. The CPS is a monthly sample survey of 50,000 households in the United States. It serves as the source of the official monthly unemployment estimates for the nation and the 50 states and the District of Columbia. Interviews are conducted with all civilian non institutionalized persons age 15 and older in the sample households. Each March, the Census Bureau includes a supplemental questionnaire in the CPS interview. The data collected in this questionnaire are the source of the official estimates of the incidence of poverty in the United States. In addition, the March supplement collects information on the health insurance coverage of all household members (including children). These data have become the source of the most widely-cited estimates of the number of uninsured children in the United States each year.

While the CPS sample was designed to support state-level estimates of unemployment rates and other labor force statistics, the sample sizes for most states are inadequate for satisfactorily precise monthly estimates that use only the survey data for that month.2 Thus, even if we are producing just a single estimate for each state--such as the number of uninsured children--that estimate will generally be imprecise due to the high sampling error associated with small samples. This means that we will be very uncertain about the true number of uninsured children and able to say for a typical state only that the number falls within a wide range, a range that is too wide to provide useful guidance for developing policy or administering a program.

This problem is even more severe when we must produce many estimates for each state, such as a table showing the distribution of uninsured children across poverty and age categories. Then, the already small sample for a state must be spread across the many cells of the table. It is not unusual for the available sample to have no observations in some cells even though the state population obviously has uninsured children with the characteristics defined by those cells.

When large samples are not available, a standard approach to improving the precision of sample estimates is to borrow strength.3 This entails the development of statistical models that allow us to derive, say, a 1998 estimate of uninsured children in Virginia using not only the 1998 data for Virginia from the main sample survey database but also data from other states, earlier years, and auxiliary sources, such as administrative records. Knowing something about the numbers of uninsured children in Virginia in 1996 and 1997 and the numbers of uninsured children in other states with economic and demographic conditions similar to Virginia's, we would generally be able to reduce substantially our uncertainty about how many uninsured children lived in Virginia in 1998.

To derive most of the estimates presented in this report, we have used a method for borrowing strength that is described in detail by Schirm and Zaslavsky (1997). With this method, we have reweighted the March 1998 CPS database to produce 51 sets of weights--one set for each state and the District of Columbia. We use the Virginia weights to derive estimates for Virginia. Each of the roughly 50,000 households in the CPS database--regardless of state of residence--gets a Virginia weight, greatly increasing the size of the sample from which to obtain estimates for Virginia. How much Virginia weight a household gets and, therefore, its relative contribution to estimates for Virginia depends on the household's characteristics. If a household from, say, Montana has characteristics that would make it unusual were it in Virginia, as opposed to other states, it receives a relatively small--probably negligible--Virginia weight. If, instead, a household of that type would be more common in Virginia than in other states, it receives a relatively large Virginia weight.

The Virginia weight assigned to a household with a particular set of characteristics depends on the aggregate characteristics of Virginia. Specifically, Virginia weights are controlled so that totals derived for Virginia using Virginia weights equal specified values. For example, the weights might be controlled so as to reproduce specified totals of children, children below 100 percent of poverty, and uninsured children. By controlling the weights, we ensure that the entire database--when weighted according to the Virginia weights--looks like Virginia in terms of totals that are relevant to ascertaining the patterns of insurance coverage among children.4

In reweighting the March 1998 CPS database, we used controls reflecting the age and racial/ethnic structure of each state's child population, the distribution of children across poverty categories, and the numbers of uninsured children (by poverty category). The full list of totals to which we controlled weights appears in the Technical Appendix.

Reweighting a database as we have described can substantially improve the precision of estimates because the samples used to derive the estimates are much larger than when we use only the observations from a single state. Using observations from all the states allows us to borrow strength. Although this alone improves precision, we have further improved precision by using administrative estimates or empirical Bayes shrinkage estimates--rather than direct sample estimates --for many of the control totals used in the reweighting.5 The administrative totals, which are population estimates derived from mainly vital records (and decennial census) data, have essentially no sampling error, and the shrinkage totals, which are derived by borrowing strength, are more precise than direct sample estimates. The specific sources of control totals are described in greater detail in the Technical Appendix.6

D. Estimates of the Uninsured

Table I.1 reports the March 1998 CPS sample sizes for all children and uninsured children by state and compares the model-based estimates of the uninsured rate with the direct sample estimates. The sample sizes underscore why it is necessary to employ procedures of the kind used here to construct state-level tabulations of uninsured children. Most state samples include fewer than 100 uninsured children, and 15 states have fewer than 50. Spreading such small numbers of observations over a table with 24 cells (the number of cells in each of the state tables in Part II) cannot yield much information of value.

TABLE I.1
COMPARISON OF DIRECT SAMPLE AND MODEL-BASED ESTIMATES OF THE PERCENTAGE OF CHILDREN UNDER 19 WHO WERE UNINSURED IN 1997

State CPS Sample Sizes Direct Sample
Percent Uninsured
Model-Based
Percent Uninsured
Difference:
Model-Based
Minus Direct
All Children Uninsured Children
U.S. Total 38,461 6,101 15.2 15.2 0.0
Alabama 430 67 15.4 16.4 1.0
Alaska 490 63 11.3 10.9 -0.4
Arizona 761 211 26.2 21.8 -4.4
Arkansas 540 147 28.0 22.8 -5.2
California 4,277 931 18.8 19.4 0.6
Colorado 533 80 14.1 14.0 -0.1
Connecticut 382 48 11.8 9.7 -2.1
Delaware 377 60 14.3 11.2 -3.1
District of Columbia 260 39 14.3 15.6 1.3
Florida 1,509 300 20.3 20.2 -0.1
Georgia 645 103 16.8 14.6 -2.2
Hawaii 328 16 5.5 8.9 3.4
Idaho 719 142 18.2 18.1 -0.1
Illinois 1,653 215 11.1 11.7 0.6
Indiana 498 64 12.7 11.5 -1.2
Iowa 474 50 11.2 10.1 -1.1
Kansas 476 51 10.2 11.3 1.1
Kentucky 446 61 14.1 14.0 -0.1
Louisiana 472 107 22.9 21.3 -1.6
Maine 318 47 14.9 11.6 -3.3
Maryland 376 39 10.4 11.6 1.2
Massachusetts 753 72 9.2 9.1 -0.1
Michigan 1,244 106 8.6 9.3 0.7
Minnesota 573 43 7.4 6.9 -0.5
Mississippi 465 86 19.1 19.9 0.8
Missouri 417 54 13.1 12.4 -0.7
Montana 552 82 13.8 16.0 2.2
Nebraska 492 49 9.7 10.2 0.5
Nevada 478 101 19.4 17.6 -1.8
New Hampshire 368 37 10.3 7.6 -2.7
New Jersey 1,078 180 15.9 14.9 -1.0
New Mexico 780 140 17.1 18.6 1.5
New York 2,449 418 15.8 15.4 -0.4
North Carolina 858 158 18.3 15.2 -3.1
North Dakota 468 64 13.6 11.5 -2.1
Ohio 1,356 153 10.6 11.3 0.7
Oklahoma 570 89 15.5 18.6 3.1
Oregon 452 54 11.2 14.4 3.2
Pennsylvania 1,370 120 8.5 9.4 0.9
Rhode Island 296 24 8.1 7.3 -0.8
South Carolina 415 76 18.8 16.2 -2.6
South Dakota 488 33 6.5 9.1 2.6
Tennessee 459 49 10.9 9.7 -1.2
Texas 2,603 700 25.2 24.7 -0.5
Utah 710 100 12.7 13.2 0.5
Vermont 350 23 6.6 5.9 -0.7
Virginia 517 62 11.9 12.9 1.0
Washington 507 40 7.9 9.5 1.6
West Virginia 369 45 12.6 11.8 -0.8
Wisconsin 487 24 4.9 7.1 2.2
Wyoming 573 78 13.7 14.1 0.4
SOURCE: Mathematica Policy Research, from the March 1998 CPS and other sources.

