Are Coresidence and Nursing Homes Substitutes? Evidence from Medicaid Spend-Down Provisions

Abstract

This paper measures the extent to which the price of nursing home care affects a potential substitute living arrangement: coresidence with adult children. Exploiting variation in state Medicaid income “spend-down” provisions over time, I find that living in a state with a spend-down provision decreases the prevalence of coresidence with adult children by 1 to 4 percentage points for single elderly individuals, with a corresponding increase in the use of nursing home care. These findings suggest that changes in Medicaid eligibility for long-term care benefits could have large impacts on living arrangements, care utilization patterns, and Medicaid expenditures.

1 Introduction

Long-term care expenditures in the United States are significant and growing. In 2013, formal care expenditures totaled over $300 billion, or over 10% of all health expenditures for all ages. As the population ages and disabling health conditions such as Alzheimer’s and obesity become increasingly common, these costs may rise dramatically.

Very few individuals have coverage for these costs: Medicare does not cover most long-term care expenses, and the private insurance market for long-term care is small and declining. Almost two-thirds of total costs are paid by the Medicaid program, and individuals who are not eligible for Medicaid predominantly pay these costs – which reach upwards of $100,000 a year for a nursing home – out-of-pocket. As a result, long-term care is one of the largest financial risks facing elderly individuals and their families.

Many individuals who need care, however, receive it outside of a nursing home. Individuals generally desire to remain in the community or even their own home, and view nursing homes as a last resort: of a sample of over 9,000 seriously ill hospitalized adults in the early 1990’s, Mattimore et al. (1997) report that 30% would “rather die” than live permanently in a nursing home. Moreover, in-home care, which can either be provided by a paid professional or provided informally by a family member (often through coresidence), may offer a more flexible level of care than an institutional setting. Nursing homes, on the other hand, offer a package of services that may be excessive for many individuals, yet suitable (or even necessary) for those who require more intensive care. In addition, the oftentimes prohibitive cost of nursing homes may push individuals to seek out care through alternative means. This price sensitivity is the focus of this paper.

The extent to which individuals are sensitive to the price of nursing home care has important implications for both individual welfare and policy. If individuals hold strong preferences for where they live and the setting in which they receive long-term care, policies that influence the relative prices of long-term care could significantly affect individual well-being. Furthermore, the large role of Medicaid in financing long-term care implies that changes to long-term care policies and utilization trends may have a large effect on federal and state Medicaid budgets. As the population ages, policymakers have raised concerns about the affordability of current Medicaid policy and indicated interest in potential solutions. For example, proposals to encourage home- and community-based services – including informal care – have gained traction in the recent decade (e.g. Medicaid’s Cash and Counseling and Money Follows the Person programs).

This paper uses price variation generated by state Medicaid eligibility differences to examine the degree to which individuals substitute between nursing home residency and a potential substitute living arrangement: coresidence with adult children. Specifically, I follow the identification strategy used in Grabowski and Gruber (2007) that leverages changes over time in optional “spend-down” policies which allow individuals to deduct health expenses from income to qualify for Medicaid. Using data on single elderly from the decennial Census and the American Community Survey (ACS) from 1980 through 2009, I test whether the presence of Medicaid spend-down policies affects living arrangements among the elderly. I find that the presence of a Medicaid spend-down provision decreases coresidence with an adult child and increases nursing home use both by 1 to 4 percentage points for individuals aged 80 and over. Most of this effect is driven by individuals who report difficulty caring independently for oneself and those with lower income. I find no effect on individuals aged 65 to 79. These findings are consistent with the fact that most elderly individuals need long-term care towards the end of their life, and the relevant margin for this Medicaid policy are those with low assets but higher income than traditional Medicaid income cut-offs would permit.

I supplement this analysis with data from the Health and Retirement Study to examine other important outcomes that are not available in the Census/ACS over the time period. First, I check the first stage of the econometric design: I verify that individuals in states with Medicaid spend-down policies are more likely to be enrolled in Medicaid, particularly if they are in poor health. Second, I show that spend-down policies reduce the likelihood of owning a private long-term care insurance policy, particularly among individuals with children. This is consistent with Brown and Finkelstein (2008), who find that Medicaid should substantially crowd out private long-term care insurance demand.

Other studies have found similar evidence that living arrangements – including nursing home residency and shared living arrangements – respond to the cost of long-term care. The Channeling Demonstration, an experiment that expanded the generosity of publicly-funded in-home care for low-income elderly in the 1980’s, led to reductions in nursing home use and reductions in shared living arrangements (Pezzin, Kemper and Reschovsky, 1996), while Orsini (2010) found that a 1997 reduction in Medicare reimbursement rates for home health care led to a significant increase in shared living arrangements. Coe, Goda and Van Houtven (2015) used state variation in subsidies to long-term care insurance policies as an instrument for insurance coverage and found that an increase in long-term care insurance coverage, which significantly lowers the marginal cost of formal care, induced less informal caregiving and less coresidence with adult children. Several papers (Van Houtven and Norton, 2004; Charles and Sevak, 2005; Van Houtven and Norton, 2008) use variation in the characteristics of children as instruments for informal care, and find that an increase in the probability of receiving informal care decreases and delays entry into a nursing home and decreases home health care.

Closely related to this paper, Grabowski and Gruber (2007) use variation in Medicaid spend-down policies and data from the National Long-Term Care Survey to examine nursing home use and find no effect of spend-down policies on nursing home use for a broadly defined sample of individuals.¹ This paper confirms this overall result, and follows up by examining a second and complementary outcome – coresidence. It also investigates heterogeneity at important margins – care needs, age, and income – to show that overall estimates of spend-down programs hide important effects for relevant sub-populations that are the main target of the policy.

In the next section I provide background on living arrangements and long-term care in the United States and the Medicaid policy variation at the core of this study. Section 3 proposes a simple conceptual framework to understand the link between living arrangements and nursing home costs. Section 4 describes the data and sample, and Section 5 presents the empirical strategy and the main results. Section 6 explores insurance mechanisms, and Section 7 concludes.

2 Background

2.1 Living arrangements

Living arrangements of the elderly vary widely, from fully independent living to 24-hour care in a nursing home. Historically, coresidence with family members was common-place, but this phenomenon has dramatically declined within the past century: in 1900, around 70% of elderly individuals lived with their adult children, while fewer than 15% do so today (Ruggles, 2007). This trend has been accompanied by a large increase in independent living. In 1900 only 15% of widows lived independently, while in 1990 over 60% did (McGarry and Schoeni, 2000). Finally, the past century has seen an increase in institutional residency, such as nursing homes, from around 3% of elderly widows to 10% from 1900-1990.

