Children’s Insurance Coverage and Crowd-Out Through the Recession: Lessons From Ohio

Abstract

Objectives. We estimated changes in children’s insurance status (publicly insured, privately insured, or uninsured) and crowd-out rates during the 2007 to 2009 US recession in Ohio.

Methods. We conducted an estimate of insurance coverage from statewide, randomized telephone surveys in 2004, 2008, 2010, and 2012. We estimated crowd-out by using regression discontinuity.

Results. From 2004 to 2012, private insurance rates dropped from 67% to 55% and public rates grew from 28% to 40%, with no change in the uninsured rate for children. Despite a 12.0% decline in private coverage and a corresponding 12.6% increase in public coverage, we found no evidence that crowd-out increased during this period.

Conclusions. Children, particularly those with household incomes lower than 400% of the federal poverty level, were enrolled increasingly in public insurance rather than private coverage. Near the Medicaid eligibility threshold, this is not from an increase in crowd-out. An alternative explanation for the increase in public coverage would be the decline in incomes for households with children.

During the economic recession of 2007 to 2009, 8 million Americans lost employer-sponsored health insurance.¹ During this same period, the national rate of uninsurance among adults increased by 15% while the rate for children decreased by 7%, as children increasingly moved from private insurance to public programs including Medicaid and the Children’s Health Insurance Program (CHIP).¹ With spending for Medicaid accounting for almost 20% of state budgets,² increases in Medicaid enrollment raise concerns for many state governments. A key debate among state legislators is whether these changes represent Medicaid fulfilling its planned role as a safety net or whether these increases represent Medicaid replacing or crowding-out private insurance coverage.

Crowd-out refers to individuals who are insured by a public program but who otherwise would have private insurance if the public program did not exist. The archetype of crowd-out is when privately insured individuals gain eligibility for Medicaid and drop their private coverage to enroll in the public plan. This scenario is often referred to as public–private “substitution” in the literature.³ It should be noted that individuals who can no longer afford private individual insurance or who lose access to, or cannot afford their employer-sponsored plan, are excluded from crowd-out estimates. For example, a child whose parent has private insurance but cannot afford to include the child on the plan⁴ would not be considered to be crowded out. The second type of crowd-out, which we refer to as continuation crowd-out, occurs when an individual on a public program becomes eligible to enroll in an employer-sponsored private plan, but chooses to remain on public coverage. This scenario could occur when an unemployed parent begins a new job that offers an affordable, employer-sponsored plan, but the parent decides to keep their child enrolled in Medicaid instead of the newly available private option.

The published literature primarily focuses on estimates of crowd-out based on increased enrollment following the expansion of eligibility, such as a state increasing Medicaid income eligibility limits. These estimates vary widely, ranging from high estimates of 50% (i.e., half of individuals gaining Medicaid coverage through an expansion would otherwise have private insurance) to other authors finding near zero crowd-out.^5–7 Some of this variability is driven by the sensitivity of the econometric models used⁸ and some may be attributable to the actual crowd-out that occurred with different expansions of eligibility in different states.⁵ A smaller literature directly measures substitution from survey data, finding low levels of this type of crowd-out.^9,10 From a legislative perspective, these crowd-out estimates reveal the budgetary cost of Medicaid expansions. For the average state in 2012, insuring a child through Medicaid cost $2700 per year.¹¹ At a 50% crowd-out rate, a state would need to budget $5400 to reduce the number of uninsured by 1 child. The $5400 would include coverage for the previously uninsured child and for a second child who previously had private insurance (1 uninsured child and 1 case of crowd-out).

The existing crowd-out literature implicitly assumes that crowd-out estimates are stable over time. The econometric approaches used in most studies require a change in Medicaid eligibility to estimate crowd-out, producing a single, national estimate for the policy change.^5,12 Absent a more recent change, policymakers assume that crowd-out rates do not change with time because those estimates are not time dependent. This implicit assumption, though, is likely invalid. Crowd-out indicates the use of public insurance while private coverage is still available; the reasons for that are likely dependent on the current cost and expected future cost of insurance, the suitability of access provided by the types of coverage, and noneconomic factors such as the stigma associated with public coverage.¹³

Each of these factors can change over time. Concern about the future cost of insurance during the recession may have been particularly important, as parents may have had strong concerns about retaining their employment or concerns that their employer would stop offering employer-sponsored health insurance. Between January 2007 and January 2010, Ohio’s unemployment rate almost doubled from 5.4% to 10.6%.¹⁴ This increase in the unemployment rate may have raised parents’ concerns about future access to health care for their children. If these concerns led to increased enrollment in public insurance, then crowd-out would increase. Previous work has not estimated state-level crowd-out levels over time. We evaluated how many children in Ohio moved from private health insurance to public health insurance and the degree to which those children were crowded out between 2004 and 2012. We estimated total crowd-out (substitution plus continuation) over time to see whether crowd-out levels changed during the recession in Ohio.

