Measuring State Medicaid Home Care Participation and Intensity Using Latent Variables

Abstract

Population aging and policies to redirect long-term care toward home- and community-based services have led to increases in Medicaid home care spending in most states. Changes in state Medicaid home care policy generosity may result from changes in the number of persons served (i.e., Participation) and/or changes in quantities of services covered (i.e., Intensity). This study measures state Medicaid home care Participation and Intensity comprehensively using latent variables, and uses those latent variables to describe changes in Medicaid home care policy generosity over time and across states. Yearly state-level data from the Medicaid Statistical Information System (1999-2012) are analyzed using exploratory and confirmatory factor analyses. Between 1999 and 2012, 29 states expanded both Participation and Intensity, whereas six states reduced both. In the remaining states, a trade-off occurred. Distinguishing between Medicaid home care Participation and Intensity deserves attention, as expansions along these two dimensions represent potentially different implications for beneficiaries.

Keywords

state Medicaid home care policy home care comprehensive measure latent variable latent construct factor analysis

Introduction

States can increase their Medicaid home care policy generosity in two main ways: by expanding access to home care programs and/or by providing larger quantities of care to home care users, once access is granted (Kemper, Weaver, Short, Shea, & Kang, 2008). Stated differently, states can increase generosity in the Participation and/or Intensity dimensions of Medicaid home care policy.

Medicaid-funded home care mainly includes home health and personal care services provided under states’ Home Health Plans, Personal Care Plans, and several waivers and demonstrations. Home care services represent the bulk of Medicaid home- and community-based services (HCBS; Harrington, Ng, Laplante, & Kaye, 2012; Konetzka, 2014). Home care services continue to expand and represent increasing proportions of Medicaid long-term services and supports (LTSS), in part, as a result of the provisions in the Affordable Care Act (ACA) that promote such home-based services. Comparing the generosity of Medicaid home care policy across states and over time is needed to inform policy decisions. Doing that requires measures of state Medicaid home care policy that capture the strategies adopted by each state as comprehensively as possible.

One indicator of overall Medicaid home care policy generosity commonly found in the literature is states’ Medicaid home care expenditures per capita. Variants of this indicator include Medicaid home care expenditures per enrollee, HCBS or home care expenditures per elderly individual, and Medicaid 1915(c) waiver expenditures per eligible or enrolled elderly individual in each state (Eiken, Sredl, Burwell, & Woodward, 2017; Gardner & Gilleskie, 2012; Kemper et al., 2008; Kenney & Rajan, 2000; Muramatsu & Campbell, 2002; Muramatsu et al., 2007; Muramatsu, Hoyem, Yin, & Campbell, 2008; Muramatsu, Yin, & Hedeker, 2010; Ng, Harrington, & Musumeci, 2011; Pezzin & Kasper, 2002; Rahman, Tyler, Thomas, Grabowski, & Mor, 2015). Another common indicator is the percentage of LTSS expenditures allocated to home care or HCBS. This can be viewed as a measure of the degree of priority given to home care or HCBS over institutional LTSS (Blackburn, Locher, Morrisey, Becker, & Kilgore, 2016; Borck, Schmitz, Doty, & Drabek, 2016; Burr, Mutchler, & Warren, 2005; Eiken et al., 2017; Gardner & Gilleskie, 2012; Konetzka, Karon, & Potter, 2012; Miller, 2011; Muramatsu et al., 2008; Muramatsu et al., 2007; Mor et al., 2007; Ng et al., 2011; Pezzin & Kasper, 2002; Rice, Kasper, & Pezzin, 2009; Thomas & Mor, 2013; Wenzlow, Eiken, & Sredl, 2016). Some studies report the number of participants on Medicaid HCBS programs per capita, or the percentage of Medicaid LTSS beneficiaries who receive home care/HCBS rather than institutional services. Both of these can be seen as indicators of participation in Medicaid home care programs (Borck et al., 2016; Eiken, 2017a, 2017b; Kitchener, Ng, & Harrington, 2007a, 2007b; Ng et al., 2011). Finally, a few studies present indicators of intensity of use of HCBS services provided under different Medicaid programs, namely, expenditures per participant in 1915(c) waivers, Personal Care Plans, and Home Health Plans, and total HCBS spending per HCBS user (Borck et al., 2016; Ng, Harrington, Musumeci, & Ubri, 2016; Rice et al., 2009).

