Insurance Premiums and Insurance Coverage of Near-Poor Children

Transcription

1 Jack Hadley James D. Reschovsky Peter Cunningham Genevieve Kenney Lisa Dubay Insurance Premiums and Insurance Coverage of Near-Poor Children States increasingly are using premiums for near-poor children in their public insurance programs (Medicaid/SCHIP) to limit private insurance crowd-out and constrain program costs. Using national data from four rounds of the Community Tracking Study Household Surveys spanning the seven years from 1996 to 2003, this study estimates a multinomial logistic regression model examining how public and private insurance premiums affect insurance coverage outcomes (Medicaid/SCHIP coverage, private coverage, and no coverage). Higher public premiums are significantly associated with a lower probability of public coverage and higher probabilities of private coverage and uninsurance; higher private premiums are significantly related to a lower probability of private coverage and higher probabilities of public coverage and uninsurance. The results imply that uninsurance rates will rise if both public and private premiums increase, and suggest that states that impose or increase public insurance premiums for near-poor children will succeed in discouraging crowd-out of private insurance, but at the expense of higher rates of uninsurance. Sustained increases in private insurance premiums will continue to create enrollment pressures on state insurance programs for children. Recent data show a drop in children s private insurance coverage from 69.8% to 65.6% between 2000 and 2004 (DeNavas-Walt, Proctor and Lee 2005), coinciding with sharp increases in private insurance premiums (Gabel et al. 2005). At the same time, a number of states increased premiums in their State Children s Health Insurance Program (SCHIP) as a way of discouraging crowd-out (substituting public coverage for private coverage) and limiting public enrollment and costs (Ross and Cox 2003; Ku and Nimalendran 2003). Fifteen states increased SCHIP premiums between January 2003 and March 2004 (Fox and Limb 2004). 1 As a result of growing enrollments and escalating program costs, the National Governors Association advocated giving states more flexibility to impose premiums in their Medicaid/SCHIP programs (National Governors Association 2005). The recent increases in both public and private Jack Hadley, Ph.D., is a principal research associate at the Urban Institute, and a senior fellow at the Center for Studying Health System Change. James D. Reschovsky, Ph.D., is a senior health researcher, and Peter Cunningham, Ph.D., is a senior fellow, both at the Center for Studying Health System Change. Genevieve Kenney, Ph.D., is a principal research associate at the Urban Institute. Lisa Dubay, Ph.D., is a research scientist at the Johns Hopkins School of Public Health. The Robert Wood Johnson Foundation provided primary funding for this research through its support of the Center for Studying Health System Change; the David and Lucile Packard Foundation provided additional support through a grant to the Urban Institute. Address correspondence to Dr. Hadley at the Urban Institute, 2100 M St., N.W., Washington, DC jhadley@ui.urban.org. Inquiry 43: (Winter 2006/2007). Ó 2006 Excellus Health Plan, Inc /06/

2 Near-Poor Children insurance premiums make it important to understand whether and how children s insurance coverage changes in response to variations in the cost of insurance. This analysis estimates the effects of public and private insurance premiums on the pattern of insurance coverage among near-poor children with family incomes between 100% and 300% of the federal poverty level (FPL). The results of this analysis, which used data from four rounds of a national survey spanning the years between 1996 and 2003, suggest that both public and private insurance premiums influence children s insurance coverage. Higher public insurance premiums are associated with a lower probability of public coverage, and higher probabilities of being uninsured or having private coverage. Conversely, higher private insurance premiums are associated with a lower probability of private coverage and a higher probability of public coverage. Background The Balanced Budget Act of 1997 encouraged states to expand children s eligibility for public insurance coverage by either raising the income and age ceilings of their Medicaid programs and/or establishing separate State Children s Health Insurance Programs to complement their Medicaid programs. To discourage the substitution of public coverage for private coverage and to limit budgetary costs, the law allows the 35 states that chose the SCHIP option to charge premiums to families with incomes above 100% of the FPL. Fourteen states and the District of Columbia chose to expand their Medicaid programs rather than establish a separate SCHIP, and five of these states have waivers from the Centers for Medicare and Medicaid Services (CMS) to charge premiums. 2 Under the federal Deficit Reduction Act of 2005 (Public Law ), Congress provided states with greater flexibility to impose premiums on children eligible for Medicaid. Several studies have examined trends in SCHIP or Medicaid enrollment in individual states that recently increased premiums for program participants (Kappel 2004; O Brien et al. 2000; Mann and Artiga 2004; Wright et al. 2005). These studies generally found that enrollment fell and substantial proportions of those who disenrolled cited premium cost as a factor. 3 While highly suggestive of a negative relationship between SCHIP premiums and children s public insurance coverage, these studies nevertheless had several limitations. They typically did not control for other factors, especially changes in the cost of private insurance that may have affected enrollment; they focused primarily on people who disenrolled from coverage rather than those who may have been discouraged from applying; and they have been difficult to generalize from because of potentially unique characteristics of individual state programs and general economic and political circumstances. Two studies have used multivariate methods to analyze individual state experiences in Florida (Shenkman et al. 2002) and in New Hampshire, Kansas, and Kentucky (Kenney et al. 2006/ 2007). Both found that disenrollment increased and new enrollment decreased following the premium hikes. Finally, only two prior studies have included premium information in multivariate analyses of national data on children s health insurance coverage, and both found that variations in premiums significantly influenced the distribution of insurance coverage. However, Cunningham, Hadley, and Reschovsky (2002) only had information on private insurance premiums, and Kronebusch and Elbel (2004) only had information on public insurance premiums and only analyzed effects on public coverage. While prior research consistently has suggested a negative relationship between public insurance premiums and public coverage, none has analyzed the effects of both public and private insurance premiums and none has provided quantitative estimates that can be applied confidently to the current policy debate. We address these limitations in several ways: by examining the effects of both public and private insurance premiums on the distribution of children s insurance coverage using data from four national surveys that span the years between 1996 and 2003, when many states implemented or changed SCHIP premiums; by controlling for unobserved differences across states and over time in multivariate models with state and time fixed effects; and by measuring public insurance premiums at the per child level, which increases intrastate variability. Methodology Conceptual Framework The empirical model posits three possible insurance choices for children eligible for public 363