Differences between the model-based and direct sample estimates vary from 0.1 to 5.2 percentage points (plus or minus). The two estimates are within one percentage point for 25 states but differ by three percentage points or more for 8 states. Sample size is clearly relevant but not the sole factor affecting the size of the difference. For example, all nine states with CPS sample sizes of more than 1,000 children have differences of 1 percentage point or less, but the two largest differences occur in states with above average sample sizes (Arizona and Arkansas). Perhaps more importantly, the model-based estimates show the impact of shrinkage toward the mean. Generally the most extreme rates--which are probably too extreme--are pulled toward the center. We see this in the states of Arizona, Arkansas, Louisiana, and Texas, where the model-based estimates are lower than the (high) direct sample estimates, and in Hawaii, South Dakota, Washington, and Wisconsin, where the model-based estimates are greater than the (low) direct sample estimates. This result is not universal, however. There are states with low direct sample uninsured rates (for example, Vermont and Minnesota) that get assigned even lower rates by the model-based procedure. As low as the direct sample estimates were in these states, the regression model predicted even lower rates. Thus the model did not indiscriminantly eliminate high and low rates.

Table I.2 reports uninsured rates by poverty level for the model-based estimates. It is quite clear from an examination of these rates that the model-based procedure does not generate homogenous uninsured rates across the states. Rather, there are distinctly different patterns in the rates for groups of states.

TABLE I.2
UNINSURED RATES BY POVERTY LEVEL: MODEL-BASED ESTIMATES

State Federal Poverty Level Based on 1997 Annual Family Income
Under 50% 50% to < 100% 100% to < 150% 150% to < 200% 200% to < 350% 350% or More
U.S.Total 25.9 24.5 27.5 21.0 11.5 5.9
Alabama 27.4 21.9 33.2 25.4 9.7 5.4
Alaska 31.2 20.0 22.9 18.2 8.7 4.9
Arizona 39.7 37.0 33.6 26.1 14.6 7.2
Arkansas 39.9 31.3 34.7 26.2 13.3 7.3
California 23.4 25.9 35.9 27.4 16.1 7.7
Colorado 28.4 25.4 26.2 20.3 12.1 6.2
Connecticut 21.2 19.0 17.2 13.8 10.3 5.1
Delaware 22.3 17.0 25.0 18.5 8.6 4.9
District of Columbia 13.3 15.5 36.1 33.0 11.5 5.1
Florida 37.3 34.8 31.2 24.4 12.7 6.6
Georgia 26.6 22.2 29.6 21.4 8.6 4.9
Hawaii 8.6 6.2 13.9 11.4 10.9 5.8
Idaho 45.3 35.1 27.8 21.6 11.1 6.2
Illinois 22.4 20.7 26.7 21.2 6.5 3.4
Indiana 32.7 23.1 20.3 16.5 8.0 4.7
Iowa 16.2 9.6 18.4 15.2 9.3 5.2
Kansas 25.6 18.3 21.7 17.4 8.8 4.8
Kentucky 25.5 16.9 22.7 18.3 11.0 6.1
Louisiana 36.1 32.0 31.8 24.4 14.3 7.5
Maine 23.8 14.8 17.4 14.2 10.5 5.7
Maryland 30.1 25.9 26.8 21.3 8.0 4.1
Massachusetts 15.1 12.2 19.0 15.3 9.5 4.9
Michigan 18.5 13.9 18.9 15.1 7.2 3.9
Minnesota 8.4 4.8 13.9 11.6 7.5 4.2
Mississippi 29.4 26.1 35.4 23.0 11.6 6.7
Missouri 23.6 16.4 21.2 16.8 10.3 5.6
Montana 32.2 20.7 25.6 20.1 11.8 6.5
Nebraska 13.9 8.3 19.4 15.7 10.1 5.5
Nevada 38.9 36.8 33.0 25.0 11.4 6.1
New Hampshire 17.8 10.7 9.7 8.1 9.0 4.9
New Jersey 27.6 26.1 31.5 24.4 13.7 7.1
New Mexico 24.7 27.5 23.1 16.8 17.1 8.2
New York 19.1 18.6 33.1 26.4 12.7 6.2
North Carolina 30.6 24.0 27.0 21.3 10.6 5.6
North Dakota 20.4 11.7 19.7 16.1 10.6 5.9
Ohio 25.3 18.2 18.4 14.8 9.3 5.1
Oklahoma 38.7 28.9 26.9 21.0 13.0 7.1
Oregon 38.2 28.9 25.0 19.8 8.2 4.5
Pennsylvania 21.0 15.2 11.9 9.7 9.3 5.0
Rhode Island 14.5 11.9 12.4 10.0 6.7 3.4
South Carolina 32.0 26.3 29.6 19.3 10.6 6.5
South Dakota 13.8 8.0 16.7 13.8 8.1 4.7
Tennessee 10.3 7.1 16.3 13.1 10.0 5.6
Texas 35.2 37.4 38.1 28.8 19.2 9.2
Utah 35.8 25.0 21.4 17.3 10.0 5.8
Vermont 4.1 2.2 6.7 5.7 8.4 4.7
Virginia 26.6 21.4 23.5 18.8 10.8 5.6
Washington 24.5 18.5 14.5 11.8 8.0 4.3
West Virginia 19.0 12.4 16.1 13.4 10.4 5.8
Wisconsin 6.9 3.9 16.7 13.5 7.3 4.1
Wyoming 32.6 23.0 20.7 16.3 11.2 6.1
SOURCE: Mathematica Policy Research, from the March 1998 CPS and other sources.

For example, there is a general tendency for children who are between 100 percent and 150 percent of poverty to have the highest uninsured rates. Children below 100 percent of poverty often have access to Medicaid while children above 150 percent of poverty are more likely to have employer-sponsored or other private insurance. But despite these tendencies we do see a number of states in which the uninsured rates are highest among children under 50 percent of poverty and then decline with each succeeding higher income level. Arizona, Florida, Idaho, Louisiana, Nevada, Oklahoma, Oregon, Utah, and Wyoming are among the states that fit this pattern. It is likely that these states have low participation in Medicaid since most children under 50 percent of poverty will be covered by Medicaid. In general, the states that fit this pattern have large Hispanic populations or western locations. The high uninsured rates of Hispanic children are well-documented. For the western states, the high uninsured rates at low income levels may reflect low participation in safety net programs generally. Whatever the reason, there is a clear pattern that the model-based estimates are able to identify.

Four states--Hawaii, Minnesota, Vermont, and Wisconsin--have single-digit uninsured rates in the two lowest poverty classes. The first three of these are noted for their broad Medicaid coverage expansions, and we would guess that Medicaid participation is very high among the eligible populations. That the model-based estimates can differentiate between these states and the rest provides additional face validity.

E. Illustrative Estimates of Uninsured Children Eligible for Medicaid and SCHIP

Development of the model-based estimation procedures employed here was motivated by an interest in applying the methodology of microsimulation to individual states. Microsimulation is particularly useful for estimating the incremental impact of small changes in program eligibility on caseloads and costs, but it requires a very large sample.7 To illustrate the application of microsimulation to the reweighted March 1998 database, we have prepared a simulation of eligibility under both Medicaid (1997 rules) and SCHIP (September 1999 rules) and applied this simulation model to the reweighted data. The simulation program captures most of the major elements of state differentiation in income eligibility limits by age, the use of gross versus net income, and, to some degree, the application of asset tests.8 We base eligibility on annual family income rather than trying to construct monthly income streams that would allow a more literal replication of the Medicaid eligibility determination. This is a widely-used practice--in large part because the CPS and other major surveys collect only annual income data. Furthermore, given that the CPS provides only annual rather than monthly estimates of insurance coverage, basing eligibility on simulated monthly rather than reported annual income would not solve the problem of relating eligibility to insurance coverage.

The first four columns of Table I.3 present state estimates of the number of uninsured children, the number of all children who were simulated to be eligible for Medicaid (without regard to insurance coverage), and both the number and percentage of these Medicaid-eligible children who were reported as uninsured. The final three columns present estimates of children who were simulated to be eligible for either Medicaid or SCHIP (again without regard to insurance coverage) and the number and percentage of these Medicaid/SCHIP-eligible children who were uninsured. Medicaid eligibility is based on program rules that were in effect in 1997 while SCHIP eligibility is based on state program provisions that were in effect in September 1999. Thus SCHIP eligibility is prospective or hypothetical rather than actual eligibility in 1997. It should be noted as well that even in the absence of SCHIP, Medicaid eligibility would have grown between 1997 and 1999. Earlier reforms extended eligibility to low income children who were born after September 30, 1983. As these children age, a larger and larger share of all children are made eligible by these provisions.

