Abstract

BACKGROUND AND LITERATURE REVIEW

Increasing health insurance coverage for low-income children has been a national health policy focus for the past two decades. Medicaid expansions beginning in the late 1980s and the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996 extended eligibility to millions of children by increasing family income limits and by separating the health insurance benefit from the cash aid programs. The SCHIP, enacted in 1997, further expanded coverage to children with incomes above the Medicaid income eligibility limit but without access to employer sponsored coverage. These programs have made a significant difference: In 1987, 30.8 percent of low-income children were uninsured; by 2002, this figure was 18.6 percent (Selden, Hudson, and Banthin 2004).

As a result of these eligibility expansions, the great majority of uninsured low income children are eligible for Medicaid or SCHIP, up to 95 percent depending on the income level used (Selden, Banthin, and Cohen 1998,Broaddus and Ku 2000). However, early enrollment in SCHIP fell far short of expectations; in the first two program years, states spent just 24 percent of the federal funding available (Kenney, Ulman, and Weil 2000). By 2002, 79 percent of children under 18 eligible for Medicaid were enrolled, and 60 percent of SCHIP eligible children were enrolled. Except for undocumented children, uninsurance for low-income children is now more a problem of participation than of eligibility (Selden Hudson, and Banthin 2004).

Several studies, using focus groups and population-based survey data, have sought to explain why many eligible families do not enroll their children in the Medicaid and SCHIP programs. Unawareness of or confusion about program eligibility were commonly cited as significant barriers to participation; many working parents assumed they and their children were ineligible (Feld, Matlock, and Sandman 1998; Perry et al. 2000; Kenney and Haley 2001; Brown et al. 2002; Kenney, Haley, and Tebay 2003). Another obstacle was lengthy and cumbersome application and enrollment procedures (Medi-Cal Policy Institute 2000; Perry et al. 2000; Ross and Cox 2000; Kenney and Haley 2001). The stigma of publicly sponsored health insurance was an issue for some families (Perry et al. 2000; Cunningham 2001).

A number of parent and family characteristics have been identified as influencing enrollment in public health insurance programs. Children more likely to be enrolled in Medicaid had parents who were younger, less educated, poorer, single, immigrants, had more children or a family member in fair or poor health (Davidoff and Garrett 2001). While nonwhites in California reported more problems navigating the enrollment system than whites (Medi-Cal Policy Institute 2000), another national study found that black or Hispanic ethnicity was positively associated with Medicaid enrollment, although Hispanic ethnicity was also a predictor of uninsurance (Davidoff and Garrett 2001). Language barriers impeded access for parents with limited English proficiency (Medi-Cal Policy Institute 2000; Perry et al. 2000), while bilingual application assisters increased new Medicaid enrollment for California Hispanics and Asians (Aizer 2003).

The effect of immigration status involves several factors. Children of immigrants are more likely to be eligible for Medicaid, because immigrant families tend to be larger and poorer than those of native-born residents. However, eligibility expansions mainly increased coverage for native-born children. Eligible immigrants face higher “transaction costs” of applying, including concerns about the use of benefits adversely affecting a future citizenship application. Also, immigrant parents aware that Medicaid covers acute care costs retrospectively may forego enrolling eligible children (Zimmerman and Fix 1998; Currie 2000; Kaushal and Kaestner 2005).

Low-income children with health problems are more likely to be enrolled in Medicaid or SCHIP, and some parents of healthy children do not perceive a need for insurance (Kenney and Haley 2001; Blumberg O'Connor, and Kenney 2005). Many parents lack knowledge about Medicaid and SCHIP; others do not recognize the benefits of health insurance coverage (Brown et al. 2002).

SCHIP programs based on private insurance models typically include several features that can impede access. “Lockouts” (imposing a minimum period of uninsurance for children with access to employer coverage) and cost sharing are two such features (Halfon, Inkelas, and Newacheck 1999), although other qualitative research suggests that families prefer to pay something toward the cost of coverage and care (Lewin Group 1998). Studies from other states indicate that parents are more likely to enroll their children in health insurance programs for which they are also eligible, and that allowing an entire family to apply at once increases the probability that children will be covered (Ku and Broaddus 2000).

