The Effects of SCHIP on Children's Health Insurance Coverage

An Update Based on Three Rounds of the Community Tracking Study, 1996-97 to 2000-01

Research Report No. 7
December 2002
Peter J. Cunningham, Jack Hadley, James D. Reschovsky

Introduction

This report provides an update to the paper, "The Effects of SCHIP on Children’s Health Insurance Coverage: Early Evidence from the Community Tracking Study," that was published in Medical Care Research and Review (Cunningham, Hadley, and Reschovsky, 2002). Whereas the earlier paper examined the effects of public program expansions on children’s health insurance coverage between the first two rounds of the Community Tracking Study household survey (between 1996-97 and 1998-99), this report extends that analysis by including the more recent data from the Round 3 CTS household survey (conducted in 2000-01). Other than the expanded time period, the analysis in this report is identical to the earlier paper. The tables in this report are also supplemental to Issue Brief #59, "SCHIP, Medicaid Expansions Lead to Shifts in Children’s Coverage" (Cunningham, Reschovsky, and Hadley, 2002).

Below is a brief summary of the methodology and results. The earlier paper (Cunningham, Hadley, and Reschovsky, 2002) provides a more detailed description of the conceptual framework and methodology for the study.

Summary of methodology

Data Source. The CTS household survey is designed to produce representative estimates of insurance coverage and people’s experiences using health care for the U.S. population as well as for the population in 60 randomly selected communities. Three rounds of the survey have been completed. The first round was completed in 1996-97, just before the passage and implementation of SCHIP, and subsequent rounds were completed in 1998-99 and 2000-01.

The sample for the analysis presented in the tables below includes about 10,000 children age 19 and under and in families with incomes below 200% of poverty. This includes 3,758 children in the 1996-97 sample, 3,372 children from the 1998-99 sample, and 2,865 children from the 2000-01 sample.

Analytical approach. We use a variant of the difference-in-differences approach to identify the effects of changes in public program eligibility and health insurance costs on children’s health insurance coverage. Essentially, the analysis compares changes in coverage for children in states that experienced larger increases in public program eligibility (and health insurance costs) to children in states that experienced smaller increases in eligibility or costs.

The following equation is used to illustrate the basic analytical specification:

Thus, health insurance coverage is expressed as a function of baseline public program eligibility and health insurance costs (prior to SCHIP), the changes in eligibility and costs between the 1996-97 and 2000-01 CTS surveys, and other individual and community factors related to the demand for health insurance.

This model is identical to that used in the study by Cunningham, Hadley, and Reschovsky (2002) with two exceptions. First, a dummy variable for the R3 survey was added to reflect the inclusion of the more recent data. In addition, the measures of changes in eligibility and premiums were extended to reflect the change between the Round 1 and Round 3 surveys (i.e. between 1996-97 and 2000-01), whereas the previous study examined changes in eligibility and premiums between the Round 1 and Round 2 survey.

The analysis is based on a multinomial logistic regression model that examines the effects of eligibility and premium changes on the choice of private insurance vs. Medicaid or other state coverage (SCHIP), the choice of private insurance vs. uninsured, and the choice of uninsured vs. Medicaid or other state coverage.

Eligibility for Medicaid/SCHIP. Eligibility is measured at the state level based on a standardized population. Specifically, the percentage of children eligible for Medicaid or other state programs is computed for each survey round by applying the state and year specific eligibility criteria to a nationally representative sample of children using the Round 1 CTS household survey. Using a standardized population to compute eligibility holds constant population characteristics across communities and over time. Thus, the state level measures of eligibility (and changes in eligibility) reflect differences and changes in states’ policies rather than variations in population characteristics and economic circumstances across states and over time.

Health insurance premiums. Direct measures of the cost of health insurance to individuals were not obtained in the survey. Instead, information obtained from the 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey was used to compute the average cost of a standardized employer-sponsored insurance policy in each of the 60 CTS communities and for each time period. The RWJF Employer Health Insurance Survey includes a representative sample of employers for each of the 60 CTS communities.

Because data on health insurance premiums were obtained for only a single year, changes in premiums between 1996 and 2000 for each of the 60 CTS communities were estimated using a regression-based model to predict average premiums for each of the time periods. The prediction model for average premiums in the community included exogenous firm, state, and site characteristics. Estimates of average premiums for 1996 and 2000 were then obtained by combining the parameters from the regression model with year and state or site-specific means for each of the variables in the prediction model. Estimates of average premiums were adjusted for cross-sectional variations in the cost of living and for general inflation.

The means and standard deviations for the eligibility and premium variables, as well as all other variables included in the multinomial regression models are shown in Table 1.

Changes in coverage. Table 2 shows that the percent of low income children with Medicaid or SCHIP coverage increased between 1996-97 and 2000-01, while the percent of children who were uninsured or had private coverage decreased during this time period. While Medicaid or SCHIP coverage increased throughout the study period, all of the decrease in private coverage occurred between 1996-97 and 1998-99, while almost all of the decrease in the percent uninsured occurred between 1998-99 and 2000-01. Near-poor children (between 100-200% of poverty) experienced the largest changes in coverage.