State Number
Uninsured
All
Simulated
Medicaid
Eligible
Simulated
Medicaid
Eligible But
Uninsured
Percent
Medicaid
Eligible But
Uninsured
All
Simulated
Medicaid
or SCHIP
Eligible
Simulated
Medicaid
or SCHIP
Eligible But
Uninsured
Percent
Medicaid
or SCHIP
Eligible But
Uninsured

TABLE I.3
STATE ESTIMATES OF UNINSURED, MEDICAID-ELIGIBLE, AND SCHIP-ELIGIBLE CHILDREN

U.S. Total 11,452,600 19,329,700 4,249,500 22.0 27,953,400 6,599,900 23.6
Alabama 195,300 292,900 64,800 22.1 532,600 141,000 26.5
Alaska 22,500 38,500 7,900 20.5 82,400 15,300 18.6
Arizona 300,500 377,500 123,000 32.6 633,200 208,800 33.0
Arkansas 164,800 197,300 58,400 29.6 213,600 65,100 30.5
California 1,893,100 3,077,600 732,200 23.8 4,649,300 1,261,200 27.1
Colorado 156,900 193,800 43,000 22.2 314,300 77,500 24.7
Connecticut 81,700 193,200 32,100 16.6 359,900 53,600 14.9
Delaware 21,900 40,100 7,600 19.0 64,300 13,000 20.2
District of Columbia 18,500 47,000 7,600 16.2 65,000 13,800 21.2
Florida 772,300 1,053,600 319,700 30.3 1,762,100 552,600 31.4
Georgia 320,400 614,700 144,100 23.4 946,300 230,800 24.4
Hawaii 29,200 95,800 8,300 8.7 103,600 9,300 9.0
Idaho 70,000 79,800 24,200 30.3 117,600 38,400 32.7
Illinois 401,500 778,500 151,900 19.5 943,800 206,500 21.9
Indiana 186,200 305,100 69,300 22.7 377,100 88,600 23.5
Iowa 78,100 143,000 16,400 11.5 251,400 36,700 14.6
Kansas 83,800 142,200 26,300 18.5 246,400 48,900 19.8
Kentucky 150,300 302,800 59,000 19.5 468,000 95,600 20.4
Louisiana 282,700 380,300 108,100 28.4 492,600 159,000 32.3
Maine 36,600 77,700 13,300 17.1 110,900 18,700 16.9
Maryland 162,400 252,300 60,800 24.1 411,300 104,800 25.5
Massachusetts 140,600 304,000 41,700 13.7 439,700 65,500 14.9
Michigan 252,000 730,100 115,400 15.8 946,400 153,000 16.2
Minnesota 92,100 222,400 15,600 7.0 247,800 17,800 7.2
Mississippi 166,200 262,900 64,300 24.5 292,200 77,900 26.7
Missouri 186,700 367,000 66,500 18.1 884,300 146,100 16.5
Montana 40,300 56,800 12,400 21.8 81,300 19,900 24.5
Nebraska 48,900 86,400 9,000 10.4 151,600 21,500 14.2
Nevada 86,000 100,700 31,700 31.5 175,000 56,800 32.5
New Hampshire 23,900 80,200 8,200 10.2 143,200 14,800 10.3
New Jersey 317,100 403,100 97,800 24.3 643,400 172,000 26.7
New Mexico 102,500 287,900 64,300 22.3 328,000 73,300 22.3
New York 752,700 1,416,200 267,700 18.9 2,060,400 470,800 22.8
North Carolina 314,800 531,700 130,100 24.5 859,200 213,300 24.8
North Dakota 20,500 36,600 5,100 13.9 38,200 5,500 14.4
Ohio 347,300 608,500 107,000 17.6 829,400 161,200 19.4
Oklahoma 179,400 234,100 64,200 27.4 349,800 94,900 27.1
Oregon 128,800 211,600 59,900 28.3 273,200 77,400 28.3
Pennsylvania 288,900 619,900 86,500 14.0 928,200 125,300 13.5
Rhode Island 18,200 106,300 11,300 10.6 124,300 13,000 10.5
South Carolina 170,900 253,600 62,700 24.7 340,000 94,100 27.7
South Dakota 19,700 50,500 5,300 10.5 55,900 6,500 11.6
Tennessee 141,000 362,500 30,700 8.5 408,200 37,400 9.2
Texas 1,517,400 1,771,700 562,100 31.7 2,023,900 678,200 33.5
Utah 99,500 145,500 33,800 23.2 250,900 57,100 22.8
Vermont 9,100 55,400 3,300 6.0 85,400 5,700 6.7
Virginia 232,500 397,700 87,000 21.9 564,000 125,100 22.2
Washington 150,600 515,000 83,800 16.3 676,600 101,700 15.0
West Virginia 53,300 140,400 20,100 14.3 169,500 25,600 15.1
Wisconsin 102,600 259,900 18,300 7.0 401,100 40,700 10.1
Wyoming 20,400 27,400 5,700 20.8 36,600 8,600 23.5
SOURCE: Mathematica Policy Research, from the March 1998 CPS and other sources.

Across all of the states, our estimates of uninsured children who were eligible for Medicaid total 4.2 million out of the 11.5 million uninsured children under 19. Our simulation of SCHIP eligibility suggests that SCHIP would have extended eligibility for public insurance coverage to about 2.4 million additional children.

The estimated percentage of simulated Medicaid-eligible children who were uninsured in each state helps us to understand the patterns of uninsurance among low income children that we saw in Table I.2. Nationally, 22.0 percent of our simulated Medicaid-eligible children were uninsured. Among the states, this rate varies from a low of 7 percent (Minnesota and Wisconsin) to a high of 33 percent (Arizona). Generally, states with higher uninsured rates among children under 100 percent of poverty than among children between 100 and 150 percent of poverty have high rates of Medicaid-eligible uninsured in Table I.3. For example, Arizona, Florida, and Idaho have higher uninsured rates in the two lowest poverty classes than in the 100 to 150 percent class, and all three have Medicaid-eligible uninsured rates in excess of 30 percent. At the other end of the distribution, we singled out Hawaii, Minnesota, Vermont, and Wisconsin for their relatively low uninsured rates among children below 100 percent of poverty, and all four of these have Medicaid-eligible uninsured rates below 10 percent.

Some important implications of the impact of SCHIP eligibility at the state level can be seen in Table I.4, which shows by state the number of uninsured children who were not simulated to be Medicaid-eligible and both the number and percentage of these who would be made eligible for coverage under SCHIP. Nationally, SCHIP would extend eligibility to about one-third of the uninsured children who were not otherwise eligible for Medicaid in 1997. This varies substantially by state--in part because some states were already covering a large part of the population that other states would now cover under SCHIP. Hawaii and Minnesota, which provide broad coverage under Medicaid, would extend coverage to fewer than 5 percent of the uninsured who were not eligible for Medicaid, whereas Alabama, with comparatively low Medicaid coverage, would extend coverage through SCHIP to nearly 60 percent of its remaining uninsured children. At the same time, Texas with comparatively low Medicaid coverage would extend coverage through SCHIP to only 12 percent of its remaining uninsured children while DC would extend coverage to 57 percent.