CALIFORNIA'S MEDICAID AND SCHIP PROGRAMS

California has more uninsured residents than any other state; over 1.1 million children were uninsured for all or part of the year in 2003, 11.2 percent of all California children, based on data from the California Health Interview Survey (CHIS). Approximately 55 percent of children who were uninsured at the time of the CHIS interview were eligible for MC or HFs (Brown et al. 2005).

California implemented its HFs program as a separate state program from MC. HF employs a private insurance model, and differs from MC in several important aspects: (1) the HF program requires monthly premiums and co-pays, while all care under MC is free; (2) parents of HF children are not eligible for this program, but parents of MC eligible children may be eligible for MC; (3) HF “locks out” children with employer-sponsored coverage within three months of applying for HF coverage, but there is no waiting period for children applying for MC; and (4) HF applications are processed by a centralized, state-level agency, creating statewide uniformity, while county welfare departments handle much of the MC eligibility determination and enrollment, creating room for county-level variation.

California developed a joint application for MC and HF, but the first version of this document was 27 pages. Program officials and advocacy group representatives labored to reduce it to four pages and to create a single point of entry for processing it. With philanthropic funding, state officials also developed an online application that is now available statewide.

HF was launched in 1998 with a statewide advertising campaign. After 2000, the state redirected funds to conduct more targeted, coordinated and community-based outreach for MC and HF. The HF administering agency funded a training and certification program for application assisters (typically bilingual, bicultural individuals with strong community ties) and, until 2003, paid health insurance brokers and application assisters fees of $50 and $25 for new and renewing children, respectively. Since 2003, the state has greatly reduced outreach funding and also eliminated enrollment payments for almost 2 years, restoring them in 2005.

Beginning in 2000, a number of California counties have developed local expansion programs to enable all children in low- to moderate-income families in a particular county to obtain subsidized health insurance at minimal cost. These local initiatives expand coverage to two groups of children ineligible for MC or HF: those with family incomes slightly above the HF limit of 250 percent of poverty (most upper income limits go up to 300, but one is 400 percent), and undocumented immigrant children. These local initiative products mirror HF, with nominal copayments and a heavily subsidized premium. They also incorporate a vigorous outreach effort to enroll all uninsured low- to moderate-income children in the county in whatever program for which they qualify, involve extensive public–private collaboration and provide strong leadership to integrated outreach and enrollment efforts. When this study took place just three expansion programs were operational. As of March 2006, coalitions in 18 counties had implemented expansion programs with a total enrollment of about 85,000 children; another 12 are planning similar initiatives (Institute for Health Policy Solutions 2006). Two private organizations, California Kids Healthcare Foundation and the Kaiser Permanente health plan, have offered subsidized coverage to a limited number of children statewide since 1992 and 1997, respectively.

CONCEPTUAL FRAMEWORK

Figure 1 is a graphic representation of how family-level demographic factors interact with county-level contextual factors to produce an enrollment outcome. This model draws upon work by Andersen (1995) measuring access to medical care and Donebedian (1980) on medical care quality to create a framework for explaining children's enrollment in MC and HF.

Open in a separate window

Figure 1

Framework for Explaining Children's Enrollment in the Medi-Cal and Healthy Families Programs.

METHODS

RESULTS

Tables 1 and ​and22 compare results from the simple logistic regressions and the hierarchical models. Because these two models generate estimates based on different assumptions, the similar results from both analytic models reinforce the robustness of the study conclusions: The results from a three-level model, which controlled for clustering at the family level, were identical to the results of the two-level model, which did not control for this clustering, suggesting that clustering at the family level did not bias the analysis. However, the survey weights used in the simple logistic regression did prove to be important. The results obtained from the hierarchical model mirror results from an unweighted simple regression model. We conclude that the simple regression model was better than the hierarchical model because it corrected for biases introduced by sampling, nonresponse, and households without telephones, and because the specific issues addressed by the hierarchical model proved to be less important: The variation among counties was explained by county-level variables, and family-level clustering did not affect the results.