METHODS

To estimate changes in insurance coverage and crowd-out through the recession, we analyzed the 2004, 2008, and 2010 Ohio Family Health Survey (OFHS) and its most recent iteration, the 2012 Ohio Medicaid Assessment Survey (OMAS), conducted by the Ohio State University Government Resource Center. These surveys examine Ohio residents’ use of health services, and their insurance and health status, and capture other determinants of population health.¹⁵ The OMAS/OFHS is a computer-assisted telephone survey that produces representative statewide estimates for Ohio as a whole and its 4 major regions. Beginning in 2008, the survey added a second sampling frame to use random digit dialing to call cellular phones, resulting in a single stratum of cellular phone surveys for the state. This dual-frame methodology (landline and cellular phone) ensures a more complete survey of Ohio’s population, including groups such as younger adults and lower-income families that are more likely to have only a cellular phone and no landline. Response rates ranged from a high of 40% in 2004 to a low of 30% in 2012. Data validation occurred by recalling a portion of surveys to confirm important variables. Survey collection was provided by outside vendors, which varied by year. Sample sizes for each survey year can be found in Table 1.

TABLE 1—

Ohio Children’s Insurance Rates and Estimated Crowd-Out, 2004–2012

Variable	2004, No., % (95% CI), or Median ±SE	2008, No., % (95% CI), or Median ±SE	2010, No., % (95% CI), or Median ±SE	2012, No., % (95% CI), or Median ±SE
No. of children in study with insurance information	15 447	13 443	2 002	5 515
Weighted no. of children	2 899 134	2 754 928	2 751 434	2 898 984
Percentage of children estimated to be crowded out near 200% FPL	8.07 (1.5, 14.6)	10.03 (2.9, 7.1)	5.64 (–5.4, 16.7)	5.55 (–0.9, 12.0)
Percentage with private insurance	66.8 (65.8, 67.8)	62.5 (61.3, 63.6)	57.7 (54.9, 60.4)	54.8 (53.1, 56.4)
Percentage with public insurance	27.8 (26.9, 28.8)	33.5 (32.4, 34.6)	37.7 (35.0, 40.5)	40.4 (38.8, 42.1)
Percentage uninsured	5.4 (4.9, 5.8)	4.0 (3.6, 4.5)	4.6 (3.4, 5.8)	4.8 (4.2, 5.5)
Median child’s federal poverty level	239 ±2.5	236 ±3.25	194 ±9.25	217 ±5.5
Median child’s household income in thousands of dollars	46 ±1	50 ±0.5	47 ±1.5	45 ±1.5

Note. CI = confidence interval; FPL = federal poverty level.

Source. Authors’ analysis of 2004, 2008, 2010 Ohio Family Health Survey, 2012 Ohio Medicaid Assessment Survey.¹⁵

The key outcome variable was children's insurance status. The OMAS/OFHS produces cross-sectional estimates of insurance status (the survey asks whether each child had health insurance the week before the call). We broke down insurance status into 3 categories—public, private, and uninsured—based on the child’s primary insurance. Public insurance includes Medicaid and Medicare. Private insurance captures all types of private health insurance, including employer-sponsored coverage and directly purchased insurance. Among children, the most common reason for gaining public coverage is because household income is below a certain threshold, but children at higher incomes may also gain public coverage because of a disability or because they have a certain disease (such as chronic renal disease requiring a transplant or dialysis, or amyotrophic lateral sclerosis).¹⁶

We used a regression discontinuity approach to estimate crowd-out; a complete description of the technique can be found in previously published research on crowd-out.^12,17–19 Briefly, regression discontinuity is an approach that compares the effect of an externally determined policy or program by comparing similar populations, which are similar in all ways except whether the policy applies to them. In this case, we compared the percentage of Ohio children who had private insurance and were barely eligible for Medicaid (household incomes that are just over 200% of the federal poverty level [FPL]²⁰) to children who are barely ineligible for Medicaid (household incomes that are just under 200% of the FPL). The survey data did not allow us to account for income disregards, so some children not meeting the income eligibility threshold still may be enrolled in Medicaid. If Medicaid has no effect on the rate of private insurance, then we expect the children just above and just below the poverty level to have approximately the same rate of private insurance. If the children who are above the poverty level have higher rates of private insurance, we interpret the difference to be caused by Medicaid eligibility.