One reason for these trends is the rise in elderly income over the past century, which allowed more elderly to afford to live independently (Costa, 1997, 1999; Engelhardt, Gruber and Perry, 2005; Goda, Golberstein and Grabowski, 2011). However, another potentially important historical determinant of elderly living arrangements has been the ability to receive care in different settings as one becomes frail in old age. Thus, decreases in fertility, increases in female labor force participation, changes in health needs, and the availability of other sources of care for the elderly may also play a role in these trends. Today, the fertility relate hovers around 2.0, and the female labor force participation rate is upwards of 67% for prime-age females, suggesting that there is much less availability of adult children to help care for the elderly than a century ago (OECD, 2017a,b). In addition, more opportunities to receive in-home assistance (e.g. the expansion of home health agencies as well as programs such as Meals on Wheels) has made it more feasible to remain in one’s home without living completely independently.

2.2 Long-term care

Long-term care in the United States, defined as assistance performing Activities of Daily Living (ADLs) or Instrumental Activities of Daily Living (IADLs),² is common and expensive. 75% of elderly individuals will depend on long-term care at some point (Brown and Finkelstein, 2008), and in 2013, formal long-term care costs in the United States added up to $310 billion, or over 10 percent of all health expenditures for all ages (Kaiser Commission on Medicaid and the Uninsured, 2015). In 2015, the average cost of nursing home care was $91,000 and the average hourly price of a home care aide was $20 per hour (Kaiser Commission on Medicaid and the Uninsured, 2015). These costs are financed through three main sources: out-of-pocket spending accounts for 35% of total costs, private insurance accounts for less than 5%, and public insurance covers the other 60% (see Mommaerts (2016) for more details). The largest public payer is Medicaid, a means-tested program described in greater detail in Section 2.3. Medicare, the public health insurance program for individuals aged 65 and over, notably does not cover most long-term care expenses.

Additionally, over half of long-term care is provided informally by family members. Mommaerts (2016) shows that among retirees 65 and over who report receiving long-term care help, over 60% of primary caregivers are family members, and for single individuals with children, almost all informal caregivers are their adult children. When care is not provided by family members, over 90% of primary caregivers are paid. The fact that some individuals elect to purchase formal care in lieu of family care suggests that there are important implicit costs of informal care. These indirect costs, such as lost wages, are not included in the annual $310 billion price tag of long-term care. Using time-use and wage data, Chari et al. (2015) place the annual implicit cost of informal caregiving at an additional $520 billion.

2.3 Medicaid spend-down policies

Medicaid, the largest payer of formal long-term care expenditures, is a means-tested public health insurance program. Eligibility for Medicaid coverage requires that an individual’s income and assets fall below defined thresholds. Because the program is jointly funded by the federal and state governments, states have some discretion in determining the level of generosity of their programs.

Traditional Medicaid eligibility is assessed through “categorically needy” provisions, which are primarily based on the rules of the Supplemental Security Income (SSI) program. Individuals who qualify as categorically needy must have income and assets below certain thresholds set by the state in which they reside. In most states, the income threshold is in the range of 100%-300% of the SSI limit (100% of the SSI limit is about 75% of the federal poverty line) and the asset threshold is $2,000-$3,000, typically excluding a primary residence. In addition, in some states individuals who are not considered categorically needy may qualify for Medicaid as “medically needy” if their health expenses are high enough such that their income net of health expenses is below the income threshold. In these states, individuals are able to “spend down” their income to Medicaid eligibility. Finally, states are allowed to impose lower income thresholds than the federal threshold, but then must allow for a spend-down provision (states that elect to do this are “209(b)” states). I classify states with Medically Needy provisions and 209(b) states as spend-down states.³ As of 2009, 33 states including the District of Columbia had spend-down policies.

State Medicaid policies can vary in generosity among several other dimensions. On the individual (demand) side, as alluded to above, states have the option to increase or decrease the income and asset eligibility thresholds. I control for, but do not focus on this variation, because the high cost of long-term care is orders of magnitude larger than this policy variation (e.g. a $1,000 difference in the asset threshold would delay nursing home entry by one week (Grabowski and Gruber, 2007)). On the supply side, states can vary in (a) how much Medicaid reimburses nursing homes for Medicaid patients, (b) whether the state has a Certificate-of-Need law that regulates the number of new nursing home beds in a state, and (c) whether and how much the state Medicaid program spends on home- and community-based services (HCBS) in lieu of institutional services. These three factors may affect the ease with which an individual can find a nursing home bed or receive Medicaid dollars for in-home care. HCBS in particular increased dramatically over the period of study: in 1980 HCBS represented less than 10 percent of the Medicaid budget, while in 2014 it represented over 50 percent (Truven Health Analytics, 2016). This trend could be important for the outcomes measured in this study, because HCBS may shift Medicaid recipients away from nursing home services towards in-home services, which could mute the effect of spend-down policies on nursing home use. However, a crucial difference between these factors and spend-down policies is that spend-down policies affect Medicaid eligibility, while these other factors should only affect individual choices conditional on being eligible for Medicaid. Thus, these other factors are important controls for the analysis but the focus of this study is the effect of changes in Medicaid eligibility, which is generated by spend-down policies. In the next section, I discuss in a simple theoretical framework how Medicaid eligibility, by changing the cost that individuals face for formal care, may affect the type of long-term care an individual receives.

3 Conceptual framework

This section outlines a basic conceptual framework for why living arrangements of single elderly may respond to Medicaid policy, particularly for older, unhealthy, and low-income populations. The underlying model is one of intergenerational family decision-making in which an elderly individual may need long-term care, which can be provided in an institution or in the community. Community-based care can consist of paid care, informal care by the family, or a combination of the two.