TABLE I.4
UNINSURED CHILDREN TO WHOM HEALTH INSURANCE COVERAGE MAY BE EXTENDED BY SCHIP

State Uninsured Who are Not
Simulated Medicaid Eligible
Number of These Who are
Eligible for SCHIP
Percent Eligible for SCHIP
U.S. Total 7,203,100 2,350,400 32.6
Alabama 130,500 76,200 58.4
Alaska 14,600 7,400 50.7
Arizona 177,500 85,800 48.3
Arkansas 106,400 6,700 6.3
California 1,160,900 529,000 45.6
Colorado 113,900 34,500 30.3
Connecticut 49,600 21,500 43.3
Delaware 14,300 5,400 37.8
District of Columbia 10,900 6,200 56.9
Florida 452,600 232,900 51.5
Georgia 176,300 86,700 49.2
Hawaii 20,900 1,000 4.8
Idaho 45,800 14,200 31.0
Illinois 249,600 54,600 21.9
Indiana 116,900 19,300 16.5
Iowa 61,700 20,300 32.9
Kansas 57,500 22,600 39.3
Kentucky 91,300 36,600 40.1
Louisiana 174,600 50,900 29.2
Maine 23,300 5,400 23.2
Maryland 101,600 44,000 43.3
Massachusetts 98,900 23,800 24.1
Michigan 136,600 37,600 27.5
Minnesota 76,500 2,200 2.9
Mississippi 101,900 13,600 13.3
Missouri 120,200 79,600 66.2
Montana 27,900 7,500 26.9
Nebraska 39,900 12,500 31.3
Nevada 54,300 25,100 46.2
New Hampshire 15,700 6,600 42.0
New Jersey 219,300 74,200 33.8
New Mexico 38,200 9,000 23.6
New York 485,000 203,100 41.9
North Carolina 184,700 83,200 45.0
North Dakota 15,400 400 2.6
Ohio 240,300 54,200 22.6
Oklahoma 115,200 30,700 26.6
Oregon 68,900 17,500 25.4
Pennsylvania 202,400 38,800 19.2
Rhode Island 6,900 1,700 24.6
South Carolina 108,200 31,400 29.0
South Dakota 14,400 1,200 8.3
Tennessee 110,300 6,700 6.1
Texas 955,300 116,100 12.2
Utah 65,700 23,300 35.5
Vermont 5,800 2,400 41.4
Virginia 145,500 38,100 26.2
Washington 66,800 17,900 26.8
West Virginia 33,200 5,500 16.6
Wisconsin 84,300 22,400 26.6
Wyoming 14,700 2,900 19.7
SOURCE: Mathematica Policy Research, from the March 1998 CPS and other sources.

Table I.5 summarizes the coverage of children under 200 percent of poverty by state, breaking down the low income population into those with insurance, those who were uninsured but eligible for Medicaid in 1997, those who were uninsured and not Medicaid-eligible in 1997 but would be eligible for Medicaid or SCHIP by the rules that were in effect in September 1999, and those who would remain uninsured.9 In the final column we see that three states would leave more than 10 percent of their low income children uninsured and ineligible for Medicaid or SCHIP: Arkansas (16 percent), Mississippi (11 percent), and Texas (13 percent).10 Most states, however, would leave fewer than 2 percent of their low income children uninsured and ineligible for Medicaid or SCHIP.

TABLE I.5
STATE COVERAGE OF CHILDREN UNDER 200 PERCENT OF POVERTY

State Number of Children
Under 200% of Poverty
Percentage of Children Under 200% of Poverty Who Are:
Insured Uninsured
But Medicaid
Eligible
(1997)
Uninsured
But Future Medicaid
or SCHIP Eligible
(1999)
Residual
Uninsured
U.S. Total 31,218,900 75.4 13.5 7.2 3.8
Alabama 541,200 73.0 12.0 14.0 1.1
Alaska 60,200 78.6 13.0 8.1 0.3
Arizona 649,700 66.0 18.9 13.1 1.9
Arkansas 399,100 67.7 14.6 1.7 16.0
California 4,733,500 72.0 15.4 11.1 1.5
Colorado 374,900 75.4 11.5 9.2 4.0
Connecticut 221,500 82.3 14.0 2.9 0.7
Delaware 64,900 79.4 11.6 8.3 0.8
District of Columbia 65,100 78.3 11.7 9.5 0.5
Florida 1,792,800 67.9 17.8 12.9 1.5
Georgia 957,100 75.1 15.0 9.0 1.0
Hawaii 122,000 90.1 6.1 0.6 3.2
Idaho 163,500 69.5 14.8 8.7 7.0
Illinois 1,329,400 77.2 11.4 4.1 7.3
Indiana 512,400 77.7 13.5 3.8 5.0
Iowa 284,200 85.1 5.7 7.1 2.1
Kansas 251,400 79.8 10.5 8.8 0.9
Kentucky 475,900 79.3 12.3 7.6 0.8
Louisiana 670,800 68.7 16.1 7.6 7.6
Maine 125,900 83.2 10.5 4.3 2.1
Maryland 413,700 73.9 14.6 10.5 1.0
Massachusetts 446,400 84.7 9.3 5.2 0.8
Michigan 958,400 83.5 12.0 3.8 0.7
Minnesota 400,200 90.0 3.7 0.3 6.1
Mississippi 454,000 71.5 14.1 3.0 11.3
Missouri 602,200 80.8 11.0 7.8 0.5
Montana 113,900 76.1 10.9 6.6 6.4
Nebraska 172,100 85.4 5.2 7.3 2.1
Nevada 179,000 66.7 17.7 13.9 1.7
New Hampshire 80,300 89.5 9.6 0.6 0.2
New Jersey 651,700 72.6 14.9 11.2 1.3
New Mexico 295,400 76.9 21.3 0.8 1.0
New York 2,152,200 76.4 12.4 9.4 1.8
North Carolina 871,200 74.7 14.9 9.4 1.0
North Dakota 65,800 83.4 7.8 0.6 8.2
Ohio 1,138,300 81.3 9.4 4.8 4.6
Oklahoma 441,700 78.7 14.5 7.0 0.0
Oregon 353,800 72.9 16.9 4.9 5.3
Pennsylvania 1,096,900 86.0 7.8 3.5 2.7
Rhode Island 87,800 87.7 10.8 0.9 0.6
South Carolina 445,100 73.5 14.0 7.1 5.4
South Dakota 85,800 87.1 6.2 1.4 5.4
Tennessee 656,700 88.1 4.6 1.0 6.2
Texas 3,082,400 64.9 18.2 3.8 13.1
Utah 256,300 77.0 13.2 8.9 0.9
Vermont 54,300 95.2 4.1 0.9 0.0
Virginia 622,700 77.6 14.0 6.1 2.3
Washington 529,200 83.3 14.9 1.3 0.5
West Virginia 229,600 84.9 8.8 2.4 4.0
Wisconsin 428,300 89.2 4.0 5.2 1.6
Wyoming 58,000 78.1 9.8 5.0 7.1
SOURCE: Mathematica Policy Research, from the March 1998 CPS and other sources.

F. Caveats About These Estimates

We highlight three caveats that apply to the tables presented in Part II and one additional caveat that applies to further use of the reweighted CPS database. The first three are the CPS's undercount of the Medicaid population, the overstatement of uninsurance among infants, and the limitations of the Medicaid simulation. The fourth involves limitations on the kinds of data that can be tabulated with the reweighted database.

1. Medicaid Undercount

It is widely known that the Medicaid enrollment in the CPS understates estimates compiled from the states' program administrative statistics. It is much less widely known that this Medicaid undercount has been growing.11 It is very likely that at least some of the children reported as uninsured were actually covered by Medicaid.12 We have not attempted to adjust our estimates in any way for this Medicaid undercount, so a portion (and perhaps a large portion) of those children that we report as eligible for Medicaid but uninsured may have actually been covered by Medicaid.13

2. Uninsured Rates for Infants

Despite their greater access to Medicaid, infants are reported to have higher uninsured rates than children 1 to 5. This is peculiar to the CPS, however, and it is very likely due to a combination of two factors: (1) the uninsured being identified as those who report no insurance (as opposed to reporting that they were uninsured) and (2) insurance coverage being measured for the previous year (Czajka and Lewis 1999). Children born between the end of the reference year and the March survey date cannot in truth be described as having had coverage of any kind the previous year, and parents who answer the questions literally will end up with their newborn infants classified as uninsured. It is not possible to identify infants born after the end of the reference year, and so it is not possible to screen out those who may have been misclassified. Users of the data need to be aware that the rate of uninsurance among infants is overstated.

3. Medicaid Eligibility

The rules governing Medicaid eligibility, which vary by state, are extraordinarily complex. A complete simulation of all the ways that a child can become eligible for Medicaid is impossible--both because of the limitations of survey data and because the full details of eligibility, state by state, are not documented in any accessible form. For this reason, any Medicaid eligibility simulation is going to involve simplifications. It is quite rare, for example, that anyone simulates eligibility under the medically needy provisions, other than indirectly, and we have not done so here. Nor have we incorporated state-specific differences in the calculation and application of disregards. The information required to do so for all of the states is not readily available, and that limits not only our own simulation but those that could be constructed by others--even with substantially more resources.