Table 1

Comparing Models: A Simple Logistic Regression Model and a Multilevel Model to Predict Children's Enrollment in Medi-Cal

	Simple Logistic Regression Model		Two-Level Logistic Regression Model
Medi–Cal Sample Dependent Variable: Enrollment Status (Yes/No)	OR	P-Value	OR	Significant*
Family-level predisposing factors
Immigration status (reference: citizen parents)
One parent citizen, one parent green card holder	0.64	.17	0.84	No
One parent citizen, one parent noncitizen with no green card	0.49	.08	0.86	No
Both parents green card holders or single green card holder parent	0.51	.02	0.57	Yes
One parent green card holder, one parent noncitizen with no green card	0.26	< 0.01	0.49	Yes
Both parents noncitizen no green card or single noncitizen parent with no green card	0.67	.16	0.84	No
Language of parent (reference: English only)
Other language spoken at home/good English proficiency	1.37	.35
Other language spoken at home/poor or no English proficiency	1.55	.20
Ethnicity (reference: white)
African American (non-Latino)	3.32	.02	2.10	Borderline
Asian/Pacific Islander (non-Latino)	2.67	.05	1.54	Yes
Latino	1.15	.66	1.33	No
American Indian/Alaskan Native (non-Latino)	5.48	< 0.01	1.29	No
Other (non-Latino)	0.50	.15	0.63	No
Household income per capita (Reference: > $20,000)
$10,000 or less per person	0.16	< 0.01	0.20	Yes
$10,001 to $20,000 per person	2.20	.33	1.52	No
Family-level perceived need factors
Age of child (reference: birth to age two)
Age 3–11	0.66	.06	0.63	Yes
Age 12–17	0.54	.02	0.58	Yes
Healthy child	1.57	.05
Family-level enabling/disabling factors
Urban residence			1.55	Yes
County-level enabling/disabling factors
Structural measures
County size (reference: large > 100,000)
Less than 10,000 children (small)	2.97	0.08
10,001–100,000 children (medium)	3.84	< 0.01	1.62	Yes
Other expansion program	2.83	.06
Number of outstations per 1,000 eligible children	0.35	.04	0.47	Yes
Media money per eligible child†			0.97	Yes
Outreach money per eligible child (excluding media)	1.06	.08
Interaction term—outreach money per eligible child × small county	0.93	.04
Interaction term—outreach money per eligible child × medium county	0.93	.06
Process measures
Applications processed per eligible child in county	0.99	.36

Open in a separate window

^*An estimate is deemed significant if it is greater than twice the standard error of the estimate.

^†The state distributed media money by Designated Market Area (DMA) and not by county. A DMA may consist of one county or a grouping of several adjacent counties. Counties that were not assigned to any DMA were grouped together. To estimate the amount of media money spent per eligible child in each county, I used the amount of media money spent per eligible child in the DMA.

Table 2

Comparing Models: A Simple Logistic Regression Model and a Multilevel Model to predict Children's Enrollment in Healthy Families

	Simple Logistic Regression Model		Two-Level Logistic Regression Model
Healthy Families Sample Dependent Variable: Enrollment Status (Yes/No)	OR	P-Value	OR	Significant*
Family-level predisposing factors
Immigration status (reference: citizen parents)
One parent citizen, one parent green card holder	0.62	.09	0.75	No
One parent citizen, one parent noncitizen with no green card	1.97	.21	1.58	No
Both parents green card holders or single green card holder parent	0.82	.47	0.98	No
One parent green card holder one parent noncitizen with no green card	0.58	.21	0.91	No
Both parents noncitizen no green card or single noncitizen parent with no green card	1.02	.96	1.08	No
Language of parent (reference: English only)
Other language spoken at home, speaks good English	1.36	.30
Other language spoken at home, speaks poor or no English	1.50	.23
Ethnicity (reference: non-Latino white)
African American (non-Latino)	3.00	.02	1.01	No
Asian/Pacific Islander (non-Latino)	2.45	.02	2.45	Yes
Latino	0.92	.78	1.28	No
American Indian/Alaskan Native (non-Latino)	0.39	.08	0.52	No
Other (non-Latino)	0.75	.52	1.03	No
Household income per capita (reference: > $20,000)
$10,000 or less per person			0.92	No
$10,001–$20,000 per person			1.80	Yes
Family-level perceived need factors
Age of child (reference: birth to age two)
Age 3–11	0.63	.10	0.79	No
Age 12–17	0.40	< 0.01	0.54	Yes
Family-level enabling/disabling factors
Hours worked per week (reference: 0 hours)
Worked some, less than 20 hours	0.72	.45
Worked between 21 and 40 hours	0.81	.32
Worked more than 40 hours per week	0.61	.05
Number of children in the family			1.15	Yes
Urban residence	0.61	.03	0.69	Yes
County-level enabling/disabling factors
Structural measures
County size (reference: large > 100,000 children)
Less than 10,000 children (small)	7.74	.01
10,001–100,000 children (medium)	2.09	.15
Outstations per 1,000 eligible children	0.94	.91
County expansion program	1.17	.76
At least one expansion program (KPCFK, CK, or a county expansion program)			1.56	Yes
Outreach money per eligible child	1.07	.05
Outreach money per eligible child/small county	0.89	.01
Outreach money per eligible child/medium county	0.94	.16
Media money spent per eligible child in the DMA	0.89	.01	0.96	Yes
Process measures
Applications per 1,000 eligible children in the county	1.02	.06