3 Inquiry/Volume 43, Winter 2006/2007 insurance public coverage (Medicaid, SCHIP, or similar state programs), private coverage (employer-sponsored insurance [ESI] or private nongroup), and being uninsured and assumes that the coverage chosen depends on the costs of alternative insurance options. Private insurance and some states public insurance plans require families to pay a premium to obtain coverage. The cost of being uninsured is the potential burden of having to pay the full cost of medical care out of pocket. This burden should be related to the child s age and health, and the family s income. The model also includes other family characteristics that represent possible differences in preferences or the perceived value of health insurance coverage. Obtaining public coverage also may impose other, nonfinancial costs, such as a waiting period or minimum number of months of prior uninsurance, the time and inconvenience of applying for public coverage, the possible stigma of obtaining care through a public insurance program, and more limited provider choice (Kronebusch and Elbel 2004; Ross and Cox 2000). These cost dimensions are very difficult to measure accurately. Therefore, except for the waiting period and the recent imposition of caps on Medicaid/SCHIP enrollment in a few states, we control for unobserved differences in other Medicaid program characteristics and other factors that may vary across states and over time through a set of state, year, and state-year interaction dummy variables. A key assumption of this model is that the cost of health insurance actually incurred by a family for the option it chooses is inherently endogenous. The family s insurance decision is likely to be influenced to some extent by the family s preferences for health insurance generally or for a particular type of health insurance, and those preferences can lead to work and employment choices that determine whether the cost of the insurance option selected is low or high. For instance, parents can influence whether their children are eligible for public coverage by altering their labor supply. They may also have some flexibility in choosing whether to seek a job that offers employer-sponsored health insurance. Parents also can decide how aggressively to shop for possible nongroup coverage in the individual insurance market. Consequently, the empirical specification of the model uses measures that are in effect instrumental variables for the public and private insurance premiums that a family faces, rather than the premium they actually pay, and imputes eligibility for public coverage based on lagged income. Data and Sample The primary data for the analysis come from pooling separate cross-sectional data from the four Community Tracking Study (CTS) Household Surveys fielded in 1996/1997, 1998/1999, 2000/2001, and (Except for the 2003 survey, data generally were collected over a 12- month field period starting in July of the first year.) Interview households were drawn randomly from 60 communities (located in 34 states and the District of Columbia) that were selected to be representative of the population in the 48 contiguous states and the District of Columbia. Information was obtained for approximately 60,000 people in each of the first three rounds and 47,000 people in The analysis sample was limited to 13,254 children in families with incomes between 100% and 300% of the federal poverty level. Children in poverty may not be charged premiums and very few children above 300% of the FPL are eligible for Medicaid/SCHIP coverage. Insurance status was based on coverage at the time of the survey. Children covered by Medicare, military insurance, or the Indian Health Service were excluded because these types of coverage typically do not involve the choices under investigation. We also excluded children in three CTS states (Minnesota, Tennessee, and Wisconsin) with public insurance premiums that applied only to families (children and parents), rather than just children. The surveys provide information on children s health insurance coverage and personal and family characteristics drawn from telephone interviews (supplemented by in-person interviews for people without a telephone). Although the data are generally very rich, they do not provide direct information on insurance premiums actually paid for children covered by public insurance or whether a near-poor child was eligible for public coverage if he/she did not have public coverage at the time of the interview. Therefore, we imputed eligibility based on the child s age, lagged family income, and state rules at the time of the interview. The surveys collect information 364

4 Near-Poor Children on premiums for private nongroup coverage, but not for employer-sponsored insurance. 4 The statistical analysis also used area-level data on Medicaid and SCHIP eligibility and premiums (by state) and private insurance premiums. Medicaid and SCHIP eligibility and premium rules and schedules were obtained by state and year from several published sources (details are given later) and applied to individual children based on age, family income relative to poverty, and family size. The private insurance premium measure was constructed from published data from the Medical Expenditure Panel Survey Insurance Component (MEPS-IC) 5 by state and year, and imputed to the county level using information from the Census Bureau s County Business Patterns (CBP) on the distribution of workers by firm size (details follow). Variables Public insurance eligibility and premiums. We imputed a child s potential eligibility for Medicaid or SCHIP by developing an algorithm that compared the child s family income relative to the FPL for the year before the interview to the age-specific maximum income allowed by the Medicaid or SCHIP program in the child s state at the time of the interview. Since the available income measure was for annual family income from all sources in the previous year, rather than current monthly income, we did not attempt to adjust income for work or child care allowances. Consequently, we assumed that the imputed eligibility dummy variable, which was used to indicate public coverage eligibility and assign a public premium to potentially eligible children, was in effect an instrumental variable for actual current eligibility, since it was based on family income relative to poverty in the prior year. 6 We obtained information on public insurance premiums from Web sites maintained by the Centers for Medicare and Medicaid Services (CMS 2004) and the Assistant Secretary for Policy and Evaluation of the Department of Health and Human Services (ASPE 2004), and from several published sources (American Academy of Pediatrics 1999; Steinberg 1999; National Conference of State Legislatures 2000, 2001). 7 We extracted information on the date when premiums went into effect or changed, the ages and family incomes relative to the FPL of children liable for a premium, the amount of the premium (which varied with income relative to poverty in some states), and the maximum premium a family could be charged, if any. Some states do not have family maximums, and some states do not vary the premium with the number of children in the family. Most states implemented premiums over approximately two years starting in Twelve states changed their premiums at least once by We imputed premiums to children potentially eligible for public coverage based on their state s premium policies as of the date the child s family was interviewed for the CTS Household Survey. The public premium was set to zero for children who were not potentially eligible for public coverage or who lived in a state that did not charge a premium. (Models estimated with data for all near-poor children include a dummy variable for potential eligibility.) Thus, for identification purposes, premiums vary across states because of differences in premium policies, within states because of differences in effective premiums per child by family size and family income relative to poverty, and over time because of differences in implementation dates and changes in premium amounts and eligibility parameters. We adjusted all premiums for general inflation using the Consumer Price Index and expressed them in 2003 dollars. Public insurance waiting periods and enrollment caps. To discourage parents from dropping private coverage and enrolling their children in Medicaid/SCHIP, a number of states require that children be uninsured for a specified number of months before being allowed to enroll. The length of the waiting period (in months) has been assigned to eligible children based on their state s waiting period in effect at the time of the interview. In addition, six states in the CTS survey instituted enrollment caps at some point between 2001 and The presence of a cap was indicated by a dummy variable that takes the value 1 if the cap was in effect at the time of the interview. Private insurance premium. We constructed the instrumental variable for the private insurance premium using a method developed by Dubay and Kenney (2006) and data from two sources: 1) published annual data collected by the MEPS- IC 8 on offer rates, employee out-of-pocket premiums, and total premiums for family coverage by firm size and state, and 2) annual county data 365