The literature relies on the above observed indicators of home care use to capture an inherently latent concept—Medicaid home care policy generosity—because states’ policies are challenging to measure. As noted, for example, by Borck et al. (2016), within the federal guidelines, states have considerable latitude to design their Medicaid systems. For instance, under Section 1915(c) waiver programs, states can provide a large range of services and supports, limit those services to specific populations or geographic locations, limit the number of participants, and operate waiting lists. The criteria regarding the level of care needs to be eligible for certain services vary considerably across states. In short, states use diverse provisions and strategies to provide home care. Observed indicators, based on expenditures or number of participants, are the results of those provisions and strategies and can be used to capture states’ generosity. Nevertheless, the choice of indicator(s) is somewhat arbitrary, subject to measurement errors, and gives a partial view of generosity.

Many studies rely on one indicator to capture states’ Medicaid home care policy generosity. Kemper et al. (2008) suggest decomposing Medicaid home care expenditures into number of users and expenditures per user, as a way to capture separately participation in Medicaid home care programs and intensity of use by participants. This approach is practical as well as conceptually appealing. Conditional on financial resources, a state may want to serve more people with lower level of home care services or provide a comparatively smaller share of its population with higher level of care, depending on the goals of the expansion. For example, serving more people may achieve the goal of reducing unmet need in the community among individuals with less severe needs. Increasing spending per person, however, may be a better strategy to reduce institutionalization among individuals with higher level of needs. These two dimensions of state Medicaid home care policy, Participation and Intensity, are seldom explicitly considered in prior studies, although they may evolve differently over time, have different implications for beneficiaries, and affect differently health care use and health outcomes.

The objectives of this study are twofold: to measure the Participation and Intensity dimensions of state Medicaid home care policy comprehensively as latent variables and to describe Medicaid home care policy generosity over time and across states based on those latent variables. Compared with observed indicators, such latent constructs have three main advantages: they provide more comprehensive measures of state Medicaid home care Participation and Intensity, overcome measurement errors inherent to observed indicators, and prevent the subjective selection of specific indicators as proxies for state Medicaid home care policy.

Using factor analysis, we develop a measurement model of state Medicaid home care policy in which Participation and Intensity are two latent variables. In such measurement models, observed correlated indicators are assumed to be reflections of the same underlying latent variable—that is, dimension of state Medicaid home care policy. Figure 1 illustrates the measurement model where the Participation and Intensity dimensions (in ellipses) are measured by three indicators each (in rectangles). The arrows pointing at the indicators from the right-hand side represent indicator-specific measurement errors—that is, this measurement strategy takes measurement errors explicitly into account. The two latent dimensions may be correlated, as indicated by the curved double-sided arrow between them. Starting with all available indicators that potentially reflect states’ Medicaid home care Participation and Intensity, factor analysis helps select a final set of indicators that best measure each dimension. The selection is based on the strength of the relationships between the indicators and the latent dimensions, preventing arbitrary selection of indicators. Thus, states’ Medicaid home care Participation and Intensity are measured comprehensively by the best set of available indicators.

Figure 1.

Model of state Medicaid home care Participation and Intensity: confirmatory factor analysis.

In addition, we consider the types of home care services, regardless of the specific Medicaid mechanisms under which they are provided; that is, state plans or waivers/demonstrations. This approach differs from what has been done in the literature so far and may provide new insights, as looking at a specific mechanism in isolation may miss an important proportion of home care received (Konetzka et al., 2012).