Which living and care arrangement the family chooses will depend on the extensiveness of care needs, preferences over living arrangements and long-term care, time costs, and financial costs. First, an institutional setting such as a nursing home provides a high level of care, which may only be necessary and relevant for individuals with extensive care needs. Living in the community, on the other hand, allows for potentially more flexible care arrangements, because care can be hired on an hourly basis, provided informally by a family member, or both. Second, preferences over living and care arrangements, such as a taste for independence, a view that nursing homes are a “last resort”, or a preference over the relationship of a caregiver, may be a major determinant of living arrangement, particularly if these preferences are strong. Third, the opportunity cost of time for family members, particularly adult children who faces other demands for their time such as market work and childcare, will likely affect which type of care the elderly individual receives. For some families, coresidence may be an appealing option because living under the same roof may lower the time cost of informal care. Finally, financial concerns may affect decisions over living arrangements. Coresidence allows families to save on housing costs compared to independent living, and depending on the out-of-pocket cost for paid services, nursing home care and in-home paid care can be a significant financial burden.

Medicaid alters the financial burden of long-term care. Along with other rules, Medicaid spend-down policies allow more individuals to qualify for Medicaid, and therefore provide them with access to nursing home care or in-home professional care at a significantly reduced out-of-pocket cost.⁴ Whether these policies affect living arrangements depends crucially on their price elasticities. In other words, it will depend not only on the financial consequences of Medicaid but also the strength of preferences, opportunity costs, and care needs.

This framework suggests that this response should vary by population characteristics. First, spend-down policies should affect the demand for nursing home care and coresidence for families with care needs, but not those without care needs. For this reason, the empirical analysis not only examines the overall effect of spend-down policies, but also separately for subpopulations that may have low care needs (younger, healthier subpopulations) and high care needs (older, sicker subpopulations). Second, spend-down policies only lowers the price for elderly individuals who are near Medicaid eligibility thresholds, and hence the policy should affect lower-resource families but not necessarily be relevant for higher-income families. For this reason, the empirical analysis also examines the effect of spend-down policies by income.

4 Data

For the empirical analysis, I merge two types of data: (1) state (Medicaid) policies over time, and (2) survey data on long-term care outcomes.

For state Medicaid spend-down programs, I use data collected by and used in Grabowski and Gruber (2007) for 1980-1999, and supplement this with policy information collected in Stone (2004), Stone (2011), and Coe (2007) for 2000-2009. In 1980, 29 states including the District of Columbia had spend-down policies. This increased to 33 states by 2009, with 10 observable policy changes in my data. Other Medicaid policies, including income and asset thresholds, nursing home reimbursement rates, and Medicaid expenditures for home- and community-based services, come from a combination of Grabowski and Gruber (2007), Kaiser (2012), the Shaping Long Term Care in America Project at Brown University (http://ltcfocus.org/), and Truven Health Analytics (2016). Finally, state Certificate-of-Need laws came from Grabowski and Gruber (2007), Rahman et al. (2016), and Cauchi and Noble (2016). For more details of the state policy data and sources, see Appendix A.

To measure living arrangements, I require data on community-based and institutionalized individuals. One of the few datasets that collects information on institutionalized (nursing home) individuals is the Census and American Community Survey (ACS).⁵ I use the Integrated Public Use Microdata Series (IPUMS) sample of the 1980, 1990, and 2000 Census and the 2006 through 2009 ACS for the main specifications.

The measure of coresidence is whether the individual lives with an adult child, as defined by whether one of the individual’s children is present in the household at the time of the survey. One caveat to this sample is that I cannot distinguish between elderly individuals who have children and those who do not, though results below from the HRS show that 87% of single individuals aged 65 and over have children. The measure of nursing home residency is whether the individual is institutionalized at the time of the survey. Except for 1980, the survey does not ask whether the institution is a nursing home or other type of institution (such as a correctional or mental institution). In 1980, which is the only year in which the type of institution is specified, 96% of institutionalized individuals in my sample were nursing home residents.⁶

Because of the limited amount of information contained in the Census/ACS, I also use data on Medicaid and long-term care insurance take-up from the 1992-2010 Health and Retirement Study (HRS) to examine the effect of these policies on insurance demand. While the HRS also contains information on informal care and nursing home care, the time span only covers three policy changes, and hence I only use the HRS for these supplemental analyses, but show robustness of the Census/ACS results using both the HRS and the National Long-Term Care Survey in the appendix.

4.1 Sample

The main sample consists of single individuals aged 65 and above. I restrict the focus of the analysis to single elderly because their main source of informal care is their adult children (in contrast, married elderly are most likely to receive informal care from their spouses).

Table 1 reports summary statistics for the Census/ACS and HRS samples of single individuals aged 65 and over. Column (1) corresponds to the Census/ACS sample, which consists of almost 3 million person-year observations weighted by person-weights. 70% of the sample lives in a state that has a spend-down provision. 20% of the sample lives with their child, and 8% live in a nursing home. 16% of the sample reports difficulty living independently and caring for oneself (which I interpret as an indicator of long-term care needs), and almost a quarter of the sample reports being enrolled in Medicaid.⁷ The sample is mostly white and female, and the average annual personal income is $24,000 (converted to 2010 dollars using the CPI-U). The second column of Table 1 reports similar statistics for the HRS sample. Of note, 10% of the sample owns a private long-term care insurance policy. 39% of the sample reports being in fair or poor health, 30% reports needing help with at least one Activity of Daily Living, and among those individuals, on average they had difficulty with 2.3 ADLs.

Table 1.

Summary statistics: Census/ACS and HRS samples

	Census/ACS	HRS
Spend-down state	0.70	0.68
Has children	–	0.87
Never married	–	0.07
Coreside	0.20	0.22
Nursing home	0.08	0.08
Informal care	–	0.17
Hours/month (if any)	–	122
Difficulty w/ independence*	0.16	–
Fair or poor health	–	0.39
Any ADLs	–	0.30
# of ADLs (if any)	–	2.37
Medicaid*	0.23	0.16
Long-term care insurance	–	0.10
White	0.84	0.79
Less than high school	0.35	0.37
College	0.26	0.28
Male	0.24	0.22
Age	77.2	78.2
Income (in 2010 $)**	24,131	23,573
Person-year observations	2,911,643	37,785

Note: Table shows means of a sample of unmarried individuals aged 65 and over from the pooled 1980-2000 Census and 2006-2009 American Community Survey (ACS) data in column (1) and from the pooled 1992-2008 HRS in column (2).

^*For the Census/ACS sample, Medicaid is only available in 2008 and 2009 and difficulty being alone is only available starting in 1990.

^**For the Census/ACS sample, income is defined as personal income, and for the HRS sample, income is defined as household income. Dollar figures are converted to 2010 dollars using the CPI-U.