The regression discontinuity approach produces estimates of crowd-out independent of any expansion of the Medicaid program, capturing both substitution and continuation crowd-out. In addition, as long as the individuals on either side of the threshold are approximately equivalent other than being eligible for Medicaid, the results of this can be treated as a random experiment.²¹

In this analysis we first estimated the percentage of children who had private insurance at specific income levels (we used blocks of 10% of the FPL; i.e., household income of > 100% to ≤ 110% of FPL, >110% to ≤ 120% of FPL, etc.) and then regressed the percentage of private insurance on either side of the eligibility threshold for Medicaid. The estimated crowd-out is the percentage of children near the eligibility threshold who have private insurance and are ineligible for Medicaid minus the percentage of children with private insurance near the eligibility threshold who are eligible for Medicaid. For this analysis, we focused on children from 50% to 350% of the FPL. Figure 1 shows a graphical representation of this, with children left of the vertical line being eligible for Medicaid and those to the right being ineligible. The difference between the intersections of the 2 regressed lines represents the crowd-out estimate.

FIGURE 1—

Estimated crowd-out levels in Ohio near 200% of the federal poverty level for (a) 2004, (b) 2008, (c) 2010, and (d) 2012.

Note. FPL = federal poverty level. This contains estimates of the percentage of children who have private insurance broken down by FPL. Each point represents a 10% bucket of poverty levels (e.g., > 100% to ≤ 110%, >110% to ≤ 120%). The vertical line in the middle represents the Medicaid eligibility threshold at 200% of the poverty level. The sloped lines are linear fits of the points on either side of the threshold. The estimated crowd-out at the eligibility threshold is the distance between the lines where they meet at the eligibility threshold.

Source. Authors’ analysis of 2004, 2008, 2010 Ohio Family Health Survey, 2012 Ohio Medicaid Assessment Survey.¹⁵

The model for our estimates is as follows:

The dependent variable, Y, is the estimated percentage of children at any Poverty Level with private insurance. Ineligible is a dummy variable where 1 indicates that the child is ineligible for Medicaid on the basis of family income, and 0 indicates that the child is eligible. By centering the data on the eligibility threshold, b0 is the intercept for the regressed line left of the eligibility threshold and is the estimated percentage of children at the eligibility threshold who have private insurance and are eligible for Medicaid. b0 + b2 is the intercept for the regressed line right of the eligibility threshold and is the estimated percentage of children at the eligibility threshold who have private insurance and are not eligible (ineligible) for Medicaid. In this model, b2, which represents the difference between the intercept for those ineligible for Medicaid and eligible for Medicaid, is the estimated effect of being eligible for Medicaid or, in other words, the estimated crowd-out at the eligibility threshold. b1 is the slope of the line left of the eligibility threshold and b1 + b3 is the slope of the line to the right of the threshold.

In this model, we used a localized linear regression to estimate crowd-out. It is localized because we focused only on children from 50% to 350% of the FPL, which is a meaningful range of children with varying amounts of private insurance; those below 50% FPL have consistently low rates of private insurance and those above 350% have consistently high rates of private insurance. We used a linear model because of its relative simplicity and because of the small number of data points on either side of the eligibility threshold. We did fit the data with polynomial regressions that produced slightly lower, but not significantly different, estimates of crowd-out across all years.

A potential endogeneity threat arises if people adjust their incomes to qualify their children for Medicaid. If true, we would see a population shift from those ineligible for Medicaid to those eligible for Medicaid, indicating that people artificially alter their income to help their children become eligible for Medicaid.²² By looking at the total population of children and by using 5% block sizes of the FPL, we found no evidence of a population shift for any of the years of the survey (at P < .2). The smaller block size was possible because we were able to use the entire sample size, rather than just the proportion with private insurance.

As a second means to observe whether incomes may have been artificially modified to meet eligibility requirements, we ran a sensitivity analysis based on the assumption that families closest to the eligibility threshold (i.e., those nearly eligible for Medicaid) would be most likely to artificially change their income. We excluded the children in families with incomes from 190% to 210% of the FPL of the eligibility threshold (those close enough to the eligibility threshold that they may realistically be willing to modify their incomes to become eligible for Medicaid) but found no significant difference in crowd-out estimates.