Method

Data

We draw data from the Medicaid Statistical Information System (MSIS) State Summary Datamarts. The datamarts provide a state-level aggregation of eligibility and claims data submitted by states to the Centers for Medicare and Medicaid Services (CMS) through MSIS—that is, aggregation of Medicaid Analytic eXtract (MAX) data. We use yearly data for the 51 states from 1999 to 2012 (including the District of Columbia). There are 685 observations (two states are missing in 2011 and 27 states in 2012).

We consider 14 indicators available in MSIS that potentially capture the Participation and Intensity dimensions of state Medicaid home care policy (Table 1). We distinguish between the two main types of home care: home health and personal care. We include not only those services provided under states’ mandatory Home Health Plans and optional Personal Care Plans but also home health and personal care services provided under Sections 1915(c), 1915(i), and 1915(j) waivers and demonstrations, among others. Below, total home care (HC) refers to the sum of home health (HH) and personal care (PC). Long-term care (LTC) includes home care and nursing home care. Potential indicators of state Medicaid home care Participation are home health, personal care, and total home care users per capita, as well as total home care users per LTC user. Potential indicators of state Medicaid home care Intensity are home health, personal care, and total home care expenditures per user. We construct analogous indicators for persons 65 years and older (65+) because they represent more than 50% of Medicaid LTSS users (Reaves, 2013). A person who receives both home health and personal care services in any given year is intentionally counted twice in the home care users total. Expenditures are adjusted for inflation and purchasing power differences between states, using the implicit regional price deflator (Bureau of Economic Analysis [BEA], 2014).

Table 1.

Summary Statistics of Available Indicators of State Medicaid Home Care Participation and Intensity.

	M	SD	Minimum	Maximum
Potential indicators of Medicaid home care Participation (%)
HH users per capita	0.33	0.37	0.00^a	2.13
			(WA, 2009)	(MN, 2011)
HH users per 65+ persons	0.68	0.88	0.00^a	4.66
			(WA, 2009)	(D.C., 2011)
PC users per capita	0.23	0.28	0.00^b	1.34
				(WA, 2005)
PC users per 65+ persons	0.92	1.29	0.00^b	6.78
				(CA, 2009)
HC users per capita	0.56	0.47	0.00^a	2.96
			(TN, 2011)	(MN, 2011)
HC users per 65+ persons	1.54	1.50	0.00^a	8.88
			(TN, 2011)	(D.C., 2011)
HC users per LTC user	44.03	18.82	0.28	88.49
			(TN, 2011)	(AK, 2012)
HC users per 65+ LTC user	29.82	17.96	2.39	72.72
			(HI, 1999)	(D.C., 2011)
Potential indicators of Medicaid home care Intensity (US$)
HH expenditures per user	4,667.04	5,444.98	58.71	37,720.02
			(WA, 2008)	(TN, 2009)
HH expenditures per 65+ user	4,391.74	4,904.50	55.21	37,276.33
			(WA, 2008)	(TN, 2007)
PC expenditures per user	6,568.85	7,378.90	0.00^b	54,496.00
				(KS, 2005)
PC expenditures per 65+ user	5,341.30	5,912.55	0.00^b	28,755.48
				(NH, 2007)
HC expenditures per user	6,648.40	5,236.16	100.62	37,720.02
			(AL, 1999)	(TN, 2009)
HC expenditures per 65+ user	6,382.37	4,830.79	99.75	37,276.33
			(TN, 2012)	(TN, 2007)

Note. State and year in parentheses. HC = HH + PC; LTC = HC + nursing home care. HH = home health care; PC = personal care; HC = total home care; LTC = long-term care.

Rounds to zero.

Various state–year observations, as some states do not provide PC.

Our focus is Medicaid home care, distinguishing between the main types of home care services, home health, and personal care. Nevertheless, at least in some states, most home health users are short-term users. Thus, it may not be fully appropriate to consider home health services to be LTC. With this in mind, we conduct a sensitivity analysis excluding Medicaid home health.