Figure 1 shows the rates of coresidence and nursing home use in the Census/ACS sample by whether the individual lives in a state with a Medicaid spend-down provision. The sample is broken down by long-term care needs, where the black lines denote individuals who report difficulty living independently and caring for oneself and the gray lines denote individuals who do not report such difficulties. The x-axis in the first row reports coresidence and nursing home rates by income quintile, and the x-axis in the second row reports rates by age.

Figure 1.

Long-term care by spend-down provision, Census/ACS sample

Note: The figures on the left (right) show the percent of individuals who coreside with their child (reside in a nursing home) among the sample of single individuals aged 65 and over in the Census/ACS. The x-axis in the top row is income quintile, and the x-axis in the bottom row is age. For each graph, the results are split by (1) whether the respondent reports difficulty living independently and caring for oneself and (2) whether the individual resides in a Medicaid spend-down state. Median annual income for each quintile is $5486; $10,181; $15,016; $23,300; and $45,587 respectively (in 2010 dollars).

As shown in the left figures, coresidence is higher in states without Medicaid spend-down provisions for the “dependent” sample (in black), suggesting that individuals are price-sensitive to formal long-term care and substitute away from formal care when formal care is costly. The right figures corroborate this. The top left figure also shows that coresidence is hump-shaped across the income distribution. For low-income individuals, this is consistent with the fact that Medicaid pays for formal long-term care and thus individuals are less likely to rely on informal care. For high-income individuals, this may indicate that they are less price sensitive to formal long-term care costs because they can more easily afford the out-of-pocket costs. The higher rate of coresidence for middle-income individuals is consistent with the fact that they do not qualify for Medicaid, but also do not have as many financial resources to devote to formal care. Again, the top right figure corroborates this, except high-income individuals are less likely to reside in nursing homes, potentially because they can afford in-home formal care. The results by age in the second row also show that “dependent” individuals who live in spend-down states are less likely to coreside with adult children and more likely to reside in a nursing home. In addition, nursing home rates increase with age, while coresidence (generally) decreases with age.

In both figures, the difference in long-term care use does not vary by whether the state has a spend-down provision for individuals who do not report difficulty caring for themselves, as shown by the gray lines. These individuals are also much less likely to reside in a nursing home. These two patterns suggest that individuals without long-term care needs are not sensitive to formal long-term care costs as proxied by their state’s spend-down status, and that coresidence captures long-term care needs relatively well. The next subsection explores this measure further.

4.2 The relationship between coresidence and informal care

One reason coresidence may be a particularly interesting outcome to compare to nursing home residency is that it may be a convenient arrangement through which to receive care, particularly from family members. However, not all care from family members is necessarily provided through coresident arrangements, and not all coresident arrangements involve informal care (for example, some families may coreside to save on housing costs or because they value communal living).

To better characterize the relationship between coresidence and informal care, Table 2 uses the Health and Retirement Study to quantify how often coresident and non-coresident elderly receive informal care (Panels A through D) and how often elderly receiving and not receiving informal care coreside (Panel E). The first column of Panel A shows that among individuals aged 65 and over, 31% of coresident households also received at least one hour of informal care in the past month, while only 10% of non-coresident households did. This gap in the rate of informal care between coresident and non-coresident households increases as the sample becomes older and sicker: the fourth column shows that 75% of coresident households and only 34% of non-coresident households received informal care in the past month. Panel B repeats this exercise for a stricter definition of informal care (20 or more hours of care in the past month), and Panels C and D reports mean and median hours of care, respectively. From each panel, it is clear that coresident households receive much more informal care than non-coresident households. Panel E flips the relationship: the final column reports that 48% of informal care recipients age 80 and over with at least 1 ADL coreside with an adult child, while only 14% of those who do not receive informal care do so.

Table 2.

Relationship between coresidence and informal care in the HRS

	Age 65+	Age 80+	Age 80+, unhealthy	Age 80+, 1+ ADL
	Panel A: Any informal care
Coresident households	0.31	0.47	0.61	0.75
Non-coresident households	0.10	0.16	0.26	0.34
	Panel B: 20+ hours informal care
Coresident households	0.26	0.42	0.55	0.69
Non-coresident households	0.06	0.10	0.16	0.22
	Panel C: Mean informal care hours
Coresident households	68	117	162	208
Non-coresident households	8	13	24	34
	Panel D: Median informal care hours
Coresident households	0	0	32	90
Non-coresident households	0	0	0	0
	Panel E: Coresidence
Receive any informal care	0.47	0.45	0.45	0.48
Receive no informal care	0.18	0.15	0.16	0.14
Observations	31,368	12,740	5,029	4,617

Note: Table shows the prevalence of informal care or coresidence as a function of the other, among a sample of unmarried individuals aged 65 and over from the 1994-2010 HRS. Panels A, B, C and D report measures of informal care as a function of coresidence: Panels A and B report whether the individual received any informal care and over 20 hours of informal care in the past month respectively, and Panel C and D report the mean and median (unconditional) number of informal care hours the individual received in the past month, respectively. Panel E report the percent of individuals that coreside as a function of whether they receive informal care. The columns restrict the sample: column (1) contains the full sample, column (2) contains individuals 80 and over, column (3) contains individuals 80 and over who report being in fair or poor health, and column (4) contains individuals 80 and over who report difficulty with at least one ADL.

In sum, informal care is much more prevalent among coresident elderly than non-coresident elderly, particularly among those who have difficulty with ADLs. In addition, coresidence is much more prevalent among elderly who receive informal care than those who do not receive informal care, again particularly among those who have difficulty with ADLs.

5 Empirical Strategy and Results

5.1 Empirical strategy

Over the last 30 years, many states have adopted and dropped spend-down policies. This allows for an analysis that exploits within-state variation to estimate the effect of Medicaid spend-down policies on various long-term care outcomes. I run the following model specification:

$$Y_{\mathit{ist}}=\beta {\mathrm{SD}}_{\mathit{st}}+\eta P_{\mathit{st}}+\gamma X_{\mathit{ist}}+{\alpha }_s+{\delta }_t+{\epsilon }_{\mathit{ist}}$$

in which i indexes the individual, s the state, and t the year. SD_st is an indicator for whether the state has a spend-down provision, and β is the main coefficient of interest. In the main results, Y_ist is either an indicator for coresidence with a child or an indicator for nursing home use; I additionally run specifications in which Y_ist is an indicator for Medicaid enrollment or holding a long-term care insurance policy. P_st is a set of other state policies, including the Medicaid income and asset thresholds, the average daily Medicaid payment rate for a nursing home stay, whether the state Medicaid program spends any money for the aged or disabled on home- and community-based services, and whether the state has a certificate-of-need law for new nursing home construction. X_ist is a set of individual control characteristics. α_s and δ_t are state and year fixed effects, so that β is identified by within-state variation in SD over time. In all specifications, I use a linear probability model and cluster the standard errors at the state level. The Census/ACS specifications are adjusted using person-level weights.