4. Tabulating the Reweighted Database

The reweighting of the CPS database for state estimation was accomplished by applying a number of controls that, depending on which of the 51 sets of weights is chosen, make the database "look like" a specific state. The controls were chosen because of their relevance to estimating the number of uninsured children by age and poverty level. While it is possible to tabulate any field on the CPS file with the state weights that we have constructed, the state-specificity of the resulting tabulation deteriorates with the declining relevance of children's age, race and Hispanic origin, poverty level, and insurance coverage (insured or not) to the fields being tabulated. In addition, fields that depend on rules that differ across states, such as AFDC participation or Medicaid participation, may be inconsistent with the rules in most other states and, for this reason, should not be tabulated or used to infer eligibility in a simulation algorithm.14

II. State Estimates

The sections that follow present ten tabulations for each of the 50 states and the District of Columbia. The tables are organized by state. Each table consists of a cross-tabulation of a defined population of children by poverty level and age. The tables utilize six poverty levels:

  • Less than 50 percent of the federal poverty level (FPL)
  • 50 percent to less than 100 percent
  • 100 percent to less than 150 percent
  • 150 percent to less than 200 percent
  • 200 percent to less than 350 percent
  • 350 percent or greater

The four age categories are:

  • Less than 1 year of age (infant)
  • 1 to 5
  • 6 to 12
  • 13 to 18

These age categories match those that the Health Care Financing Administration is using in reports of SCHIP enrollment.

Each table describes a different population or sub population within the state. The first seven tables were produced from the reweighted CPS database. That is, they are model-based estimates. They are dated January 1998 rather than March 1998 because the population estimates that were used as external controls are effectively January 1998 numbers.

  • Table II.A: Model-based Estimates of the Number of Children by Age and Poverty Level, January 1998
  • Table II.B: Model-based Estimates of the Number of Children in January 1998 Who Were Uninsured in the Previous Year
  • Table II.C: Model-based Estimates of the Percent of Children in January 1998 Who Were Uninsured in the Previous Year
  • Table II.D: Model-based Estimates of Simulated Medicaid-Eligible Children, January 1998
  • Table II.E: Model-based Estimates of Simulated Medicaid-Eligible Children in January 1998 Who Were Uninsured in the Previous Year
  • Table II.F: Model-based Estimates of Simulated SCHIP- or Medicaid-eligible Children, January 1998
  • Table II.G: Model-based Estimates of Simulated SCHIP- or Medicaid-eligible Children in January 1998 Who Were Uninsured in the Previous Year

The final three tables are direct sample estimates from the March 1998 survey. They use no other data. They provide counterparts to the first three of the tables listed above and were included so that we could demonstrate on a state-by-state basis how much our borrowing strength methodology has improved our ability to produce usable state-level information. These final three tables are:

  • Table II.H: Direct Sample Estimates of the Number of Children by Age and Poverty Level, March 1998
  • Table II.I: Direct Sample Estimates of the Number of Children in March 1998 Who Were Uninsured in the Previous Year
  • Table II.J: Direct Sample Estimates of the Percent of Children in March 1998 Who Were Uninsured in the Previous Year

These tables are dated March 1998 because the March CPS sample is weighted to population estimates for the month of March.15

We walk through these tables for Alabama to illustrate their interpretation. Table II.A.1 indicates that there were about 1.2 million children in Alabama in January 1998 and provides a breakdown by poverty level and age.16 These estimates indicate, for example, that there were 122,000 children in families with incomes below 50 percent of poverty and that about 33,000 were 13 to 18 years old. Table II.B.1 shows the number of uninsured children, which totaled 195,000. Children in families between 100 and 150 percent of poverty accounted for the largest number of uninsured (47,100) in any poverty class and nearly three times as many as children in families above 350 percent of poverty. Table II.C.1 gives the percentage of children who were uninsured in the previous year. The estimates in this table were prepared by dividing each cell entry in Table II.B.1 by the corresponding cell entry in Table II.A.1. Table II.C.1 indicates, for example, that 19 percent of all Alabama children 13 to 18 years old were uninsured while 38 percent of the 13 to 18 year-olds in families under 50 percent of poverty were uninsured.

Table II.D.1 presents estimates of the number of Alabama children who were simulated to be eligible for Medicaid, without regard to their existing insurance coverage. Very small numbers with family incomes above 200 percent of poverty appear to be eligible, indicating the impact of income disregards that extend eligibility beyond the nominal limits. Table II.E.1 shows how many of these simulated Medicaid-eligible children were uninsured the previous year. Note that none of the children who were simulated to be eligible with family incomes above 200 percent of poverty were uninsured. Table II.F.1 provides estimates of the number of children who were simulated to be eligible for either Medicaid or SCHIP.17 Table II.F.1 shows that about 533,000 Alabama children were simulated to be eligible for either program when existing insurance coverage was ignored while Table II.G.1 shows that only 141,000 of these eligible children were actually uninsured the previous year.

Tables II.H.1, II.I.1, and II.J.1 are direct sample estimates from the CPS, and they correspond in content to Tables II.A.1, II.B.1, and II.C.1. Table II.H.1 shows all children, and it is about 90,000 lower than the total represented in Table II.A.1. The difference is due to our use of population controls that are external to the CPS and that include detail for children under 15.18 While the totals are nearly the same, however, there are striking differences between Tables II.A.1 and II.H.1 in the estimates for combinations of poverty level and age. For example, Table II.A.1 shows almost 7,000 infants below 50 percent of poverty while table II.H.1 shows double that number (and half as many as the number 1 to 5). The differences between Tables II.B.1 and II.I.1, which show the number of uninsured children, are even more dramatic. Table II.I.1 contains several cells for which there were no sample observations whereas Table II.B.1 has no empty cells. The totals are different because our model-based estimate of the uninsured rate in Alabama was higher than the CPS direct sample estimate (see Table I.1). The relative numbers of children in adjacent age categories in Table II.I.1 fluctuate dramatically across poverty levels. For example, there are more than twice as many 13 to 18 year-olds as 6 to 12 year-olds between 100 and 150 percent of poverty but 10 times as many 6 to 12 year-olds as 13 to 18-year olds between 200 and 350 percent of poverty. In Table II.B.1 the relative numbers of children in the two age groups vary little by poverty level until the top category. Finally, the uninsured rates reported in Table II.J.1 show sharp fluctuations across the table and even between adjoining cells whereas those reported in Table II.C.1 are much smoother.

Footnotes:

1. An earlier application of the methodology employed in this report was used to assist the State of New Jersey in its SCHIP design and planning efforts (see Czajka, Rosenbach, and Schirm 1999).

2. Recognizing the limitations of the CPS, Congress has appropriated funds for enhancing the CPS to support more precise state estimates of the numbers of uninsured children. But, such enhancements have not yet been designed, let alone implemented, and it is unlikely that the expanded sample will support detailed breakdowns of uninsured children.

3. Estimators that borrow strength have been used successfully in administering important public programs. For example, the Bureau of Labor Statistics (BLS) uses data from administrative sources to help construct the monthly estimates of state unemployment rates. Another estimator has been used for several years to derive state estimates for allocating federal funds under the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) (Schirm and Long 1995). Similar estimators have also been used to obtain state and county estimates of poor school-aged children for allocating federal Title I funds for compensatory education in elementary and secondary schools (National Research Council 1998). Thus the methodology that we employ in producing the tables presented in this report follows a long line of successful applications of statistical enhancements to the CPS for the purpose of preparing estimates at the state level.

4. In contrast, with the original CPS sample weights, the database looks like the whole United States.

5. Empirical Bayes shrinkage methods average direct sample estimates with predictions from regression models that derive their predictions based on state characteristics measured by decennial census and administrative records data (e.g., the poverty rate according to the census or the ratio of children enrolled in Medicaid, according to Medicaid administrative data, to the total population of children, which is derived from a combination of census and administrative data).