MSIS is the only source of Medicaid data rich enough to provide all the indicators considered. As happens with any Medicaid data source, it has some limitations. We employed a number of strategies to guarantee the quality of the data.¹ Overall, our measurement model of Medicaid home care Participation and Intensity takes measurement errors in each indicator explicitly into account. This is an advantage of measuring states’ generosity using latent variables that helps limit potential issues with the data. As some states do not report data for managed LTSS programs through MSIS, we conduct a sensitivity check, where we reproduce our analyses on the subset of states without managed LTSS programs.² We conduct two additional sensitivity checks, where we exclude one state or one year of data at a time to see whether it has an impact on our measurement model of state Medicaid home care Participation and Intensity.

Table 1 shows the summary statistics of all available indicators. Overall, there are wide variations over time and across states. For example, nearly 0% of Tennessee’s population in 2011 uses Medicaid home care services, compared with 2.96% in Minnesota in 2011. Medicaid home care expenditures per user vary between US$101 (Alabama, 1999) and US$37,720 (Tennessee, 2009).

Factor Analyses

We conduct exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) (Jöreskog, 1969; Spearman, 1904). In EFA, all possible relationships between the latent dimensions and available indicators are estimated; that is, no structure is imposed. In CFA, only the relevant relationships are taken into account, based on EFA results. Those relationships are measured by coefficients called “factor loadings” or simply “loadings.” Indicators with loadings above 0.7 are typically considered relevant measures of one latent dimension. Smaller loadings mean that the latent dimension accounts for less than 50% of the variation in the indicator (0.7²). As a preliminary step, one typically starts with EFA of all available indicators. EFA shows whether the two dimensions we are interested in—Participation and Intensity—truly exist and are reflected by the available indicators. CFA is then used to validate and evaluate the measurement model established in our conceptual framework and initially informed by EFA results (Brown, 2006). It is considered good practice to conduct the EFA and CFA using distinct samples, sometimes called the derivation and validation samples (e.g., Moyo, Huang, Simoni-Wastila, & Harrington, 2018). For this purpose, we split the sample in two (twenty-five/six states each).

The goal is to achieve a final CFA model with good fit. Therefore, first, we estimate a CFA model using all indicators with loadings above 0.7 in the EFA, and then, we drop indicators until we achieve a good fitting model. Once the best CFA model is obtained in the validation sample (25 states), it is applied to all 51 states. This is a way to verify external consistency. Relying on current best practice, the following indices are used to evaluate the CFA model fit: chi-square statistic, root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), Tucker–Lewis index (TLI), and comparative fit index (CFI).³ The full information maximum likelihood estimator is used, as it relies on all available information without deleting observations with missing values. Standard errors are clustered at the state level.

Finally, to assess internal consistency of the final latent constructs, we use Cronbach’s alpha (Cronbach, 1951). Internal consistency is considered good if Cronbach’s alpha exceeds 0.8. The final optimal CFA model is used to estimate the values of the latent dimensions for each state in each year—that is, factor scores. These values are used to classify states by their levels of generosity in Medicaid home care Participation and Intensity; factor scores have no numeric meaning per se.

Results

Measurement of State Medicaid Home Care Participation and Intensity

EFA results, based on the first half of the states and all 14 indicators considered, are reported in Table 2. As expected, we get two latent dimensions that correspond to the Participation and Intensity dimensions of state Medicaid home care policy, based on the indicators that reflect them. The Participation dimension is reflected by five indicators with large loadings, above 0.7. They are personal care users per capita, personal care users per 65+ persons, home care users per 65+ persons, home care users per LTC user, and home care users per 65+ LTC user. The Intensity dimension is reflected by four indicators: home health expenditures per user, home health expenditures per 65+ user, home care expenditures per user, and home care expenditures per 65+ user. The two dimensions are independent, as indicated by a nonsignificant correlation.

Table 2.

Exploratory Factor Analysis Results (Unstructured Model): Factor Loadings and Dimensions.