5.2 Results

Table 3 reports estimates of the effect of living in a state with a Medicaid spend-down provision on coresidence with a child and nursing home use. Columns (1) and (2) show that Medicaid spend-down policies have a small negative and significant overall effect on coresidence with an adult child, and a small and imprecisely estimated overall effect on residing in a nursing home. Columns (3) and (4) report effects interacted with 5-year age groups, and show that there is significant heterogeneity by age: there are small and insignificant impacts of spend-down policies on living arrangements for individuals under age 80, and significant effects that increase in magnitude with age starting around age 80. For 85-89 year-olds, the coefficients imply that the introduction of a Medicaid spend-down policy induces a 2.4 percentage point decrease in the number of individuals living with their children and a 1.6 percentage point increase in the number of individuals living in nursing homes. This is a 12% decrease and 21% increase from a base of 20% and 8% of individuals living with children and in nursing homes, respectively, in states without Medicaid spend-down provisions. This age-effect is consistent with the conceptual framework, which hypothesized that living arrangements such as coresidence and nursing home residency are much more likely to be sensitive to Medicaid policy for individuals who are less healthy. Since individuals are more likely to be unhealthy at older ages, we should expect to see this age gradient.

Table 3.

Effect of spend-down provisions on coresidence and nursing home use

	(1)	(2)	(3)	(4)
Sample:	Age 65+	Age 65+	Age 65+	Age 65+
Dependent variable:	Coresidence	Nursing Home	Coresidence	Nursing Home
Spend-down provision	-0.013** (0.005)	0.004 (0.004)
Spend-down, 65-69			0.001 (0.006)	-0.006 (0.005)
Spend-down, 70-74			-0.008 (0.006)	-0.005 (0.004)
Spend-down, 75-79			-0.009 (0.007)	0.000 (0.004)
Spend-down, 80-84			-0.018*** (0.006)	0.006 (0.005)
Spend-down, 85-89			-0.024*** (0.008)	0.016* (0.009)
Spend-down, 90+			-0.044*** (0.009)	0.045** (0.018)
Mean dept variable	0.199	0.076	0.199	0.076
N	2911364	2911364	2911364	2911364

Note: The sample includes single individuals aged 65 and over in the pooled 1980-2000 Census and 2006-2009 American Community Survey. The estimates in each column are from a linear probability model in which the dependent variable in columns (1) and (3) are whether the individual coresides with an adult child and in columns (2) and (4) are whether the individual is institutionalized. The main independent variables in columns (1) and (2) are whether the individual lives in a state with a Medicaid spend-down provision. Columns (3) and (4) interact the spend-down variable with the 5-year age bin of the individual. In addition, each regression is weighted using person-weights and includes controls for Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, whether the state has a certificate-of-need law, year, state, income quintile, age quintile, race, education, gender, and marital status. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01

Table 4 explores other sources of heterogeneity through which the effect may work, focusing on the sample of individuals aged 80 and over. I first examine the effect of spend-down policies by income quintile, under the hypothesis that lower income individuals are more likely to be on the margin of qualifying for Medicaid through the spend-down provisions. If income and assets are positively correlated, one would expect this effect to be even stronger. Columns (1) and (2) show that the effects are concentrated in the lower half of the income distribution: the introduction of a Medicaid spend-down policy induces almost a 3 percentage point decrease in coresidence with children among the bottom 60% of the income distribution (the median annual income in the third quintile is $15,016 in 2010 dollars), and no detectable decrease for the top 40%. Analogously, there is almost a 4 percentage point increase in nursing home use for the bottom 20% of the income distribution in response to the introduction of a Medicaid spend-down policy.

Table 4.

Effect of spend-down provisions on coresidence and nursing home use

	(1)	(2)	(3)	(4)
Sample:	Age 80+	Age 80+	Age 80+	Age 80+
Dependent variable:	Coresidence	Nursing Home	Coresidence	Nursing Home
Spend-down, Income Q1	-0.023 (0.017)	0.038*** (0.012)
Spend-down, Income Q2	-0.025*** (0.008)	-0.004 (0.009)
Spend-down, Income Q3	-0.029*** (0.007)	-0.008 (0.009)
Spend-down, Income Q4	-0.012 (0.015)	-0.002 (0.007)
Spend-down, Income Q5	-0.004 (0.018)	-0.009 (0.011)
Spend-down, Independent			-0.009 (0.006)	-0.016** (0.007)
Spend-down, Dependent			-0.041*** (0.011)	0.046* (0.025)
Mean dept variable	0.210	0.140	0.208	0.135
N	1047313	1047313	863340	863340

Note: The sample includes single individuals aged 80 and over in the pooled 1980-2000 Census and 2006-2009 American Community Survey. The estimates in each column are from a linear probability model in which the dependent variable in columns (1) and (3) are whether the individual coresides with an adult child and in columns (2) and (4) are whether the individual is institutionalized. The main independent variables in columns (1) and (2) are whether the individual lives in a state with a Medicaid spend-down provision, interacted with the income quintile of the individual. Columns (3) and (4) interact the spend-down variable with whether the individual reports difficulty living independently and caring for oneself, which is only available beginning in 1990. In addition, each regression is weighted using person-weights and includes controls for Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, whether the state has a certificate-of-need law, year, state, income quintile, age quintile, race, education, gender, and marital status. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01

Columns (3) and (4) report estimates by whether the individual reports difficulty caring for herself and living independently, to explore the hypothesis that individuals who need more long-term care are likely to be more affected by a policy that activates once one needs care. One caveat to these estimates is that this information was not collected in the 1980 Census, and thus does not exploit several of the points of policy variation for identification. Nonetheless, using 1990-2009 data, the results show that the effect of Medicaid spend-down provisions are concentrated among individuals who need care (i.e. “dependent”): the introduction of a spend-down program induces over a 4 percentage point decrease in living with children and an equal increase in living in a nursing home.⁸