6. Prior to reweighting the CPS database to borrow strength, we controlled the weights of households within each state to a set of population totals characterizing the age and race/ethnic structure of the state's child population (the same set of totals that is later used in the reweighting). Such control totals are not used by the Census Bureau in weighting the CPS because of the emphasis placed on using the survey for producing employment estimates. Indeed, the only state-level control on weights used by the Census Bureau is the total population aged 16 and over, which is regarded as the population of working age. Because we are developing estimates pertaining to the child population, our within-state adjustment of weights to child population totals should improve the precision of the estimates. Then, borrowing strength by reweighting should lead to further improvements in precision.

7. If the underlying database is too small, the estimates of incremental changes to programs tend to be either zero or excessively large.

8. For Medicaid, the most important limitation of our simulation is the exclusion of eligibility under state medically needy provisions. The CPS lacks the medical expenditure data required to simulate these provisions. For SCHIP the most important limitation is our inability to take into account the waiting periods that a number of states have introduced to discourage parents from dropping employer-sponsored coverage for their children and enrolling them in SCHIP. The CPS provides no information on duration of uninsurance and, furthermore, the details on state waiting periods that we would require in order to simulate them are not readily available. These waiting periods can reduce potential eligibility by a significant amount.

9. Poverty in this table is measured relative to the Census Bureau's poverty thresholds while program eligibility is based on the poverty guidelines released by the Department of Health and Human Services (DHHS). While generally similar, the two series differ in a number of respects, and this may account for the fact that small percentages of children below 200 percent of poverty are simulated to be ineligible for Medicaid or SCHIP in states with SCHIP eligibility limits of 200 percent. The Census Bureau poverty threshold may define 200 percent of poverty as a slightly higher dollar figure than the corresponding DHHS guideline used to determine eligibility, such that a small number of children fall between the two figures. In addition, the poverty guidelines recognize the markedly higher living costs in Hawaii and Alaska than the rest of the nation and are set at higher levels while the Census Bureau thresholds are undifferentiated across the states.

10. Access to coverage among low income children in Arkansas, currently, is actually better than this suggests. In September 1997, Arkansas implemented the ARKids First program under a section 1115 Medicaid waiver. Because of its late introduction, this program which provides coverage for children up to 200 percent of poverty is not included in our simulation of Medicaid eligibility in 1997 for the state of Arkansas. Nor is it included in the combined Medicaid and SCHIP eligibility simulation, which mixes Medicaid eligibility rules in 1997 with SCHIP eligibility rules as of September 1999. A simulation based on a later point in time would very likely include this extended coverage under SCHIP; state officials have indicated that there are plans to transform ARKids First into a state SCHIP (Irvin and Czajka 2000).

11. Comparison of CPS estimates with Medicaid administrative statistics indicates that the CPS undercount of Medicaid children under 15 grew by about 3 million children between the March 1994 and March 1998 surveys. In March 1994 the undercount was estimated to be about 17 percent or 3.2 million children. See Czajka and Lewis (1999) for a discussion.

12. Underreporting of coverage probably accounts for most of the Medicaid undercount. Persons who failed to report their Medicaid coverage could have reported another form of coverage during the year--either as an incorrect description of their Medicaid coverage or as other coverage that they actually had during the year. Part of the undercount could also be due to CPS underrepresentation of low income households. This has quite different implications, however. If low income households are underrepresented, then not only Medicaid enrollees but uninsured people will be undercounted as well. We are not aware of any evidence of an under-representation of low income households in the CPS, but the possibility is one that must be recognized. Under-representation of low income households might arise for many of the same reasons that the decennial census undercounts the population.

13. Both our ASPE project officers and the majority of a technical advisory committee recommended against any adjustment for the Medicaid undercount. The concern is not about the reality of a sizable undercount but the uncertainty as to what it implies about the estimate of the uninsured.

14. Researchers who apply microsimulation to estimate changes in eligibility under hypothetical program reforms are familiar with the limitations of reported participation under such scenarios. Schirm and Zaslavsky (1998) propose a solution to this problem.

15. As we explain in the Technical Appendix, one of the ways in which our methodology reduces the sampling error in state estimates of uninsured children is by incorporating more state-specific detail for children than the Census Bureau uses in weighting the March CPS. The added detail includes race and Hispanic origin and multiple age categories. In order to include this additional detail, however, we had to use population estimates that the Census Bureau prepares for July 1 of each year. We averaged the July 1 estimates for 1997 and 1998, yielding estimates that we then characterize as January 1, 1998.

16. The numeral 1 is appended to each of the 10 tables for Alabama, the numeral 2 to each of the 10 tables for Alaska, and so on.

17. We combine the two programs because the SCHIP eligibility criteria that we were asked to use refer to September 1999 while the Medicaid eligibility criteria apply to 1997. Medicaid eligibility is continuing to grow because of the phase-in of poverty-related eligiblity provisions, so subtracting the 1997 Medicaid eligibles from the 1999 SCHIP eligibles would attribute too much of the growth in eligibility to SCHIP.

18. As we noted in an earlier footnote, the Census Bureau applies no state-specific controls to the population under 16 when it constructs weights for the CPS.

Technical Appendix

In this Technical Appendix we document the methodology employed in creating the state estimates presented in this report. We begin with a brief overview and then present a step-by-step description of the procedures that we followed.

A. Overview

The Census Bureau assigned a sample weight to each of the approximately 50,000 households that responded to the March 1998 CPS. This weight indicates the number of households in the population that each sample household represents. The weight incorporates a number of factors in addition to a sample household's probability of being selected into the sample. These factors include an adjustment for nonresponding households and a series of corrections designed to bring the sample into closer agreement with independent estimates of the size, age and sex structure, and racial/ethnic composition of the population. The population controls used in weighting the CPS include only one total that is specific to each state: the number of persons age 16 and older. There are no state controls for the size or composition of the child population or the composition of the adult population. While several controls are applied at the national level, CPS state estimates of many characteristics are less accurate than they would be if controls were applied at the state level.

Each observation in the CPS sample was selected to represent households in only one state. For example, a sample household from Maryland with a weight of 4,000 represents 4,000 households in Maryland. When we borrow strength across states by the method of reweighting used here, we create 51 new weights for this Maryland household, and we distribute the household weight of 4,000 across the 51 states. The sample household continues to represent 4,000 households nationally, but it may now represent only 400 households in Maryland and 3,600 households spread across the other states. The other 3,600 Maryland households that this sample household previously represented are now represented by similar households from other states. The sample weights of all 50,000 CPS households are allocated across the 51 states in such a way that a set of state-specific control totals defined and constructed for this application is reproduced in each state.1

The step that creates the 51 state weights for each CPS household is in fact the last of 10 steps. While the assignment of 51 weights per sample household may sound complex, the nine steps that precede it constitute the bulk of the work in producing the reweighted database, and most of the detailed description that follows pertains to those nine steps. First, however, we review the basic design decisions that precede these estimation steps.

The specification of control totals is of great importance in determining how well a particular reweighting of a CPS database accomplishes its objective of supporting accurate state estimates. The process of specifying these control totals involves the interaction between what we would like to control, given the tabulations that we intend to produce, and what we are best able to control, given the data that are available.2 Control totals developed from external sources with little or no sampling error yield the greatest improvement in the accuracy of state estimates, providing that they are relevant to the tabulations that we wish to produce. Such controls allow us to borrow strength across data sources. Clearly, if we had access to error-free counts of uninsured children in each state we could greatly improve the accuracy of our state tabulations of uninsured children by poverty level and age. That we lack such controls, however, is one reason why we must use other methods of estimation.

The state control totals and the corresponding household-level characteristics to which the controls are applied are displayed in Table A.1. The variables are grouped according to the source of the control totals. For the variables that capture the age, race, and ethnic structure of a state's child population, we use population totals derived from administrative (mainly vital records) and decennial census data. The totals for the class A variables have essentially no sampling error, which, as we said, is a highly desirable property. Unfortunately, there are no such totals for the numbers of children in various poverty categories or the numbers of uninsured children. For those totals, we must rely on sample-based estimates. But rather than using direct sample estimates from the CPS, we improve their precision by using empirical Bayes shrinkage methods to produce totals for the class E variables. These methods average direct sample estimates with predictions from regression models. The dependent variable in such a regression is the direct sample estimate, and the predictors are state characteristics measured by decennial census and administrative records data (e.g., the poverty rate according to the census; the infant mortality rate, obtained from vital records; or the ratio of children enrolled in Medicaid, according to Medicaid administrative data, to the total population of children, derived from a combination of census data and administrative data).