An important limitation to this model is that regression discontinuity produces “local” rather than global crowd-out estimates. Because the regression is around the eligibility threshold, the estimated crowd-out is localized to this population. Therefore, we were able to estimate the percentage of children who are crowded out near 200% of the FPL (income eligibility for children has remained unchanged at 200% of FPL since 2000¹⁵). However, we were unable to estimate the total percentage of children who were crowded out because it is unlikely that crowd-out levels are the same at every income level. Our estimates of crowd-out, then, can be interpreted as an estimate of the percentage of children in the state of Ohio with family incomes near 200% of the FPL who have public health insurance but, if public insurance were unavailable, would have private health insurance.

In addition, we have created population density charts and stacked bar charts of insurance status for each of the years (Figures 2 and 3) to indicate how the population of children shifted during the study period. Population density charts show the total number of children in Ohio that fall within certain income bands; the different colors refer to different types of insurance under which the children are covered. The stacked bar charts show the percentage of children with different types of health insurance by income level. These charts reveal how the total population of children has shifted over time, as well as how insurance levels have changed.

FIGURE 2—

Population density of children in Ohio by household income for (a) 2004, (b) 2008, (c) 2010, and (d) 2012.

Note. FPL = federal poverty level. This contains estimates of the total number of children in Ohio who are in different FPLs. Each bar represents a 20% bucket of poverty levels (e.g., > 0% to ≤ 20%, > 20% to ≤ 40%) and within each bar the population is broken down into the types of health insurance that each child has. Over time the overall population of children has shifted toward the left (lower incomes) and within population categories the rate of insurance has shifted toward public insurance.

Source. Authors’ analysis of 2004, 2008, 2010 Ohio Family Health Survey, 2012 Ohio Medicaid Assessment Survey.¹⁵

FIGURE 3—

Children’s insurance coverage status by household income in Ohio for (a) 2004, (b) 2008, (c) 2010, and (d) 2012.

Note. FPL = federal poverty level. This contains estimates of the type of insurance that children in Ohio have who are in different FPLs. Each bar represents a 20% bucket of poverty levels (e.g., > 0% to ≤ 20%, > 20% to ≤ 40%). Over time the overall population of children has shifted toward public insurance.

Source. Authors’ analysis of 2004, 2008, 2010 Ohio Family Health Survey, 2012 Ohio Medicaid Assessment Survey.¹⁵

We performed all population estimates in Stata version 13.1 (StataCorp LP, College Station, TX) by using the SVY command that accounts for the complex sampling design of the surveys. We obtained median estimates by using the EPCTILE command that estimates standard errors by using complex survey designs.^23,24

RESULTS

Table 1 presents the regression discontinuity estimates of crowd-out and source of insurance by year. Private health insurance coverage declined between 2004 and 2012 while public coverage for children increased. In 2004, 66.8% of Ohio children were covered by private insurance plans. By 2012, private coverage had declined to 54.8% of children (statistically different at P = .01). Over the same period, the share of children covered by public insurance increased from 27.8% to 40.4% (P = .01). The uninsured rate for children showed no statistically significant difference (5.4% in 2004 and 4.8% in 2012; P = .18). In 2012, the rate of uninsurance for Ohio children was below the national average for children (6.6%).²⁵ Even though private coverage declined by 12.0% and public insurance increased by a similar 12.6%, these trends cannot be interpreted as public insurance crowding out private coverage.

Although the 12% decrease in private insurance was offset by a similar increase in public coverage, the regression discontinuity estimates find no evidence for an increase in crowd-out. Table 1 contains the numeric estimates and Figure 1 has a graphical representation of the crowd-out estimates. The regression discontinuity models estimate the percentage of children whose household incomes were near the eligibility threshold who were crowded out of private insurance by public plans for each of the years. When we compared with another study, which used the American Community Survey, the 2012 estimate of 5.55% is very close to that study’s 2012 estimate of 5.23%.¹⁹ Although the trend from 2004 to 2012 suggests that local crowd-out may have declined modestly from 8.07% to 5.55%, the confidence intervals show no statistically significant change over time.

Declining private coverage with increasing public insurance is not always evidence of crowd-out. Medicaid serves as an important safety net for children and families with falling incomes, especially during recessions. A key component of the increase in public coverage may be explained by examining the Medicaid eligibility and incomes for households with children over this period. From 2004 to 2012, the median child’s household income in Ohio dropped from 239% of the FPL to 217%. Over the same period, the proportion of children eligible for Medicaid (200% FPL) increased from 36% of children to 40%.