The results in this section indicate that Medicaid spend-down policies affect the living arrangements of elderly individuals. The results from the sub-population analyses suggest that the mechanism through which Medicaid spend-down policies affect coresidence and nursing home use is through access to regular care associated with the difficulties of aging: the effects are concentrated among individuals aged 80 and over, lower-income individuals, and individuals who need help caring for themselves.⁹

These findings are not inconsistent with the null nursing home result in Grabowski and Gruber (2007): while the effect sizes are large for certain populations, the overall effect is small and insignificant. To corroborate this further, Appendix Table 1 repeats the analysis using the same data source as Grabowski and Gruber (2007) and finds that (1) the overall nursing home effect is small and insignificant, but (2) the sub-population analysis results in similar effects as the Census/ACS analysis (although less precise), and (3) there are also strong(er) effects on coresidence, particularly at older ages and for individuals in poor health. In sum, these analyses show that the overall null result for nursing home use hides important heterogeneity by age, income, and health status.

For almost all of the results, I cannot reject that the spend-down coefficient for coresidence and nursing home are equal but opposite in sign at the 5% level. This implies that I cannot reject that the substitution between coresidence and nursing home use is one-for-one, which suggests that individuals appear to be elastic to price on the intensive margin of care (i.e. the setting in which to receive care) but inelastic to price on the extensive margin (i.e. whether to receive care or not). The main exception to this is the effect of spend-down policies for individuals in the second lowest and middle quintile of the permanent income distribution: the effect is significantly larger on coresidence than on nursing home use (in absolute value) at the 1% level.¹⁰ This could be evidence that some individuals in need of care substitute away from coresidence into independent home care (instead of nursing home care), which Medicaid covers in some circumstances.

5.3 Multiple hypotheses tests and placebo tests

One may be concerned about multiple inference problems when testing many hypotheses as well as the validity of state-year difference-in-difference tests. To jointly test hypotheses, I implement a step-down procedure proposed by Romano and Wolf (2005) that controls the probability of falsely rejecting at least one true null hypothesis. First, a joint test of the twelve age-related hypotheses in columns (3) and (4) of Table 3 finds that the main conclusions about coresidence survive at the 10% level and are unlikely to be due to false rejections (Romano and Wolf, 2005). Second, a joint test of the ten income-related hypotheses also finds that the main conclusions survive, and finally, a joint test of the four health-related hypotheses finds that the coresidence results survive. Thus, the results in Tables 3 and 4 – particularly for coresidence – do not appear to be the result of false rejections due to the multiple hypotheses being tested.

In addition to multiple inference issues, one may be concerned that our difference-indifference approach leads to spurious effects. While it would be ideal to show a standard check for parallel trends before and after state rule changes, most of the spend-down policy changes occurred in the late 1980s and early 1990s, so there are only one or two datapoints to examine pre-trends. Instead, I run a placebo test. Specifically, I use randomization inference in which I rerun the analysis 100 times, each time randomly assigning state spend-down policies to different states (Bertrand, Duflo and Mullainathan, 2004).

Figure 2 shows the kernel distribution of each coefficient from this procedure corresponding to the age-interacted coefficients as in Table 3, with dotted vertical lines marking the 10th and 90th percentile of this distribution, and the solid gray vertical line marking the coefficient from Table 3. The first column of figures shows that the effect of spenddown policies on coresidence for younger individuals is not outside 10th or 90th percentile of the placebo distribution, as expected, but for older individuals the coefficient is outside or near the border of this placebo distribution. This is also the case for the nursing home effects in the second column of figures. Similarly, Figures 3 and 4 show similar results for the dependence- and income-interacted coefficients in Table 4: the coefficients on coresidence and nursing home residency for dependent individuals are outside the 10th and 90th percentile of the placebo distributions, and they are outside or near the border for the second and third income quintile for coresidence and the first income quintile for nursing home residency.

Figure 2.

Placebo estimates of spend-down provisions on coresidence and nursing home use, by age

Note: This figure plots the distribution of coefficients from a placebo exercise in which states are randomly given spend-down policies from the distribution of true state policies, repeated 100 times. The solid vertical gray line in each figure represents the estimated coefficient from Table 3 and the dashed vertical lines represent the 10th and 90th percentiles of the placebo distribution.

Figure 3.

Placebo estimates of spend-down provisions on coresidence and nursing home use, by dependency

Note: This figure plots the distribution of coefficients from a placebo exercise in which states are randomly given spend-down policies from the distribution of true state policies, repeated 100 times. The solid vertical gray line in each figure represents the estimated coefficient from Table 4 and the dashed vertical lines represent the 10th and 90th percentiles of the placebo distribution.

Figure 4.

Placebo estimates of spend-down provisions on coresidence and nursing home use, by income

Note: This figure plots the distribution of coefficients from a placebo exercise in which states are randomly given spend-down policies from the distribution of true state policies, repeated 100 times. The solid vertical gray line in each figure represents the estimated coefficient from Table 4 and the dashed vertical lines represent the 10th and 90th percentiles of the placebo distribution.

In sum, the results in this section show that Medicaid spend-down policies affect the living arrangements of elderly individuals, particularly older individuals (who are more likely to need care that could be paid by Medicaid), low income elderly (who are more likely to qualify for Medicaid), and sicker elderly (who are also more likely to need care). These findings generally withstand tests that correct for multiple hypotheses as well as placebo tests, and oftentimes I cannot reject that the coefficients on coresidence and nursing home residency are equal-but-opposite signed, suggesting that they substitute for one another. The next section verifies that these effects are also associated with increases in Medicaid enrollment and investigates whether spend-down policies affect the demand for private long-term care insurance.

6 Mechanisms: public and private insurance

The mechanism through which Medicaid spend-down policies should affect the living arrangement outcomes in Section 5 is through changes in Medicaid enrollment by way of eligibility. Unfortunately, the Census does not collect information on Medicaid enrollment, and the ACS only began to collect this data in 2008. To continue to exploit the difference-in-difference strategy, I use 1992-2010 data from the Health and Retirement Study. This is still not ideal, since most policy changes occurred in the late 1980’s and early 1990’s, but offers suggestive evidence of a “first-stage”.¹¹

Table 5 reports the effect of Medicaid spend-down policies on Medicaid enrollment for single individuals aged 65 and over. Column (1) shows that the introduction of a Medicaid spend-down provision induces a 3.7 percentage point increase in Medicaid enrollment, confirming that Medicaid spend-down provisions allow more individuals to enroll in Medicaid. Column (2) shows that this effect is concentrated among individuals with adult children. One interpretation of this result is that individuals with children have more opportunities for alternative care arrangements, such as coresidence, so that this ability to substitute makes them more price sensitive than individuals who lack alternative arrangements. Finally, column (3) shows that while this effect is relatively large for both individuals who do and do not report being in fair or poor health, the effect is more precise for unhealthy individuals. This is consistent with the findings from the Census/ACS sample and with the fact that spend-down policies should only affect those who have difficulty living independently.