5 Inquiry/Volume 43, Winter 2006/2007 on the distribution of employment by firm size from the CBP. We constructed county-level estimates of private insurance premiums that vary over time and across counties using equation 1 to combine the MEPS-IC and CBP data: privprem jk ¼ empshr ijk ½ðoffrate ik * empprm ik Þð1Þ þð1 offrate ik Þ* totprm ik Š; where empshr ijk ¼ share of workers; firm size i; county j; state k; offrate ik ¼ average offer rate; firm size i; state k; empprm ik totprm ik ¼ employees average out-of-pocket premium; firm size i; state k; ¼ employers average total premium; firm size i; state k: The term in brackets represents a weighted average of employees average out-of-pocket premium and employers average total premium, where the weight is the offer rate (offrate ik ) for firms of different sizes (fewer than 10, 10 to 24, 25 to 99, 100 to 999, and 1,000 or more workers). This expression was calculated for each firm size by state and year. We assume that the average offer rate represents the probability that a family has access to employer-sponsored insurance and faces the lower employee contribution as its out-of-pocket premium (Blumberg, Nichols, and Banthin 2001; Chernew, Frick, and McLaughlin 1997). In the second part of the expression, (1- offrate ik ) represents the probability of purchasing insurance in the nongroup market. We assume that the employer s average total premium represents the cost of nongroup insurance if the family does not have an insurance offer from an employer. 9 The firm size-specific premiums for each state then were averaged and assigned to individual counties within the state using the shares of workers in firms of different sizes in the county (empshr ijk ) as weights. Private premiums were assigned to children based on parents work status and county of residence. Children in families with either no workers or only self-employed workers were assumed not to have access to employer-sponsored insurance and were assigned the state-level total premium paid by firms with fewer than 10 employees. 10 Other children were assigned the county-level private premium computed from the previous formula. 11 This approach to constructing an instrumental variable for the endogenous (and largely unobserved) actual private premium effectively assumes that: a family faces a lower cost for private insurance if at least one parent is an employee and has potential access to employersponsored insurance, and lives in a county where the offer rate is relatively high and where employees pay a relatively low out-of-pocket premium. In counties where the offer rate is low and the total premium for a family policy is high, private insurance will be more costly. The county-level estimates of the cost of private insurance were assigned to families by year and county of residence and expressed in 2003 dollars. Unlike the public insurance premium, which is measured per child, the private insurance premium is measured per family. 12 Although the constructed variable attempts to compensate for the lack of data on families ESI out-of-pocket premiums and the endogeneity of the premium actually paid, it may be subject to measurement error because of the assumptions implicit in its construction. Subsequent sensitivity analysis explored alternative measures developed from CTS data on families ESI offers and premiums paid for nongroup coverage. State and year fixed effects. Differences exist among states and over time in many aspects of design and administration of their public insurance programs for children. These include: the complexity, timing and convenience of the eligibility determination process; the extent of outreach and information provided to potentially eligible families; the need to satisfy an asset test; the length of time public coverage is provided without redetermining eligibility; the extent to which covered children are enrolled in managed care organizations; and potential coverage of parents. Rather than try to control explicitly for the effects of all the other Medicaid/SCHIP program characteristics in addition to premiums, waiting periods, and enrollment caps, the empirical model controls for all unobserved state and secular effects by including dummy variables that represent 31 states and the District of Columbia, the six years covered by the CTS survey data, and 155 state-year interaction variables. 366

6 Near-Poor Children Table 1. Percentage distribution of near-poor (100% to 300% FPL) children by estimated SCHIP/Medicaid eligibility and premium liability, by year Year Not eligible (%) Eligible, no premium (%) Eligible, with premium (%) All years Source: Community Tracking Study Household Surveys, Center for Studying Health System Change. Note: The sample from CTS states excludes Minnesota, Tennessee, and Wisconsin, which charge only family premiums. We adopted this strategy because it is difficult to identify and specify all the relevant policies and structures that might affect actual enrollment and, even if the relevant policies could be identified, it is difficult to measure them accurately, especially over time. Moreover, it is possible that the specific set of policies a state adopts is correlated with its decision to require premiums. If so, it would be very difficult to identify and distinguish the effect of premiums from the effects of other state policies. Finally, there might be other state-specific and/or secular differences beyond Medicaid/SCHIP policies that could affect the pattern of children s coverage. Employing stateyear fixed effects and interactions controlled for these effects as well. Individual and family characteristics. The child s individual and family characteristics serve as indicators of the need and/or preference for the alternative health insurance coverage choices. The individual characteristics are the child s age (younger than 5, 5 to 9 years old, and 10 to 13 years old, relative to 14 to 18 years old), gender, race/ethnicity (African American, white Hispanic, or other race, relative to white non-hispanic), and health status (excellent or very good, or good, relative to fair or poor). Family characteristics include prior year income (measured by a set of dummy variables in $10,000 increments from $15,000 to $45,000 or more), family structure and parents work status (single working parent, two parents with one worker, and two parents with no workers, relative to a single nonworking parent), the number of children in the family, highest year of education for either parent, whether the interview with the family informant was conducted in Spanish, and whether either parent is a strong or moderate risk taker (relative to not being a risk taker). Statistical Estimation and Identification The model was estimated as a multinomial logistic regression with three outcomes: public coverage (the reference coverage), private coverage, or no coverage (uninsured). We treated the insurance premium variables as exogenous instrumental variables, since the SCHIP premium was calculated from state policies, family size, and potential eligibility based on lagged family income; the private insurance premium was imputed from state and county data. Identification of the public premium s effect on coverage derives from a combination of interstate variation in premium rules and intrastate variation in premium levels due to family size, family income, and changes over time in premium amounts. From the broadest perspective, the analysis can be viewed as a comparison between near-poor children who face no public premiums and those who face premiums of varying amounts. However, it is possible that some of these implicit comparisons embody structural differences that go beyond the presence or absence of a public premium. Table 1 shows the distribution of children in the sample by estimated eligibility and premium liability for each year of the study. The percentage of children potentially eligible for public coverage was much lower in 1996 and 1997 prior to SCHIP implementation, and very few potentially eligible children faced premiums for public coverage. During 1998 when SCHIP was being implemented 45% of nearpoor children were eligible for coverage; a third of them faced premiums. However, the effects of premiums during this period may have been confounded by the lack of knowledge about the program in some states. More generally, states that never charged premiums may differ from other states in some fundamental, unobserved way. Finally, children who face no public premium because they are not eligible for public coverage may have a fundamentally different choice from those who are eligible but do not have to pay a premium to enroll. 367