Open in a separate window

^*An estimate is deemed significant if it is greater than twice the standard error of the estimate.

Tested as a block, family-level predisposing factors are significant predictors of enrollment for both the MC and the HF programs (p ≤ 0.05). Neither recent arrival in the United States nor language is associated with enrollment in either program sample. Immigration status was a strong predictor of enrollment in the MC model (p ≤ 0.05), but not in the HF model. Children of noncitizen parents were less likely to be enrolled in MC than children with two citizen parents. However, it was not the children of undocumented parents (or “noncitizens without green cards”) who had the lowest program participation, but rather the children of parents with permanent legal residency. Ethnicity was a strong predictor of enrollment in both the MC and the HF models (p ≤ 0.05). Whites and Latinos were much less likely to enroll in MC than African Americans, Asians, American Indians, and Alaskan Natives (AIANs). Among HF eligible children, Asians and African Americans had higher odds of being enrolled compared with whites, and Latinos had similar odds. AIANs had only 40 percent the odds of whites, a reversal of what was found in the MC program (p = 0.07).

When tested as a block, perceived need factors were significant predictors of enrollment in both programs (p ≤ 0.05). A child's age was a strong predictor of enrollment in the HF program. Health status, a potentially endogenous variable, was positively associated with enrollment status in the MC model, but not in the HF model. Healthy children had 50 percent better odds of being enrolled in MC compared with those reported as being in fair or poor health (p ≤ 0.05).

Taken together, family-level enabling/disabling factors were significant predictors of enrollment in both the MC and HF programs, although individual variables were significant only in the HF models. Family-level enabling/disabling factors measured in this study include hours worked per week, emotional well-being, the number of children in the household, marital status, and rural versus urban residence. While children with working parents had lower odds of enrollment compared to those whose parents did not work, only those whose parents worked more than 40 hours per week had significantly lower odds (p = 0.05). Children living in urban settings had over 40 percent lower odds of enrollment in HF than children living in nonurban settings (p ≤ 0.05).

When tested as a group, county-level factors were significant predictors of MC enrollment (p ≤ 0.05), but were not significant for the HF program. However, individual variables were significant for both programs. Media money spent per eligible child in the Designated Market Area (DMA) was a significant predictor of HF enrollment; however, the direction of the effect was contrary to expectations. For every additional dollar per eligible child in the DMA that was spent on media, the odds of enrollment actually decreased among the HF sample by nine percentage points (p ≤ 0.05). The amount of money spent on outreach (excluding the media campaign) was associated with greater enrollment in both the MC and HF programs. The odds of being enrolled in MC increased by 6 percentage points for each additional dollar spent on outreach per eligible child in the county (p = 0.08), and for HF eligible children the odds increased by seven percentage points (p ≤ 0.05).