	Participation dimension	Intensity dimension
HH users per capita	0.040	−0.109
HH users per 65+ persons	−0.026	−0.072
PC users per capita	0.941*	−0.004
PC users per 65+ persons	0.938*	0.047
HC users per capita	0.646*	−0.089
HC users per 65+ persons	0.855*	−0.001
HC users per LTC user	0.715*	0.151
HC users per 65+ LTC user	0.781*	−0.174
HH expenditures per user	−0.351*	0.872*
HH expenditures per 65+ user	−0.436*	0.821*
PC expenditures per user	0.367*	0.083
PC expenditures per 65+ user	0.601*	0.209
HC expenditures per user	0.014	0.980*
HC expenditures per 65+ user	0.132	0.855*

Note. Sample includes half the states. HC = HH + PC; LTC = HC + nursing home care. HC = total home care; HH = home health care; PC = personal care; LTC = long-term care.

p < 0.05. Geomin-rotated loadings.

To conduct CFA on the second half of the states, we start with the nine indicators with loadings above 0.7 in the EFA. In subsequent CFA, we drop three additional indicators until we achieve the best fitting CFA model. This best fitting CFA model includes three indicators per dimension. It is applied to the entire country; the results are presented in Figure 1. The model has good or at least reasonable fit, depending on the criterion—for example, RMSEA = 0.05, TLI = 0.94. Interpreting the loadings, standardized to lie between zero and one, is straightforward: the Participation dimension accounts for 67% of the variation in personal care users per capita (0.82²), 88% of the variation in home care users per 65+ persons (0.94²), and 67% of the variation in home care users per LTC user (0.82²), similarly for Intensity. As suggested by the EFA results, the correlation between the two dimensions is set to zero, that is, they are independent. Cronbach’s alpha takes values 0.89 for the Participation dimension and 0.94 for the Intensity dimension; that is, the model has good internal consistency.

Trends and Geographic Variations in State Medicaid Home Care Participation and Intensity

The measurement model in Figure 1 is used to estimate the values of state Medicaid home care Participation and Intensity. These values are standardized to have mean zero over all 685 state–year observations. Thus, a negative value indicates that, in the year considered, the state has a level of generosity (in the Participation or Intensity dimension) below the 1999-2012 sample mean, the opposite for a positive value. Globally, if generosity steadily increases over time, negative values would be concentrated in the earlier years of the sample. The estimated levels of generosity in each dimension in 1999 and 2012 are reported in Table A1 in the appendix. When generosity is higher in 2012 than in 1999, the 2012 values are highlighted in gray. Generosity in the Participation dimension of state Medicaid home care policy increased in 34 of the 51 states between 1999 and 2012. Generosity in the Intensity dimension increased in 40 states during the same period. Overall, 29 states expanded both Medicaid home care Participation and Intensity over the 14 years, whereas six states reduced generosity in both dimensions. A trade-off seems to have occurred in five states that expanded Participation but reduced Intensity (e.g., Tennessee, Texas) and in 11 states that expanded Intensity but reduced Participation (e.g., Oregon, West Virginia).

To illustrate geographic patterns in addition to the time trends, Figures 2 and 3 map the different levels of generosity in state Medicaid home care Participation and Intensity in 1999 and 2012. We group states into quintiles, based on the distribution of each latent construct (see Table A1 in the appendix for the actual estimated values). In the maps, darker shades of gray represent higher quintiles of the distribution of generosity, based on the entire 1999 to 2012 sample. Globally, for Participation, generosity increased in many states during the 1999 to 2012 period, but the increases are not large enough to be clearly visible. Still, the 2012 map has a few more states with a darker shade of gray than the 1999 one (e.g., Michigan, Washington; Figure 2). In 2012, there is limited geographic clustering in the Participation dimension of state Medicaid home care policy. The 10 states in the top quintile are distributed across all the U.S. census regions (e.g., California, Michigan, New York, Oklahoma). The same happens with the 12 states in the bottom quintile, which include, for example, Arizona, Florida, Kansas, and New Hampshire.

Figure 2.

Maps of generosity in state Medicaid home care Participation in 1999 and 2012.

Figure 3.

Maps of generosity in state Medicaid home care Intensity in 1999 and 2012.