7 Inquiry/Volume 43, Winter 2006/2007 Although the model s state, year, and stateyear dummy variables should control for unobserved differences across states and over time, we explored the sensitivity of the premium coefficients to the implicit identification strategy by estimating the multinomial logit model for five different samples. Our primary sample consisted of potentially eligible children in the years 1999 through 2003, following the SCHIP implementation year of This sample is arguably the most relevant for policy purposes, since the pre- SCHIP environment was very different from the present situation and some states that currently do not charge premiums may do so in the future. The alternative samples for the sensitivity analyses were: all near-poor children pre- and post- SCHIP (i.e., ), all potentially eligible children pre- and post-schip, all near-poor children following SCHIP implementation ( ), and potentially eligible children in states that charged a public premium post-schip. Table 2. Percentage distribution of children by annual public premium, children liable for a premium, Annual public premium (2003 $) Results Premium Liabilities and Sample Characteristics Over time, Medicaid/SCHIP eligibility increased substantially and the percentage of children estimated to be both eligible and liable for a public premium doubled (Table 1). Among eligible children facing a premium, the average annual premium per child (in 2003 dollars) over the years was $119. Table 2 shows the distribution of children charged a premium by annual premium amount. (The median premium was $90.) Almost 75% faced premiums less than $120 per year, while 5.4% paid more than $360 per year (2.9% faced annual premiums of $600 or more). The average annual, private out-of-pocket premium calculated using equation 1 was $3,847 per family over the entire time period (median ¼ $3,231). Twenty-five percent of children faced premiums of $2,722 or less, and 25% faced premiums of $4,370 or more. Between 1999 and 2003, the average private premium increased by 50%, from $3,324 in 1999 to $4,994. Table 3 reports the mean values of the variables used in the regression models for the primary sample of potentially eligible children in the years and the maximum available CTS sample of all near-poor children for the years Overall, 45.7% of the full sample was imputed to be potentially eligible for public coverage. Average annual public premiums were higher in the primary sample than in the full sample because a higher proportion of children in the primary sample faced public premiums. Average private premiums also were higher in the primary sample because private premiums were relatively stable in the late 1990s, due primarily to the effects of managed care, and then started to increase much more rapidly beginning in 2000 in response to changing market conditions and accelerated growth in underlying medical costs (Gabel et al. 2005). Since potential eligibility is strongly related to income, the primary sample had a much higher proportion of families with incomes less than $25,000, and a much smaller proportion with incomes of $45,000 or more. The difference in incomes between the two samples reflects differences in parental work status, parents education, and race, ethnicity, and interview language (Spanish or English). Higher proportions of children in the primary sample were in families with: no working parents or only a single, working parent; parents who either did not finish high school or who did not pursue any post-high school education; and parents who are racial minorities or were interviewed in Spanish. The values of the other variables were fairly similar in the two samples, although a smaller proportion of children in the primary sample were in excellent or very good health. Multinomial Regression Coefficients Percentage of children Less than $ $60 up to $ $120 up to $ $240 up to $ $360 or more 5.4 Source: States SCHIP premium information linked to the Community Tracking Study Household Surveys, Center for Studying Health System Change. Note: The sample from CTS states excludes Minnesota, Tennessee, and Wisconsin, which charge only family premiums. Table 4 shows the multinomial logistic regression coefficients of the model s independent 368

8 Near-Poor Children Table 3. Variable means for full and primary samples Full sample All near-poor children, Primary sample Near-poor children potentially eligible for Medicaid/SCHIP, (Unweighted N) (13,254) (4,418) Potentially eligible (%) 45.7 Public premium ($) a,b Private premium ($) a 3,847 4,371 Waiting period (months) Enrollment cap (%) Family income (%),$15,000 (ref.) $15,000 25, $25,000 35, $35,000 45, $45,000 or more Parental work status (%) Single nonworking parent (ref.) Single working parent Two parents, 11 workers Two nonworking parents Child s age (%) (ref.) Female child (%) Spanish interview (%) Child s race/ethnicity (%) White, non-hispanic (ref.) White, Hispanic African American Other races Child s health (%) Excellent or very good Good Fair or poor (ref.) Parent(s) highest education (%) Less than high school (ref.) High school graduate Some college College graduate Parent(s) risk attitudes (%) Not a risk taker (ref.) Moderate risk taker Strong risk taker Risk attitude missing Number of children in family Source: Community Tracking Study Household Surveys, Center for Studying Health System Change. a 2003 dollars. b The public premium is set to 0 for children not liable for a premium. 369