Children who lived in counties with a county-sponsored expansion program had almost three times the odds of being enrolled in MC compared with children from counties without such a program in place (p = 0.06). The private KPCK and CK programs were not associated with increased MC enrollment. For HF, county-sponsored expansion programs, KPCK and CK were all positively associated with enrollment, but these results were not statistically significant. The application rate was associated with enrollment in the HF program only, with an increase of two percentage points in the odds of enrollment for every additional application per 1,000 eligible children in the county (p = 0.06). This variable specifically applies to joint applications received for the HF program and was only a proxy measure of the number of applications received by the MC program. The number of outstations per 1,000 eligible children in the county was negatively associated with enrollment, and the presence of evening, weekend, or walk-in hours at outstations was not associated with enrollment in either program.

DISCUSSION

The theoretical framework for predicting enrollment in the MC and HF programs set forth in this study is a useful paradigm for studying program enrollment. Most of the variation in enrollment status occurs at the family level, but both family- and county-level factors are important predictors of enrollment in both the MC and the HF programs.

While study findings do not establish causality, they suggest relationships that may be explained in context. For example, the lack of association between language and enrollment in the regression model suggests that state efforts to target linguistic minorities with outreach, including the translation of enrollment materials into 10 different languages, have been working. Children of undocumented immigrants and green card holders are less likely to enroll, likely reflecting concerns about the use of benefits prejudicing a future citizenship application (Zimmerman and Fix 1998). Lower participation by AIANs in the HF program when this group's MC participation was high may be partially explained by the failure of California to implement a federal policy to exempt AIANs from paying HF premiums (Satter et al. 2002).

Parents who work full time were less likely to enroll their child compared with nonworking parents, perhaps reflecting that working parents have less time to enroll their children. Children from large counties were less likely to enroll, suggesting that large county bureaucracies may be less responsive and connected to the communities they serve compared with smaller agencies, and thus less able to foster program enrollment.

Dollars spent on the statewide media campaign had a negative association with HF enrollment, although the media campaign may have generated greater awareness of the HF program. Initially, the HF program featured a burdensome application process, and parents experienced problems with enrollment. These experiences may have fostered negative reports in the community about the program and discouraged participation. Increased product awareness can interact with negative information, resulting in greater dissatisfaction compared with areas with lower product awareness (Gaskell et al. 1999). This same relationship was not seen with MC, which was already well known.

OEWs were associated with lower enrollment. Outstations are resource intensive and OEWs process fewer applications than their counterparts stationed in welfare offices (Blackburn and Happoldt 1999). Also, many safety net hospitals use OEWs to offset uncompensated care, channeling high-cost uninsured patients to the worker to apply for emergency MC coverage. These patients are less likely to be children (Vilander 2004). Outstations are partially funded by MC and HF outreach subventions from the state to counties, and take funding away from other outreach efforts that may be more effective with children. Further research is needed to establish whether certain types of outstations are more effective for reaching children than others. For example, hospital outstations may reach adults but few children. However, school-based outstations may be effective with children and warrant the high resource expenditure associated with outstationing.

Residency in a county with a county-sponsored expansion programs was strongly associated with enrollment in MC but not HF, although local initiatives aim to increase enrollment in both programs. The positive association with MC enrollment may be due to the vigorous implementation of outreach and enrollment activities when local public–private coalitions and their leaders become invested in achieving strong results. The reason that a similar association was not seen for HFs may be due to the difficulties that local initiatives report in following up with HF applications because these applications are processed by state contractors who observe strict confidentiality. In contrast, local initiatives collaborate with county agencies that determine MC eligibility (Alameda County Health Services Agency 2002). Research published subsequent to this study found that the Santa Clara County Children's Health Initiative led to large enrollment gains in both MC and HF (Trenton and Orzol 2004). Further comparative research examining changes in children's MC and HF program enrollment over time in counties with and without local initiatives is needed to assess the broader impact of these initiatives.

CONCLUSION AND POLICY IMPLICATIONS

Progress has been made in enrolling eligible children in public programs, and further efforts are needed to broaden coverage. Although the design of this study does not identify causal relationships, findings from it suggest that to better target eligible but uninsured children we can: (1) target linguistic minorities and immigrants, especially Latinos; (2) reassure immigrants that use of MC or HF is not considered being a public charge; (3) communicate culturally relevant messages about the benefits of having health insurance; (4) further simplify enrollment; (5) fund outreach by community-based organizations and agencies; and (6) expand eligibility to include more children or support county-based expansion programs that do so.

Acknowledgments

REFERENCES