Multinomial regression results. Table 3 summarizes the effects of changes in eligibility and premiums on children’s health insurance coverage between 1996-97 and 2000-01 for different income groups, and updates the findings from Table 3 in the earlier paper (Cunningham, Hadley, and Reschovsky, 2002). Table 4 of this report presents the full multinomial logistic regression results for low income children (i.e. less than 200 percent of the federal poverty level), and updates the findings in Table 4 of the earlier paper. All results are presented as odds ratios. Thus, when examining the results for the effects of increases in eligibility on the likelihood of having private vs. Medicaid or other state coverage, an odds ratio of less than one indicates that an increase in eligibility decreases the odds of having private insurance vs. Medicaid or other state coverage. An odds ratio of greater than one indicates a higher likelihood of having private insurance vs. Medicaid or other state coverage, while an odds ratio of one indicates no effect.

Among all low income children (less than 200% of poverty), an increase in eligibility for public coverage (as reflected by the change in eligibility * R3 parameter) was associated with a decreased likelihood of having private versus Medicaid or other state coverage (Table 3). Although the magnitude of the coefficient is very similar to the earlier result (Cunningham, Hadley, and Reschovsky 2002), it is statistically significant at the .05 level, while the earlier result was not statistically significant. As with the earlier study, there was no statistically significant effect of the change in eligibility on private coverage vs. uninsured, as well as uninsured vs. Medicaid/other state coverage.

The effects of changes in eligibility on coverage for poor children (below poverty) as well as near poor children (between 100-199% of poverty) were also similar to the earlier study. The effect of changes in eligibility on any of the coverage variables for poor children was not statistically significant. Although the magnitude and direction of some of these effects differs somewhat from the earlier study, neither the earlier study nor the updated results show any statistically significant effects of eligibility changes on coverage of poor children. Estimates for poor children are less precise than for all low income children due to smaller sample sizes.

For near poor children (between 100-199% of poverty), eligibility changes decreased the likelihood of having private insurance coverage vs. Medicaid or other state coverage (p < .05), and the magnitude of the effect was similar to the earlier study (O.R = .91 compared to .88 in the earlier study). Although eligibility changes decreased the likelihood of being uninsured vs. having Medicaid/other state coverage, the result was not statistically significant. This differs somewhat from the earlier study in that the effect of eligibility changes between 1996-97 and 1998-99 on the likelihood of being uninsured vs. Medicaid/other state coverage was marginally significant (p < .10), and the magnitude of the coefficient in the earlier study was somewhat larger than that in Table 3. As the simulation results in Table 5 will show, however, these small differences with the earlier study do not change the basic result that eligibility expansions did not affect children’s uninsurance rates.

In general, the effects of changes in premiums on coverage were somewhat weaker than in the earlier paper. This may due to the fact that the premium measures were based on a prediction model from 1997, and there is greater error in the predictions for later years. Nevertheless, the overall conclusion is unchanged, which is that premium increases decreased the likelihood of having private insurance coverage, and these effects were concentrated primarily among poor children.

Impact of changes in eligibility on coverage. The results of the multinomial logistic regression model are used to quantify the changes in coverage due to public program expansions by simulating coverage rates for 2000-01, assuming that eligibility remained at 1996-97 levels. In computing these simulations, we calculated the predicted probabilities for each of the 2000-01 observations in our sample, setting the change in eligibility measure to 0. These individual predictions were then averaged using survey weights. The results are shown in Table 5, which is an update of Table 5 from the earlier paper.

These results show that the decreases in private insurance coverage and the increases in Medicaid/other state coverage would have been smaller had there been no expansions in eligibility for public coverage. Based on these results, we estimate that about 23 percent of the total increase in Medicaid/state coverage among all low income children was due to substitution (sometimes referred to as crowd-out) of public for private coverage. Among near-poor children (between 100-200% of poverty), about 39 percent of the increase in Medicaid/state coverage can be attributed to substitution. These estimates are almost identical to the earlier estimates based on changes between 1996-97 and 1998-99.

By contrast, the results imply that the decrease in uninsurance rates would have been virtually the same even without the increases in eligibility expansions. However, we suspect that the decrease in uninsurance rates is related to other characteristics of the SCHIP and Medicaid program that are correlated with the eligibility expansions. One indication of this is that the dummy variable for R3 observations has a large and statistically significant effect on the likelihood of being uninsured vs. Medicaid and other state coverage (see Table 4). In effect, the coefficient on this dummy variable incorporates all other unmeasured factors associated with changes in coverage, such as state outreach and enrollment efforts. The fact that it has such a large effect on the likelihood of being uninsured vs. Medicaid and other state coverage indicates that much of the decrease in uninsurance rates between 1996-97 and 2000-01 is not explained by the other factors in the regression model.