Table 5.

Effect of spend-down provisions on Medicaid take-up

	(1)	(2)	(3)
Sample:	Age 65+	Age 65+	Age 65+
Dependent Variable:	Medicaid	Medicaid	Medicaid
Spend-down provision	0.037** (0.014)
Spend-down, no kids		-0.005 (0.021)
Spend-down, has kids		0.043*** (0.015)
Spend-down, healthy			0.035* (0.019)
Spend-down, unhealthy			0.038** (0.017)
Mean dept var	0.163	0.163	0.163
N	29464	29464	29464

Note: The sample includes single individuals in the 1992-2008 Health and Retirement Study. The estimates in each column are from a linear probability model in which the dependent variable is whether the individual is covered by Medicaid. The main independent variable in column (1) is whether the individual lives in a state with a Medicaid spend-down provision. Column (2) interacts the spend-down variable with whether the individual has children, and column (3) interacts the spend-down variable with whether the individual reports being in fair or poor health. In addition, each regression includes controls for Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, whether the state has a certificate-of-need law, year, state, income quintile, 5-year age bin, race, education, and gender. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01

Aside from altering living arrangements and care choices, Medicaid coverage – or the prospect of future Medicaid coverage – may alter earlier decisions in preparation for the risk of needing long-term care, such as private long-term care insurance coverage. Other studies posit that Medicaid causes significant crowd-out of private long-term care insurance (Pauly, 1990; Brown and Finkelstein, 2008) but do not empirically test these claims using policy variation (one notable exception is Brown, Coe and Finkelstein (2007), who find that Medicaid asset limits can have a large effect on private long-term care insurance demand). Table 6 reports the effect of Medicaid spend-down policies on private long-term care insurance ownership for a sample of 60-69 year olds, which is the main decade during which individuals make this purchase decision (Brown and Finkelstein, 2007). There is no significant effect for the overall sample (column (1)), but there is a 3.8 percentage point decrease in insurance take-up when I restrict the sample to healthy individuals, since unhealthy individuals are often unable to purchase long-term care insurance (column 2). Finally, column (3) splits this effect by whether the individual has children, and finds that it is only precise for individuals with children. This result is consistent with the mechanisms discussed in Mommaerts (2016), which shows that insurance demand is crowded out by both Medicaid, which provides a cheaper substitute source of insurance, and informal care by children, which provides a potentially cheaper and preferred source of care than that covered by private insurance.

Table 6.

Effect of spend-down provisions on long-term care insurance in the HRS

	(1)	(2)	(3)
Sample:	Age 60-70	Healthy, 60-70	Healthy, 60-70
Dependent Variable:	LTC Insurance	LTC Insurance	LTC Insurance
Spend-down provision	-0.009 (0.011)	-0.038*** (0.014)
Spend-down, no kids			-0.051 (0.045)
Spend-down, has kids			-0.037** (0.014)
Mean dept var	0.084	0.102	0.102
N	12360	8164	8164

Note: The sample includes single individuals aged 60-70 in the 1992-2008 Health and Retirement Study. Columns (2) and (3) restrict the sample to individuals who do not report being in fair or poor health. The estimates in each column are from a linear probability model in which the dependent variable is whether the individual owns a long-term care insurance policy and the independent variable is whether the individual lives in a state with a Medicaid spend-down provision, and column (3) interacts the spend-down provision with whether the individual has children. In addition, each regression includes controls for Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, whether the state has a certificate-of-need law, year, state, income quintile, 5-year age bin, race, education, and gender. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01

7 Conclusion

Long-term care is an important source of risk for the elderly in the United States. The interaction between different sources of care, such as nursing homes, in-home care, and informal care, is important not only for individual welfare but also for the design of social insurance programs. This paper exploits variation in Medicaid policies to determine whether coresidence with adult children can substitute for formal sources of long-term care in response to price changes. If individuals are price sensitive to the source of care, the ability to substitute between living arrangements and alternative sources of care can have important implications for Medicaid policy.

I use data from the decennial Census, the American Community Survey, and the Health and Retirement Study over a 30 year period to estimate the responsiveness of living arrangements to changes in the existence of Medicaid spend-down policies across states and over time. I find that fewer elderly reside with their children, and more reside in nursing homes in states that have this optional Medicaid policy that results in more generous Medicaid eligibility standards. These effects are driven by subpopulations that we would expect to be most sensitive: individuals aged 80 and over, as well as individuals with low income and those who have difficulty caring for themselves. In addition, I find that individuals are more likely to report Medicaid enrollment and less likely to purchase private long-term care insurance, suggesting a role of Medicaid in crowding out private insurance demand. Overall these findings suggest that changes in Medicaid coverage of long-term care benefits could have large impacts on long-term care utilization patterns, Medicaid expenditures, and, ultimately, individual welfare.

Appendix A

State policy sources and coverage

The following list contains the data sources and coverage for the state policies used in the empirical analysis.

1. Medicaid spend-down policies: I use data collected by and used in Grabowski and Gruber (2007) for 1980-1999, and supplement this with policy information collected in Stone (2004), Stone (2011), and Coe (2007) for 2000-2009. In 1980, 29 states including the District of Columbia had spend-down policies. This increased to 33 states by 2009, with 10 observable policy changes in my data: GA added in 1990 and lost in 2005, LA added in 1992, NJ added in 1995, OR added in 1986 and lost in 1991, RI added in 1986, TN added in 1987, lost in 2005, and added in 2008.

2. Medicaid income and asset thresholds: I use data collected by and used in Grabowski and Gruber (2007) for 1980-2000, and supplement this with policy information collected by Kaiser Commission on Medicaid and the Uninsured (2010) for 2009. Data for 2006-2008 are missing, so in the empirical analyses I flag these observations.