For the last two control variables listed in Table A.1, the class D variables, we use direct sample estimates of their totals. The main purpose for including these two variables is to restrict somewhat the borrowing of strength from reweighting. Specifically, for the one state (the District of Columbia) with no households outside large central cities with substantial black or Hispanic populations, no weight is given to a household if it is not from such a central city.3 Likewise, for the 21 states that have no large central cities with substantial black or Hispanic populations, no weight is given to a household from such a central city in another state. For example, no Wyoming weight is given to a household from New York City.

CONTROL VARIABLES/TOTALS USED IN REWEIGHTING
TABLE A.1
Household Control Variable State Control Total
Class A: Variables for which we use Administrative estimates of totals
Number of children age 0 Population age 0
Number of children ages 1-5 Population ages 1-5
Number of children ages 6-13 Population ages 6-13
Number of children ages 14-18 Population ages 14-18
Number of Hispanic children ages 0-18 Hispanic population ages 0-18
Number of non-Hispanic black children ages 0-18 Non-Hispanic black population ages 0-18
Class E: Variables for which we use Empirical Bayes shrinkage estimates of totals
Number of children < 50% FPL Number of children < 50% FPL
Number of children 50 to < 100% FPL Number of children 50 to < 100% FPL
Number of children 100 to < 200% FPL Number of children 100 to < 200% FPL
Number of children 200 to < 350% FPL Number of children 200 to < 350% FPL
Number of uninsured children < 100% FPL Number of uninsured children < 100% FPL
Number of uninsured children 100 to < 200% FPL Number of uninsured children 100 to < 200% FPL
Number of uninsured children 200% FPL or greater Number of uninsured children 200% FPL or greater
Class D: Variables for which we use Direct sample estimates of totals
Indicator that household is in a large central city with a substantial black or Hispanic population Number of households in large central cities with substantial black or Hispanic populations
Indicator that household is not in a large central city with a substantial black or Hispanic population Number of households not in large central cities with substantial black or Hispanic populations

The steps needed to derive control totals can become complex for at least two reasons. First, empirical Bayes estimation is itself complex and may also include steps that entail elaborate operations--such as smoothing estimated variances. Second, if a state control is obtained from an external source or by using empirical Bayes estimation, its introduction as a control is likely to change--and generally improve--the estimates of other totals to which it is related. For example, the CPS does not control the size of the Hispanic population at the state level, and Hispanic children tend to have higher uninsured rates than non-Hispanic children. By introducing estimates of the state Hispanic population as controls, we may improve the precision of the state estimates of uninsured children that we also want to use as controls. Rather than introducing all of the controls simultaneously in one step, it is desirable to introduce them sequentially so that the controls introduced at one step can allow us to obtain better estimates of controls that can be introduced--along with the earlier controls--at a later step.

B. Step-by-Step Procedure

Development of the reweighted database required 10 steps:

  1. Derive estimates of Class A totals
  2. Adjust the weights within each state to reproduce the totals derived in Step 1
  3. Derive direct sample estimates of Class E totals using the weights from Step 2
  4. Select regression models to predict the Class E totals
  5. Derive empirical Bayes shrinkage estimates of Class E totals
  6. Adjust the weights within each state to reproduce the totals from Steps 1 and 5
  7. Derive direct sample estimates of Class D totals using the weights from Step 6
  8. Obtain adjusted totals for the Class A variables pertaining to numbers of Hispanic children and non-Hispanic black children
  9. Adjust the weights within each state to reproduce the totals from Step 1 (for the first four Class A variables), Step 8 (for the last two Class A variables), Step 5 (for the Class E variables), and Step 7 (for the Class D variables)
  10. Reweight the March 1998 CPS database from Step 9 to borrow strength across states, using the control totals from Step 1 for Class A variables, Step 5 for Class E variables, and Step 7 for Class D variables

We describe the 10 steps in detail below.

1. Derive Estimates of Class A Totals

The source of Class A totals was the Census Bureau's state population estimates by age, race, sex, and Hispanic origin. These estimates are based on the most recent decennial census and carried forward by a combination of vital statistics and other administrative data.

The population estimates published by the Census Bureau are sometimes described as "census-level" estimates because they are intended to represent the population counts that would be obtained if a decennial census were conducted. As is well known, there is a net undercount of the population by the census when it is actually conducted, and there would surely be a net undercount if a census were conducted sometime between 1990 and 2000. The Census Bureau's estimate of what the net undercount would have been had a census been conducted in, say, 1997 is the estimated undercount in the 1990 census. Accordingly, the Bureau has developed and published a "net population adjustment matrix" that contains for each state the estimated undercount by single year of age, sex, race, and Hispanic origin. When the Bureau publishes population estimates, the net undercounts are subtracted from the Bureau's best estimates of the actual population totals to obtain the published totals. To develop adjusted population estimates, we "undid" this last step, adding the net undercounts to the published population totals.

The published estimates refer to July 1 of each year. We averaged successive July 1 estimates to obtain estimates for January 1, the end of the reference period for much of the data collected in the March CPS.4

2. Control the Weights to the Totals Derived in Step 1

Applying the controls from Step 1 may alter individual state estimates of uninsured children and low income children. Before developing empirical Bayes estimates of uninsured children and low income children, therefore, we "raked" the CPS weights to the Class A totals derived in Step 1. Raking is a widely used procedure for adjusting sample weights. For a specified set of characteristics of the sampled population, it brings weighted sums obtained from the sample into agreement with totals obtained from external sources. The raking was done within each state--that is, there was no borrowing of strength across states at this point.

In an effort to avoid extremely large upward adjustments to weights in states with small numbers of Hispanics or blacks, we used four different raking models: (1) rake to all totals except the totals for Hispanics and non-Hispanic blacks, (2) rake to all totals except the total for Hispanics, (3) rake to all totals except the total for non-Hispanic blacks, and (4) rake to all totals. In general, we used the first model if a state has few Hispanics and non-Hispanic blacks. We used the second and third models if a state has relatively few Hispanics or relatively few non-Hispanic blacks, respectively. We used the fourth model for the remaining states.5

3. Derive Direct Sample Estimates of Class E Totals

Using the weights from Step 2, we calculated direct sample estimates of the Class E totals, which are needed in Steps 4 and 5.

We estimated percentages rather than counts (that is, the percentage uninsured rather than the total number) in order to standardize for state population size, which is necessary for the next two steps. For each of the four income variables, the denominator of the percentage is the total number of children. For the three uninsured variables, the denominator is the total number of children in the indicated poverty category.

We estimated variances and covariances for the direct sample estimates using a jackknife estimator, treating the CPS rotation groups as replicate samples. These estimates are required for the calculation of empirical Bayes estimates in Step 5.

4. Select Regression Models to Predict the Class E Totals

In developing regression models to predict state income distributions and uninsured rates, we considered a wide range of potential predictors, summarized in Table A.2. We selected models based on their predictive abilities. In addition, we checked for and did not find strong evidence of correctable, persistent bias in the predictions for groups of states defined by diverse characteristics, such as population size, percent Hispanic, and the other variables considered as predictors.

Regression models predicting the Class E totals were estimated for the March CPS samples for 1995, 1996, 1997, and 1998 so that we could evaluate the performance of alternative models in different years and select final model specifications based on their fit across the four years.