Figure 2 contains population density charts of Ohio’s children as a function of household income for each survey year. There is a noticeable shift in the population density of children toward the left (lower incomes), indicating that more children live in households with lower incomes. We adjusted for this shift in population toward lower incomes by adding an individual income covariate, and still found an average yearly decrease in private insurance rates among Ohio’s children of 1.3% (P = .01) and a concomitant increase in public insurance of 1.4% (P = .01). There is also a very small decrease in rates of uninsurance of 0.1% (P = .034).

This shift of children from private to public plans is not uniform across all household income levels. Figure 3 contains area charts of the percentage of children with different types of insurance as a function of their household income for each of the survey years. Above 400% FPL, there is little change in insurance types. Below 400% FPL, the area representing children with public insurance expands over time, indicating that, at each of the FPLs below 400%, there are fewer privately insured children at the end of the study period.

DISCUSSION

Despite a 12.0% decline in private coverage for children and a corresponding 12.6% increase in public coverage, we find no evidence that crowd-out increased during the recent recession. These findings indicate that crowd-out, at least near the eligibility threshold, should not be a primary concern for policymakers. Not only did the crowd-out point estimate decrease in absolute terms, but it also decreased as a percentage of publicly insured children, meaning that fewer total children were estimated to be crowded out. Although there was a small increase in crowd-out from 2004 to 2008, the increase (2%) was much less than the decrease of children on private insurance (9%). During this 1-time increase, only 22% of children who moved to public insurance could have been classified as being crowded out, substantially less than previous estimates^5,8,26

Often overlooked in the crowd-out debate is the fact that Medicaid serves as a safety net for children and families facing falling incomes. One of the original crowd-out studies is repeatedly cited for the finding that 50% of children enrolling in Medicaid after a coverage expansion would otherwise have had or were crowded out of private coverage²⁷ (subsequent studies have reported a wide range of crowd-out estimates^4–6). Often overlooked but also important is the authors’ other finding that crowd-out explained only 15% of the decline in private insurance from 1987 to 1992. Similarly, our study found no evidence that crowd-out is a major contributor to the decline in private insurance from 2004 to 2012.

During the recent recession, households in Ohio faced declining incomes resulting in more children becoming eligible for Medicaid as is seen by the population shift in Figure 2. With more children in lower-income households, there was likely a greater need for public insurance as a safety net. Crowd-out is not the only explanation for the increase in children enrolling in public insurance. An alternative explanation would be an increased need for public insurance because of the recession.

Limitations

This research presents several limitations. First, the survey sample size varied across years. During the recession, funding for projects such as this one was cut throughout Ohio. Because of financial considerations, the sample size was decreased. Subsequently, fewer child interviews were conducted in 2010 and 2012 (2002 and 5515 children) compared with 15 447 and 13 443 in 2004 and 2008. The surveys were weighted to allow for state-level estimates, but the smaller sample size does lead to decreased precision and increased standard errors for these years’ estimates. In addition, the survey response rate was only 30% to 40%. With a low response rate, there is the possibility that there is some selection bias of those who responded compared with those who did not.

In addition, our crowd-out estimates only examine children near the eligibility threshold. There is no guarantee that children at other poverty levels will be crowded out at the same rate. We hypothesized that there would be a similar relative trend at other eligibility levels (a relative decrease in crowd-out over time), but further work will be needed to test that empirically.

Another important limitation of this study is its generalizability to other states. Although most states suffered a similar economic downturn during the years in this study, baseline uninsurance rates do vary by state. Crowd-out levels also vary depending on individual state policies, structure of Medicaid and CHIP programs, and barriers to enrollment.^13,27 The general trend nationwide is toward more children being covered by public programs, similar to Ohio.²⁸ We believe that our core finding that the movement of children toward public programs being driven by changing economics as opposed to increased crowd-out is likely generalizable across the United States.

Conclusions

From the period of 2004 through 2012, the uninsurance rates for children in Ohio remained stable with approximately 1 out of 20 children lacking health insurance. During this time, children increasingly moved from private to public insurance, but the percentage of children who are estimated to be crowded out of private insurance did not increase near the Medicaid eligibility threshold. An alternative explanation of the broad movement of children toward public insurance would be declining household incomes during the recession.