3. Medicaid nursing home reimbursement rates: I use data collected by the Shaping Long Term Care in America Project at Brown University (http://ltcfocus.org/) for 1980-2009.

4. Medicaid spending on home- and community-based services (HCBS): I use data from Truven Health Analytics (2016) for 1981-2009. I break down HCBS spending by (1) 1915(c) waivers, (2) home health, (3) personal care, and (4) other HCBS programs, and only count spending for the aging and disabled population (one exception is the 1915(c) waiver program, which only includes overall spending prior to 1995). I assign all states a value of zero for 1980 due to missing data and because some Medicaid HCBS programs were not yet in existence (e.g. the 1915(c) waiver program).

5. Certificate-of-Need laws: I use data collected by and used in Grabowski and Gruber (2007) for 1981-2000 and supplement this with Rahman et al. (2016) for 1980 and Cauchi and Noble (2016) 2006-2009.

Appendix B

Appendix Table 1.

Effect of spend-down provisions on coresidence and nursing home use (NLTCS)

Dependent variable:	Nursing Home			Coresidence
	(1)	(2)	(3)	(4)	(5)	(6)
Spend-down provision	-0.010 (0.016)			-0.028 (0.029)
SD × Age 65-69		-0.013 (0.016)			-0.034 (0.026)
SD × Age 70-74		-0.012 (0.015)			-0.006 (0.032)
SD × Age 75-79		-0.011 (0.017)			-0.031 (0.032)
SD × Age 80-84		-0.017 (0.018)			-0.038 (0.034)
SD × Age 85-89		0.006 (0.018)			-0.064* (0.035)
SD × Age 90+		0.018 (0.025)			-0.065* (0.035)
SD × Not many ADLs			-0.014 (0.016)			-0.024 (0.030)
SD × Many ADLs			0.038 (0.034)			-0.077*** (0.028)
Mean dept variable	0.091	0.091	0.091	0.288	0.288	0.288
N	33205	33205	33205	33111	33111	33111

Note: The sample includes single individuals aged 65 and over in the pooled 1984, 1989, 1994, and 1999 National Long-Term Care Survey. The estimates in each column are from a linear probability model in which the dependent variable in columns (1)-(3) are whether the individual resides in a nursing home and in columns (4)-(6) are whether the individual coresides with an adult child. The main independent variable is whether the individual lives in a state with a Medicaid spend-down provision. Columns (2) and (5) interact the spend-down variable with the 5-year age bin of the individual, and columns (3) and (6) interact the spend-down variable with whether the individual reports difficulty with over three Activities of Daily Living. In addition, each regression is weighted using person-weights and includes controls for Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, whether the state has a certificate-of-need law, year, state, income quintile, age quintile, race, education, gender, number of children, marital status, and marital status interacted with year. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01

Appendix Table 2.

Effect of spend-down provisions on coresidence and nursing home use, varying controls

	(1)	(2)	(3)	(4)	(5)	(6)
Sample:	Age 65+	Age 65+	Age 65+	Age 65+	Age 65+	Age 65+
Dependent Variable:	Coresidence	Nursing Home	Coresidence	Nursing Home	Coresidence	Nursing Home
Spend-down provision	-0.007 (0.007)	0.006** (0.003)	-0.015* (0.008)	0.006* (0.003)	-0.013** (0.005)	0.004 (0.004)
Income test ($10,000)			-0.031 (0.035)	-0.021 (0.019)	-0.022 (0.026)	-0.025 (0.019)
Asset test ($10,000)			-0.009 (0.011)	0.007 (0.007)	0.001 (0.010)	0.002 (0.007)
Medicaid rate ($100)			-0.003 (0.014)	0.002 (0.007)	-0.010 (0.013)	0.006 (0.007)
CON law					-0.011 (0.013)	0.005 (0.006)
Any 1915(c) HCBS spending					-0.001 (0.012)	-0.010*** (0.003)
Any home health HCBS spending					0.012 (0.008)	-0.004 (0.004)
Any personal care HCBS spending					0.015*** (0.005)	-0.007* (0.004)
Any other HCBS spending					0.008** (0.004)	-0.005 (0.003)
Mean dept variable	0.199	0.076	0.199	0.076	0.199	0.076
N	2911364	2911364	2911364	2911364	2911364	2911364

Note: The sample includes single individuals aged 65 and over in the pooled 1980-2000 Census and 2006-2009 American Community Survey. The estimates in each column are from a linear probability model in which the dependent variable in columns (1), (3), and (5) are whether the individual coresides with an adult child and in columns (2), (4), and (6) are whether the individual is institutionalized. The main independent variables are whether the individual lives in a state with a Medicaid spend-down provision, Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, and whether the state has a certificate-of-need law. In addition, each regression is weighted using person-weights and includes controls for year, state, income quintile, 5-year age bin, race, education, gender, and marital status. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01

Appendix Table 3.

Effect of spend-down provisions on outcomes in the HRS

	(1)	(2)	(3)	(4)
Sample:	Age 80+	Age 80+	Age 80+	Age 80+
Dependent variable:	Coresidence	Nursing Home	Informal Care, any	Informal Care, 40+ hrs
Spend-down, healthy	0.001 (0.042)	0.031* (0.018)	0.006 (0.019)	-0.012 (0.016)
Spend-down, unhealthy	-0.065 (0.043)	0.032 (0.022)	0.009 (0.026)	-0.028* (0.016)
Mean dept variable	0.205	0.145	0.262	0.122
N	12607	11165	11455	11455

Note: The sample includes single individuals aged 80 and over with children in the 1992-2008 Health and Retirement Study. The estimates in each column are from a linear probability model in which the dependent variable is (1) whether the individual coresides with an adult child, (2) whether the individual is institutionalized, (3) whether the individual received any informal care from an adult child in the previous month, and (4) whether the individual received 40 or more hours of informal care from an adult child in the previous month. The independent variable is whether the individual lives in a state with a Medicaid spend-down provision, interacted with whether the individual is in fair or poor health. In addition, each regression includes controls for Medicaid income and asset thresholds, Medicaid average nursing home payment rate, whether the state Medicaid program spends any money for the aged or disabled on 1915(c) waivers, home health, personal care, and other HCBS programs, whether the state has a certificate-of-need law, year, state, income quintile, 5-year age bin, race, education, and gender. Standard errors, clustered by state, are in parentheses.

^*p < 0.10,

^**p < 0.05,

^***p < 0.01