As a test, we re-estimated the multinomial logistic regression models excluding the round 2 and round 3 dummy variables. When these variables are excluded, the effects of eligibility expansions on the likelihood of being uninsured vs. having Medicaid/other state coverage were much larger and had a high level of statistical significance (i.e. larger expansions greatly reduced the likelihood of being uninsured vs. having Medicaid/other state coverage). This indicates that increases in eligibility are correlated with characteristics-possibly other state program characteristics-that may have a much more direct impact on reducing the uninsurance rate. These may include efforts to reduce administrative barriers to enrollment, such as shortening application forms and eliminating asset tests, as well as outreach efforts designed to identify eligible uninsured children and encourage parents to enroll their children in the program.

This report updates an earlier analysis of the effects of SCHIP on the health insurance coverage of low income children between 1996-97 and 1998-99 (Cunningham, Hadley, and Reschovsky 2002) by including more recent data from the 2000-01 CTS household survey.

In general, the results are very similar to the earlier analysis. Increases in public program eligibility between 1996-97 and 2000-01 resulted in a decrease in private insurance coverage relative to Medicaid and other state coverage among low income children. The results were strongest among the primary SCHIP target population: children in families with incomes between 100-200% of poverty. Overall, we estimate that about 23 percent of the increase in Medicaid/SCHIP coverage for low income children was due to substitution of public for private coverage. For near-poor children (incomes between 100-200% of poverty), 39 percent of the increase in Medicaid/SCHIP coverage is due to substitution.

As with the earlier analysis, increases in public program eligibility appeared to have little effect on uninsurance rates among low income children. This was not surprising for the earlier analysis of changes between 1996-97 and 1998-99, since uninsurance rates were essentially unchanged during this period. However, uninsurance rates among low income children decreased by 4 percentage points between 1998-99 and 2000-01. Although eligibility expansions alone do not appear to have caused this decrease, it is possible that other program characteristics correlated with eligibility expansions account for this decrease. In future research, we hope to explicitly identify other Medicaid and SCHIP program characteristics in the analyses, especially state efforts to reduce administrative barriers and outreach activities designed to encourage enrollment.

Table 1
Means and Standard Deviations of Independent Variables Used in Multivariate Analysis. Children in Familis with Incomes Below 200% of Poverty

Omitted categories include white, poor health, not a risk-taker, and South region.
1Adjusted for differences in the cost-of-living across the 60 CTS sites as well as for general inflation between 1996 and 2000.
2Values of parents were used for children.
Source: HSC Community Tracking Study Household Survey, 1996-97, 1998-99 and 2000-01.

Table 2
Health Insurance Coverage if Low Income Children (age 19 and under)

*Difference with previous year is statistically significant at .05 level.
#Difference between 1997 and 2001 is statistically significant at .05 level.
"Other" includes those covered by military insurance, Indian Health Service, Medicare, as well as a few covered by miscellaneous other public programs.
Source: HSC Community Tracking Study Household Survey, 1996-97, 1998-99 and 2000-01.

Table 3
Effects of Eligibility and Health Insurance Premiums on Likelihood of choosing Private Insurance Coverage (Odds Ratios)

*p<.10 level.
**p<.05
Results are based on multinominal regressions that also control for other individual and community characteristics (see Table 4 for full results).
1Includes state-specific SCHIP programs.
Source: HSC Community Tracking Study Household Survey, 1996-97, 1998-99 and 2000-01.

Table 4
Results of Multinominal Logits of the Likelihood of Choosing Private Insurance Coverage vs. Medicaid/Other State Coverage or Uninsured. Children with Family Incomes Less than 200% of Poverty. Results are Presented as Odds Ratios.

** p<.05
*p<.10
Omitted categories include white, poor health, not a risk taker, and South region.
1 Includes state-specific SCHIP programs.
Source: HSC Community Tracking Study Household Survey, 1996-97, 1998-99 and 2000-01.

Table 5
Changes in Coverage Between 1996-97 and 1998-99. Actual Change in Coverage and Simulated Change in Coverage Assuming No Change in Eligibility.

1Includes state-specific SCHIP programs
2Includes Medicare, CHAMPUS, and Indian Health Service.
3Based on the weighted average of the individual predictions from the model, computed separately for the 1996-97 sample and the 2000-01 sample using their actual values for the independent variables.
4Based on the weighted average of the individual predictions for the 1998-99 sample, setting the change in eligibility parameter equal to 0.

Source: HSC Community Tracking Study Household Survey, 1996-97, 1998-99 and 2000-01.

Cunningham, Peter J., Hadley, Jack, and James Reschovsky. 2002. "The Effects of SCHIP on Children’s Health Insurance Coverage: Early Evidence from the Community Tracking Study," Medical Care Research and Review, Vol. 59 (4): 359-383,

Cunningham, Peter J., Reschovsky, James, and Jack Hadley. 2002. "SCHIP, Medicaid Expansions Lead to Shifts in Children’s Coverage," Issue Brief #59. Washington: Center for Studying Health System Change (December).