9 Inquiry/Volume 43, Winter 2006/2007 Table 4. Multinomial logistic regression coefficients ( p-values): primary sample (near-poor children eligible for SCHIP/Medicaid, ) Coefficient ( p-value) Variable Private coverage Uninsured Public premium ($) (.01) (.02) Private premium ($) (,.01) (.21) Waiting period (months).186 (.53) (.12) Enrollment cap (%) (.17).530 (.50) Family income (%),$15,000 (ref.) $15,000 25, (,.01).262 (.37) $25,000 35, (,.01).650 (.05) $35,000 45, (,.01).910 (.02) $45,000 or more (,.01) (.03) Parental work status (%) Single nonworking parent (ref.) Single working parent.721 (.01).872 (.01) Two parents, 11 workers (,.01) (.93) Two nonworking parents.968 (,.01) (,.01) Child s age (%) (,.01) (.02) (,.01) (,.01) (.09) (.44) (ref.) Female child (%) (.99).314 (.05) Spanish interview (%) (.05).283 (.31) Child s race/ethnicity (%) White, non-hispanic (ref.) White, Hispanic (.35) (.85) African American (.37) (.67) Other races (.49).392 (.30) Child s health (%) Excellent or very good.175 (.51).457 (.11) Good (.56).493 (.13) Fair or poor (ref.) Parent(s) highest education (%) Less than high school (ref.) High school graduate.394 (.02) (.39) Some college.736 (,.01) (.15) College graduate (,.01) (.82) Parent(s) risk attitudes (%) Not a risk taker (ref.) Moderate risk taker (.11).132 (.56) Strong risk taker (.01).102 (.67) Risk attitude missing.898 (.01).910 (.01) Number of children in family (.03) (,.01) Source: Community Tracking Study Household Surveys, Center for Studying Health System Change. Note: P-values are in parentheses. variables (excluding the state, year, and stateyear dummy variables, which are available on request) for the primary sample of all eligible children in the years Relative to public coverage, the omitted reference category, the public premium variable had the hypothesized positive effects on the probabilities of being covered by private insurance or of being uninsured, 370

10 Table 5. Sensitivity test comparisons of public and private premium variables coefficients ( p-values), multinomial logistic regressions Sample specification Public premium coefficient ( p), by insurance outcome a Private premium coefficient ( p), by insurance outcome Years Eligible children only CTS states b Sample size Private coverage Uninsured Private coverage Uninsured Yes All 4, (.03) (.02) (,.01) (.21) No All 7, (.01) (.04) (,.01) (.04) Yes Premium 3, (.03) (.04) (,.01) (.26) states No All 13, (,.01) (,.01) (,.01) (.01) Yes All 5, (.02) (.01) (.01) (.14) Source: Community Tracking Study Household Surveys, Center for Studying Health System Change. Note: P-values are in parentheses. a Public coverage is the omitted reference outcome. b Excludes Minnesota, Tennessee, and Wisconsin, which charge only family premiums. Near-Poor Children and its estimated coefficients both were significantly different from zero. A higher public premium reduced the probability of a child being covered by public insurance and increased the probabilities of each of the alternative insurance options. The private premium variable had a statistically significant negative effect on the private coverage option, which means that higher private premiums are associated with a lower probability of private coverage, and higher probabilities of either public coverage or being uninsured. The estimated coefficient on the uninsured option was not statistically significant, but (as shown later) this appears to be a consequence of the smaller sample size when the analysis was limited to potentially eligible children. Marginal probabilities calculated from these coefficients for small changes in premiums indicate that if the probability of private coverage decreases by 1%, the probability of public coverage will increase by.55% and the probability of being uninsured will increase by.45%. An increase in the public premium that leads to a 1% decrease in public coverage has a larger marginal impact on the probability of private coverage, which increases by.62%, while the probability of being uninsured increases by.38%. Other variables in the model indicate that neither a SCHIP waiting period nor a cap on SCHIP enrollment had statistically significant coefficients. 13 The effects of the sociodemographic variables generally were as expected. Increasing family income increased the probabilities of both private coverage and of being uninsured. Since all of these children were estimated to be potentially eligible, the effect on the probability of being uninsured might reflect an increase in parents perception that they were not likely to be eligible as their incomes rose. Similarly, children in families with working parents were more likely to have either private coverage or to be uninsured relative to public coverage, while a child with two nonworking parents was more likely to have private insurance but not more likely to be uninsured than a child with a single nonworking parent. Parents highest level of education was significantly related to the probability of private coverage, but not to the probability of being uninsured. Younger children had a higher probability of public coverage. Gender, health, race, and ethnicity generally were not statistically significant, although girls appeared somewhat more likely to be uninsured, and children whose parent was interviewed in Spanish were significantly less likely to have private coverage. Children with a strong risk-taker parent were significantly less likely to have private insurance coverage. Finally, the probabilities of both private coverage and of being uninsured decreased as the number of children in the family increased. Sensitivity Analysis Table 5 reports the coefficients and p-values of the premium variables from the multinomial logistic regression model estimated with alternative samples of near-poor children. The first row 371