The increases in state Medicaid home care Intensity are stronger (Figure 3). For the Intensity dimension, the 2012 map has several more states with a darker shade of gray than the 1999 one (e.g., Rhode Island, Washington). Turning to geographic clusters, in 2012, the highest levels of generosity in the Intensity dimension are found in the Northeast and in the West. In these two regions, most states are in the two upper quintiles. In contrast, most Southern states have comparatively low Medicaid home care Intensity.

In sum, although generosity in both dimensions of Medicaid home care policy increased in most states between 1999 and 2012, the expansions in Intensity were much more pronounced. This result indicates that states have increased home care use among those with access to Medicaid home care services. The Northeast and the West are easily identified as generous regions in the Intensity dimension in 2012. States with comparatively high or low levels of generosity in the Participation dimension are spread out across all regions.

Comparison of States’ Generosity Based on the Latent Constructs and Observed Indicators

To show the importance of measuring state Medicaid home care Participation and Intensity comprehensively, we contrast the rankings of states by quintiles of generosity in each latent dimension to the rankings resulting from two observed indicators found in the literature: percentage of Medicaid LTSS beneficiaries who received HCBS (Eiken, 2016) and total Medicaid HCBS expenditures per participant (Ng et al., 2016). As a reminder, home care is the main type of HCBS. These comparisons are presented in Table 3. Differences in states’ generosity of three or four quintiles between the latent dimension and the observed indicator are highlighted in gray. In some cases, whether we use the comprehensive latent measure or the observed indicator gives a very different perception of states’ generosity. For example, there are striking differences in Participation in Oklahoma and in Intensity in Tennessee and New Mexico. Overall, the correlation between the quintile groupings of states by each latent construct and observed indicator is only 0.34 for Participation and 0.28 for Intensity.

Table 3.

Comparison of States’ Generosity in the Participation and Intensity Latent Dimensions and Observed Indicators (Quintiles).

States	Participation dimension^a	Home care users per LTSS user^b	States	Intensity dimension^a	Home care expenditures per user^b
Arizona	First	NA	Texas	First	First
Wisconsin	First	NA	Vermont	First	First
Tennessee	First	NA	Mississippi	First	First
Kansas	First	NA	Georgia	First	Second
Delaware	First	First	Kentucky	First	Second
Indiana	First	First	Kansas	First	Third
Utah	First	Second	Maine	First	Third
Georgia	First	Second	Arizona	First	Fourth
Florida	First	Second	Alabama	First	Fourth
Pennsylvania	First	Third	Tennessee	First	Fifth
Wyoming	First	Third	New Mexico	First	Fifth
Maine	Second	NA	North Carolina	Second	First
Mississippi	Second	First	Michigan	Second	First
Kentucky	Second	First	Arkansas	Second	First
Nebraska	Second	Second	Oklahoma	Second	First
Ohio	Second	Second	Iowa	Second	Second
New Hampshire	Second	Second	South Dakota	Second	Second
North Dakota	Second	Third	Utah	Second	Third
Illinois	Second	Fourth	New Hampshire	Second	Fourth
Colorado	Second	Fifth	Wisconsin	Second	Fourth
Oregon	Second	Fifth	Delaware	Second	Fifth
New Mexico	Third	NA	Missouri	Third	First
Maryland	Third	First	California	Third	First
South Dakota	Third	First	Ohio	Third	Second
Virginia	Third	Second	Nebraska	Third	Third
West Virginia	Third	Third	New Jersey	Third	Third
Iowa	Third	Fourth	Oregon	Third	Third
Hawaii	Third	Fourth	Wyoming	Third	Third
Montana	Third	Fourth	District of Columbia	Third	Fourth
Nevada	Third	Fourth	Minnesota	Third	Fourth
Idaho	Third	Fifth	West Virginia	Third	Fifth
Texas	Fourth	NA	Idaho	Fourth	First
New Jersey	Fourth	NA	Rhode Island	Fourth	First
Alabama	Fourth	NA	Illinois	Fourth	Second
Louisiana	Fourth	First	Nevada	Fourth	Second
Arkansas	Fourth	First	Florida	Fourth	Second
South Carolina	Fourth	Third	Louisiana	Fourth	Second
Massachusetts	Fourth	Third	South Carolina	Fourth	Third
Rhode Island	Fourth	Third	Montana	Fourth	Third
Vermont	Fourth	Fifth	Connecticut	Fourth	Fourth
Washington	Fourth	Fifth	Pennsylvania	Fourth	Fifth
Oklahoma	Fifth	First	Colorado	Fifth	Second
Connecticut	Fifth	Second	Washington	Fifth	Third
Michigan	Fifth	Third	Massachusetts	Fifth	Fourth
New York	Fifth	Third	Indiana	Fifth	Fourth
District of Columbia	Fifth	Fourth	Hawaii	Fifth	Fourth
Missouri	Fifth	Fourth	North Dakota	Fifth	Fifth
North Carolina	Fifth	Fourth	Virginia	Fifth	Fifth
Alaska	Fifth	Fifth	New York	Fifth	Fifth
Minnesota	Fifth	Fifth	Alaska	Fifth	Fifth
California	Fifth	Fifth	Maryland	Fifth	Fifth
Correlation	0.34		Correlation	0.28