TABLE A.2

POTENTIAL PREDICTORS EVALUATED IN REGRESSION MODELS FOR POVERTY LEVEL AND UNINSURED RATE

Characteristics of and Participation in Social Welfare Programs

Participation in:

Food Stamp Program
National School Lunch Program
Supplemental Security Income
Medicaid
Unemployment Insurance

Fraction of children eligible for Medicaid by poverty category
Age distribution of children enrolled in Medicaid

Income and Poverty

Poverty rate among federal tax return filers and the nonfiler rate
Per capita income (from National Income and Product Accounts)
Median household income (from census)
Percentage of population by poverty category (census)
Percentage of child population by poverty category (census)

Demographic Characteristics of Population

Population total and population growth Racial/ethnic distribution--percentage black, percentage Hispanic
Migration--percentage noncitizen (census), net international migration rate
Urban/rural distribution (census)

Health and Vital Statistics

Immunization rate
Infant mortality rate and low birth weight rate Child death rate and teen violent death rate
Teen birth rate

Employment and Education

Proportion of jobs by sector (e.g., agriculture, manufacturing, services, government)
Proportion of jobs in small establishments
Proportion of adults who are self-employed (census)
Educational attainment of adults (no HS diploma, at least a BA) (census)

Living Arrangements

Percentage of children by number and employment status of parents in household (census)
Percentage of children institutionalized (census)
Percentage of nonelderly persons in nonfamily households (census)
Percentage of households with no children or nonfamily (census)


We selected a single best model for all four of the Class E poverty variables (that is, the percentage of children in each of the poverty classes listed in Table A.1). This model included the following predictors:

  • the child poverty rate according to individual income tax data, that is, the percent of child exemptions that are claimed on tax returns with income below the poverty level
  • the percentage of the population receiving food stamps
  • the percentage of children ages 1 to 18 in families with incomes less than or equal to 75% of the federal poverty level (FPL) according to 1990 census data
  • the percentage of children ages 1 to 18 > 75% to 100% FPL (from 1990 census)
  • the percentage of children ages 1 to 18 > 100% to 185% FPL (from 1990 census)
  • median household income (from 1990 census)
  • percentage of children ages 0 to 17 who are noncitizens (from 1990 census)

For the three Class E insurance coverage variables (the percentage uninsured in each of three poverty classes, which are listed in Table A.1) we identified separate best models; however, all three models included the following predictors:

  • the percentage of the population that is Hispanic
  • the nonelderly poverty rate according to individual income tax data, that is, the percentage of nonelderly exemptions that are claimed on tax returns with income below the poverty level
  • percentage of children ages 1 to 20 enrolled in Medicaid

The best model for the first insurance coverage variable, the percentage uninsured among children below 100 percent of poverty, also included the following predictors:

  • the percentage of children ages 0 to 17 who are living with two parents, only one of whom is in the labor force (from 1990 census)
  • the percentage of children ages 0 to 17 who are living with one parent who is in the labor force (from 1990 census)
  • the proportion of jobs that are in agricultural services, forestry, or fishing (as of 1996)

The best model for the second insurance coverage variable, the percentage uninsured among children between 100 and 200 percent of poverty, included the following predictors (in addition to the three listed above):

  • the percentage of children between 100 and 200 percent of poverty who are income-eligible for Medicaid
  • the percentage of children ages 1 to 18 > 100% to 185% FPL (from 1990 census)
  • the proportion of jobs that are in agricultural services, forestry, or fishing (as of 1996)
  • the proportion of jobs that are in retail trade (as of 1996)

The best model for the third insurance coverage variable, the percentage uninsured among children at or above 200 percent of poverty, included the following predictor (in addition to the three that were included in all three models):

  • the percentage of children ages 0 to 17 who are living with one parent who is in the labor force (from 1990 census)

The models have reasonable face validity; that is, the predictors have plausible relationships, generally, to the variables being predicted.

5. Derive Empirical Bayes Shrinkage Estimates of Class E Totals

While regression models were estimated for each of four years, empirical Bayes estimates were ultimately needed for just March 1998. We estimated the four poverty variables and three uninsured variables as weighted averages of the direct sample estimates calculated in Step 3 and regression predictions from the models selected in Step 4. The relative weighting of the direct sample estimates and regression predictions varied by state, depending on the state-specific variances of the direct sample estimates and the overall fit of the regression models (which does not vary by state).

We obtained estimated counts from the estimated percentages. We used estimates of the population ages 0 to 18 from Step 1 to convert the empirical Bayes estimates of poverty percentages to poverty counts. We then ratio-adjusted the state counts so that they would sum to direct sample estimates of national totals (from Step 3) for each of the four poverty variables. A ratio adjustment of this kind is standard practice in small area estimation; it is necessary because the counts derived from the empirical Bayes estimates do not necessarily sum to the national totals. We used the adjusted state poverty counts to convert the empirical Bayes estimates of uninsured percentages to uninsured counts, and we ratio-adjusted the state estimates of these three variables to the direct sample estimates of national totals.

6. Rake Weights to Totals from Steps 1 and 5

Using the four raking models that we applied in Step 2, but modified to include the Class E variables, we raked the weights within each state to the totals obtained in Steps 1 and 5. This step was necessary so that the Class D controls calculated as direct sample estimates in Step 7 would be consistent with the Class A and Class E controls.

7. Derive Direct Sample Estimates of Class D Totals

Using the weights obtained from Step 6, we calculated direct sample estimates of the two Class D totals in each state.

8. Obtain Adjusted Totals for the Class A Variables

The Class A controls pertaining to the numbers of Hispanic children and non-Hispanic black children could not be applied fully in Steps 2 and 5 because some states had too few sample observations in one or both of these two groups. At this point, then, the weighted sums of Hispanic children and non-Hispanic black children do not agree with the Step 1 controls at the national level. To correct this problem, we grouped the states in which the Hispanic control could not be applied earlier, and we ratio adjusted the direct sample estimates of Hispanic children for this group of states as a whole so that the adjusted totals sum to the Step 1 totals for the group. We then repeated the process to obtain adjusted totals of non-Hispanic black children.

9. Rake Weights within Each State

Within each state we raked the weights to totals derived from Step 1 (for the first four Class A variables), Step 8 (for the last two Class A variables), Step 5 (for the Class E variables), and Step 7 (for the Class D variables). In this step, there is just one raking model. We do not have to treat states with low percentages of Hispanics or blacks differently because we are raking to the state totals created in Step 8 rather than the Step 1 totals for Hispanic children and non-Hispanic black children.

10. Reweight the March 1998 CPS Database to Borrow Strength Across States

Using the control totals from Step 1 for Class A variables, Step 5 for Class E variables, and Step 7 for Class D variables, we applied the reweighting procedure to obtain 51 state weights for each sample household. With this procedure there are two constraints on the state weights: (1) all control totals must be satisfied for all states and (2) for each household, the national weight given to the household after reweighting--that is, the sum of the household's state weights--must equal the weight given to the household at the conclusion of Step 9. These constraints and the maximum likelihood estimation algorithm are described in detail in Schirm and Zaslavsky (1997).

References

Czajka, John L., and Kimball Lewis. "Using National Survey Data to Analyze Children’s Health Insurance Coverage: An Assessment of Issues." Washington, DC: Mathematica Policy Research, May 1999.

Czajka, John L., Margo L. Rosenbach, and Allen L. Schirm. "Uninsured Children in New Jersey: Estimates of Their Number and Characteristics." Washington, DC: Mathematica Policy Research, April 1999.

Irvin, Carol, and John L. Czajka. "Simulation of Medicaid and SCHIP Eligibility: Implications of Findings from 10 States." Washington, DC: Mathematica Policy Research, April 2000.

National Research Council. Small Area Estimates of School Age Children in Poverty, Interim Report 2, Evaluation of Revised 1993 County Estimates for Title I Allocations, edited by Constance F. Citro, Michael L. Cohen, and Graham Kalton. Panel on Estimates of Poverty for Small Geographic Areas, Committee on National Statistics. Washington, DC: National Academy Press, 1998.

Schirm, Allen L., and Alan M. Zaslavsky. "Model-based Microsimulation Estimates for States When State Programs Vary." 1998 Proceedings of the Section on Survey Research Methods. Alexandria, VA: American Statistical Association, 1998.

Schirm, Allen L., and Alan M. Zaslavsky. "Reweighting Households to Develop Microsimulation Estimates for States." 1997 Proceedings of the Section on Survey Research Methods. Alexandria, VA: American Statistical Association, 1997.

Schirm, Allen L., and Cindy Long. "Fund Allocation and Small Area Estimation in the WIC Program." 1995 Proceedings of the Section on Survey Research Methods. Alexandria, VA: American Statistical Association, 1995.

Zaslavsky, Alan. "Representing Local Area Adjustments by Reweighting of Households." Survey Methodology, vol. 14, no. 2, December 1988.

Topics
Child Welfare
Populations
Uninsured & Underinsured | Children