11 Inquiry/Volume 43, Winter 2006/2007 reproduces the values for the primary sample of potentially eligible children surveyed between 1999 and The next two rows show the coefficients for children interviewed in the same time period, first expanding the sample to include all near-poor children, 14 and then restricting the sample to potentially eligible children in states that imposed public premiums. The last two rows expand the sample to include all years in the CTS surveys ( ), first for potentially eligible children and then for all near-poor children. These comparisons suggest that the underlying factors that identify the premium coefficients are robust and not sensitive to sample definition. Both premium variables had the expected signs in all subsamples that is, the public premium consistently had positive effects on the probabilities of having private coverage or being uninsured, and the private premium had a negative effect on the probability of private coverage and a positive effect on the probability of being uninsured. Moreover, the estimated premium coefficients for a particular type of coverage generally were very similar in magnitude, especially when the sample was limited to eligible children in the post-schip years. Other sensitivity tests examined the effects of alternative approaches to constructing the private premium variable. One used the CTS data to estimate a Heckman selection-adjusted nongroup premium model, which then was extrapolated to all children in the sample, plus a logistic model for generating a predicted probability of having an ESI offer in the family. The private premium variable then was constructed as a weighted average of the family ESI contribution (from the MEPS-IC) and the selection-adjusted nongroup premium (from the CTS data), using the predicted probability of having an ESI offer as the weight. The second version used the same premium variables (from the MEPS-IC and the selectionadjusted nongroup CTS premiums), but used the actual value of the ESI offer variable to assign the child the family ESI contribution if the family had an ESI offer and the selection-adjusted nongroup premium if the family did not have an ESI offer. Comparing the original premium coefficients for the primary sample (eligible children in the time period) with these two alternatives indicates that the public premium coefficient is quite robust. 15 However, the private premium coefficients differed substantially across the two alternatives. In the first version, which used predicted values for both the ESI offer probability and the nongroup premium, the private premium coefficients became much smaller in magnitude and did not approach statistical significance. This lack of significance probably reflects a weak instrument problem, since the factors that influence whether a family has a nongroup policy (the Heckman selection equation) are essentially the same as the factors that determine whether the family has an ESI offer. Ideally, one needs separate and independent exogenous variables to identify the nongroup premium and the family ESI offer probability. The second version, which used the actual ESI offer value to assign a private premium to the child s family, produced coefficients that were two to three times larger in magnitude and which were statistically significant. 16 Since having an ESI offer in the family is likely to be endogenous, the increases in magnitude may reflect endogeneity bias. However, it is also possible that the coefficient estimates in the primary model were biased downward because of measurement error. Better data on both ESI and nongroup out-of-pocket premiums and independent exogenous identifying variables for the underlying premium and offer models are needed to obtain a more precise estimate of the private premium s effects. Overall, the sensitivity tests implied a generally robust relationship between public premium increases and changes in the distribution of insurance coverage. The private premium coefficients were less robust, and the coefficients reported in Table 5 may understate their true magnitude. Nevertheless, the two samples of eligible children in the post-schip years (rows 1 and 2) indicated that both public and private premiums had statistically significant and negative own-price effects, and significant and positive cross-price effects. The estimates imply that from 23% to 45% of the children who appear to lose coverage (or do not obtain coverage) because of higher public premiums become (or remain) uninsured. Simulated Effects of Alternative Premium Amounts To further illustrate the magnitudes of changes in the distribution of insurance coverage associated 372

12 Near-Poor Children Table 6. Simulated effects of changing public and private premiums on the distribution of insurance coverage for near-poor children eligible for Medicaid/SCHIP (based on characteristics of potentially eligible children in 2003) Insurance coverage distribution (%) Premium value Public (%) Private (%) Uninsured (%) Baseline distribution Public premium policy change Eliminate all premiums Impose a minimum premium of $120 and increase existing premiums by $ Institute graduated premium increases a Percentage change in private premium 10% % % Source: Community Tracking Study Household Surveys, Center for Studying Health System Change. a $60 for 100% 149% FPL; $120 for 150% 199% FPL; $240 for 200% 249% FPL; $360 for 250% 300% FPL. with alternative premium amounts, we conducted simple simulations of the probabilities of each type of coverage for alternative SCHIP and private premiums. Predicted probabilities were calculated using the coefficients for the primary sample (Table 4) combined with the characteristics of the sample of eligible children in 2003 because their characteristics should be most similar to the current population of Medicaid/SCHIPeligible children. We simulated three alternative public premiums: eliminating all premiums, increasing the annual premium by $120 for all eligible children (including those not currently liable for a premium), and instituting a graduated premium schedule ($60 per year for children between 100% and 149% FPL, $120 per year for children between 150% and 199% FPL, $240 per year for children between 200% and 249% FPL, and $360 per year for all other eligible children up to 300% FPL). The private premium simulations assumed increases of 10%, 20% and 30% above the baseline private premium values, reflecting the fact that the inflation-adjusted cost of ESI coverage increased by about 30% between 2000 and 2003 (Gabel et al. 2005). Table 6 shows the predicted probabilities under the alternative premium amounts and the baseline predicted probabilities (row 1) using the actual values of all variables, including the premiums. 17 Eliminating public premiums, which were relatively small and applied to about half of potentially eligible children in 2003 (Table 1), would increase public coverage by 1.32 percentage points and reduce both private coverage (by.76 percentage points) and uninsurance (by.58 percentage points). In contrast, imposing a minimum premium of $120 and increasing all other premiums by $120 represents a substantial increase in premium liabilities among all eligible children and would reduce public coverage from 37.25% to 34.16%. The percentage uninsured would increase from 11.95% to 13.04%, and the balance of the reduced public coverage would show up as increased private coverage. A graduated public premium policy, which would be more equitable in terms of family burden, has a very similar impact on the distribution of coverage, with public coverage falling to 34.59%, private coverage increasing to 52.49%, and uninsurance increasing to 12.92%. Similarly, increasing private insurance premiums by various amounts would reduce private coverage and increase both public coverage and uninsurance. The 30% increase in the cost of private insurance, which parallels the real increase observed between 2000 and 2003, would reduce private coverage from 50.8% to 43.19% and would increase public coverage by 4.51 percentage points and uninsurance by 3.1 percentage points. Discussion This analysis reinforces and extends earlier studies of the relationship between public insurance 373

13 Inquiry/Volume 43, Winter 2006/2007 premiums and children s insurance coverage. Consistent with prior research, we find that higher premiums for public coverage are associated with a lower probability of near-poor children being covered by public insurance. Extending the prior research, our analysis also finds that higher public premiums are associated with higher probabilities of a near-poor child being uninsured or having private coverage. At the same time, public insurance coverage and uninsurance are significantly related to private insurance premiums. The relatively large increases in the cost of private insurance since 2000 may have contributed to a form of public insurance crowd-in if near-poor families have substituted public insurance for private coverage in response to rapidly increasing private premiums. Our analysis also suggests that declining family incomes contribute to a higher probability of public coverage. Between 2000 and 2003, families in the lowest fifth of the income distribution experienced an 8% decrease in constant-dollar income, and families in the second-lowest fifth of the income distribution had a 4.6% decrease. 18 Thus, decreases in family income due to rising unemployment since 2001 likely contributed to recent increases in SCHIP enrollment, even though some states were increasing their SCHIP premiums. While this study s estimates are generally robust with respect to the years, states, and samples that can be varied within the CTS surveys, the results may be limited by the fact that the 60 CTS communities are drawn from only 35 states. Although the sample is nationally representative and these states account for about 90% of the total U.S. population, the data are not necessarily representative of each state s population. In a parallel study, Kenney, Hadley and Blavin (2006/2007) investigated the effects of public and private insurance premiums on children s insurance coverage using a much larger sample drawn from the Current Population Surveys (CPS), which include all 50 states (plus the District of Columbia) for all years from 1999 through While there are some essential differences between the CPS and CTS data with respect to the measurement of insurance status, family income, and the private premium, comparable models estimated with the two data sets produce qualitatively similar results. Overall, it appears that analyses of two independent data sets drawn from the CTS and CPS have produced broadly similar results. While more research would be useful to increase the precision and narrow the range of estimates across alternative analyses, especially for the effects of private insurance premiums, we believe that our qualitative conclusions are sound. The pattern of nearpoor children s health insurance coverage is sensitive to premiums charged by both public programs and private insurance. Other things equal, states that increase premiums for public coverage should expect fewer children to have public coverage and more children to be uninsured. If private insurance premiums keep growing at a high rate and access to ESI continues to erode, future increases in public premiums are likely to lead to higher uninsurance rates among near-poor children. Moreover, the impact of higher premiums probably will be greater for lower-income children within the near-poor population because their families are less likely to have an ESI offer as an alternative to public coverage and less able to afford premium increases (Kenney, Hadley and Blavin 2006/2007). States seek greater authority and flexibility to increase premiums in their public insurance programs as a way of constraining enrollment and program costs. Our analysis suggests that higher public premiums will reduce enrollment, but they also will increase uninsurance among near-poor children. Although states have less control over private insurance premiums, their budgetary goals would be well served by advocating for national policies to constrain the growth of private insurance premiums. Successful cost containment of private insurance will help slow nearpoor children s enrollment in public insurance programs and reduce uninsurance. Notes The authors are very grateful to Beny Wu of Social and Scientific Systems, Inc., for her excellent computer programming support. 1 One of the 15 states is Wisconsin, which has a waiver from CMS to charge premiums to higher-income families in its Medicaid program. 374

14 Near-Poor Children 2 The five states are Missouri, Rhode Island, Tennessee, Vermont, and Wisconsin. 3 See Artiga and O Malley (2005) for a more detailed summary of these and related analyses of recent experience in states that imposed or increased premiums. 4 Data on ESI premiums are not collected because of low reporting accuracy, presumably due to variations across employers in the methods for deducting or collecting the employee s contribution. Reporting accuracy for nongroup premiums is higher, since the respondent typically pays the full cost, but it is subject to selection bias (Hadley and Reschovsky 2003). 5 The MEPS-IC is an establishment survey with sample sizes ranging from 27,000 in 1996 to 44,000 in 2003 and an average response rate of 78%. (See Sommers 1999, 2004 for details.) 6 Using lagged family income will not completely solve the potential endogeneity problem if family income is correlated over time. Sensitivity analysis will indicate whether the premium coefficients are robust with respect to the eligibility designation by estimating the model for all near-poor children as well as those imputed to be potentially eligible. 7 We used multiple sources because no single source provided sufficient detail for the full time period of our analysis. The multiple sources also provided a means of validating the information obtained. In the case of discrepancies, priority was given to information provided on the state s Web site or the CMS and ASPE Web sites. 8 MEPS-IC data for 2003 were not available at the time of this analysis. Premium values for 2003 were estimated by applying regional estimates of the percentage change in total and out-of-pocket premiums between 2002 and 2003 from Gabel et al. (2005). 9 Although research has shown that the total premium for employer-sponsored coverage is generally less expensive than the premium for nongroup insurance (Pauly and Nichols 2002), we assume that variations in the total premium are a reasonable proxy for variations in the premium for nongroup coverage. Kenney, Hadley and Blavin (2006/ 2007) investigated alternative assumptions that increased the value of the total premium to reflect the generally higher cost of nongroup policies and found that the coefficient estimates were not sensitive to the alternative assumptions. 10 This assumption adds an element of possible endogeneity, since nonworking or self-employed adults have the option of seeking employment with a firm that offers insurance. 11 Starting in 2001, the MEPS-IC reported separate premiums for employee-plus-one coverage and family (three or more people) coverage. We applied the formula for calculating the premium variable separately to these two types of coverage for those years and assigned the value based on employeeplus-one coverage to children in two-person families, (i.e., a single parent with one child). All other children were assigned the premium based on family coverage. Kenney, Hadley and Blavin (2006/ 2007) explored the sensitivity of the results to the change in reporting by the MEPS-IC and found that the coefficient estimates were robust. 12 In general, it is not possible to determine the share of a family premium that is attributable to covering children. Data on ESI premiums from the MEPS-IC show that the ESI premium for covering two people (either employee plus a spouse or employee plus a child) was $6,043 in 2002 compared to $8,469 for family coverage. While this suggests that the cost of covering children is less than the per-person cost for family coverage, there is no clear basis for allocating family premium costs between adults and children. Similarly, Hadley and Reschovsky (2003) found that the marginal cost of covering a child on a nongroup policy was considerably less than the cost of covering an additional adult. 13 Omitting the enrollment cap and waiting period variables does not affect the coefficients of the premium variables. 14 Models estimated with samples of all near-poor children include a dummy variable that indicates potential eligibility and is also interacted with the public premium and other Medicaid program characteristics. In effect, the public premium, waiting period, and enrollment cap apply only to potentially eligible children even when the sample is expanded to all near-poor children. 15 The alternative values are.0012 and.00177, compared to for the effect on private coverage, and and.00175, compared to for the effect on uninsurance. 16 The coefficient for private coverage increases from to.0008, and the coefficient for uninsurance increases from to This means that some children who currently face a relatively high premium are assumed to have a lower premium under the simulation f03ar.html. References American Academy of Pediatrics Summary of Title XXI Programs in States: American Academy of Pediatrics. Elk Village Grove, Ill.: American Academy of Pediatrics. Artiga, S., and M. O Malley Increasing Premiums and Cost Sharing in Medicaid and SCHIP: Recent State Experiences. Washington, D.C.: Kaiser Commission on Medicaid and the Uninsured. Assistant Secretary of Planning and Evaluation, U.S. Department of Health and Human Services State Children s Health Insurance Program 375

15 Inquiry/Volume 43, Winter 2006/2007 Database. Available at: schip2. Accessed July 15, Blumberg, L. J., L. M. Nichols, and J. S. Banthin Worker Decisions to Purchase Health Insurance. International Journal of Health Care Finance and Economics 1(3/4): Centers for Medicare and Medicaid Services (CMS) SCHIP Fact Sheet. Available at: Accessed July 15, Chernew, M. E., K. D. Frick, and C. McLaughlin The Demand for Health Insurance Coverage by Low-Income Workers. Health Services Research 32(4): Cunningham, P., J. Hadley, and J. Reschovsky The Effects of SCHIP on Children s Health Insurance Coverage: Early Evidence from the Community Tracking Study. Medical Care Research and Review 59(4): DeNavas-Walt, C., B. Proctor, and C. Lee Income, Poverty, and Health Insurance Coverage in the United States: Current Population Reports P Washington, D.C.: U. S. Census Bureau. Dubay, L., and G. Kenney The Impact of SCHIP on Insurance Coverage. Unpublished manuscript. Washington, D.C.: The Urban Institute. Fox, H. B., and S. J. Limb SCHIP Programs More Likely to Increase Children s Cost Sharing than Reduce Their Eligibility or Benefits to Control Costs. Fact Sheet 4. Washington, D.C.: Maternal & Child Health Policy Research Center. Gabel, J., G. Claxton, I. Gil, et al Health Benefits in 2005: Premium Increases Slow Down, Coverage Continues to Erode. Health Affairs 24(5): Hadley, J., and J. Holahan The Cost of Care for the Uninsured: What Do We Spend, Who Pays, and What Would Full Coverage Add to Medical Spending? Washington, D.C.: The Henry J. Kaiser Family Foundation. Hadley, J., P. Cunningham, and L. Hargraves Insurance Coverage, the Safety Net, and Racial and Ethnic Differences in Access to Care. Washington, D.C.: Center for Studying Health System Change. Hadley, J., and J. D. Reschovsky Health and the Cost of Nongroup Insurance. Inquiry 40(3): Henry J. Kaiser Family Foundation Employer Health Benefits 2003 Summary of Findings. Washington, D.C.: Henry J. Kaiser Family Foundation. Kappel, S Effects of Medicaid Premiums on Program Enrollment Preliminary Analysis. Joint Fiscal Office, Vermont State Legislature. April 8. Available at: Healthcare/Medicaid%20Premiums.pdf. Accessed July 21, Kenney, G., R. A. Allison, J. Costich, J. Marton, and J. McFeeters. 2006/2007. The Effect of Premium Increases on Enrollment in SCHIP: Findings from Three States. Inquiry 43(4): Kenney, G., J. Hadley, and F. Blavin. 2006/2007. The Effects of Public Premiums on Children s Health Insurance Coverage: Evidence from SCHIP. Inquiry 43(4): Kronebusch, K., and B. Elbel Simplifying Children s Medicaid and SCHIP. Health Affairs 23(3): Ku, L., and T. A. Coughlin Sliding-Scale Premium Health Insurance Programs: Four States Experiences. Inquiry 36(4): Ku, L., and S. Nimalendran Losing Out: States are Cutting 1.2 to 1.6 Million Low-Income People from Medicaid, SCHIP and Other State Health Insurance Programs. Washington, D.C.: Center on Budget and Policy Priorities. Mann, C., and S. Artiga The Impact of Recent Changes in Health Care Coverage for Low-Income People: A First Look at the Research Following Changes in Oregon s Medicaid Program. Washington, D.C.: The Henry J. Kaiser Family Foundation. National Conference of State Legislatures State Children s Health Insurance Program (SCHIP): Cost-Sharing. Washington, D.C.: National Conference of State Legislatures State Children s Health Insurance Program Chartbook. Washington, D.C.: National Conference of State Legislatures. National Governors Association Short-Run Medicaid Reform. Washington, D.C.: National Governors Association. Available at: Files/pdf/0508MEDICAIDREFORM.PDF. Accessed Aug. 29, O Brien, M. J., M. Archdeacon, M. Barrett, et al State Experiences With Cost-Sharing Mechanisms in Children s Health Insurance Expansions. Publication N.385. New York: The Commonwealth Fund. Pauly, M., and L. Nichols The Nongroup Health Insurance Market: Short on Facts, Long on Opinions and Policy Disputes. Health Affairs supplement Web exclusive (Oct. 23): W Ross, D. C., and L. Cox Preserving Recent Progress on Health Coverage for Children and Families: New Tensions Emerge. Washington, D.C.: Henry J. Kaiser Family Foundation Making It Simple: Medicaid for Children and CHIP Income Eligibility Guidelines and Enrollment Procedures. Washington, D.C.: Henry J. Kaiser Family Foundation. Shenkman, E. A., B. Vogel, J. M. Boyett, and R. Naff Disenrollment and Re-enrollment patterns in a SCHIP. Health Care Financing Review 23(3): Sommers, J List Sample Design of the 1996 Medical Expenditure Panel Survey Insurance Component. MEPS Methodology Report No. 6, AHCPR Pub. No Rockville, Md.: Agency for Health Care Policy and Research Enrollment Rates for Employer- Sponsored Health Insurance in the Private Sector 376

16 Near-Poor Children between 1999 and Statistical Brief #49. Rockville, Md.: Agency for Healthcare Research and Quality. Steinberg, D Keeping Kids Enrolled: Continuity of Coverage under SCHIP and Medicaid. Washington, D.C.: National Conference of State Legislatures. Strunk, B. C., and J. Reschovsky Trends in Health Insurance Coverage, Tracking Report No. 10. Washington, D.C.: Center for Studying Health System Change. The Lewin Group Issues in Developing Programs for Uninsured Children: A Resource Book for States. Washington, D.C.: The Lewin Group. Wright, B., M. Carlson, J. Smith, and T. Edlund Impact of Changes to Premiums, Cost- Sharing, and Benefits on Adult Medicaid Beneficiaries: Results from an Ongoing Study of the Oregon Health Plan. New York: The Commonwealth Fund. 377