Note. Gray cells highlight states with quintile differences of three or more between the latent measure and the observed indicator. Higher quintiles mean higher generosity. LTSS = long-term services and supports; HCBS = home- and community-based services.

2012 or last year available.

Percentage of Medicaid LTSS beneficiaries who received HCBS, 2012 (source: Eiken, 2016) and Total Medicaid HCBS expenditures per participant served, 2012 (source: Ng et al., 2016).

Sensitivity Analyses

As discussed, we cannot distinguish between Medicaid home health services provided for rehabilitation or chronic needs; thus, it may not be appropriate to consider all Medicaid home health services to be LTC. To explore whether the Participation and Intensity dimensions still emerge when excluding home health care, we conduct EFA and CFA on indicators of personal care only, namely, personal care users per capita/per 65+ persons, personal care users per LTC user, and personal care expenditures per (65+) user. Here, LTC users include only personal care and nursing home care users. Results reveal that the two dimensions do emerge and each one is reflected by the expected indicators (Figure A1 in the appendix). Interestingly, in this case, the two dimensions are positively correlated, suggesting a tendency of states to expand generosity in both dimensions simultaneously. The measurement model of Medicaid personal care generosity has good fit and good internal validity as measured by Cronbach’s alpha (0.96 for Participation, 0.87 for Intensity).

As mentioned above, some states provide LTSS under managed care programs, but not all report managed LTSS data through MSIS. In our next sensitivity check, we investigate whether excluding states with managed LTSS programs affects our measurement model and the resulting ranking of states by their generosity in the Participation or Intensity latent dimensions. Globally, our model is robust to the exclusion of those states, with most loadings remaining virtually the same. All model fit criteria are about the same values as before (Figure A2 in the appendix). We also use this model, estimated on the restricted sample, to predict generosity. The state rankings by their level of generosity in any given year are practically the same (see, for example, 2010 in Table A2 in the appendix).

In our last sensitivity check, we exclude one state or one year of data at a time to see how the estimated factor loadings change. In this way, we can investigate how sensitive the measurement model may be to potential outlier states/years. The results are presented in Figures A3 and A4 in the appendix. Only one factor loading—the one associated with home health expenditures per 65+ user—varies by more than +0.05 or −0.05 from the median value of all 51 estimates, when Alaska or Tennessee are excluded, respectively (Figure A3). Overall, the measurement model is robust to the exclusion of each state. The exclusion of any year of data has very limited impact on the estimated factor loadings (Figure A4).

Discussion and Conclusion

This study measures comprehensively the Participation and Intensity dimensions of state Medicaid home care policy and uses those comprehensive measures to describe Medicaid home care policy generosity over time and across states.

Our results suggest that the Participation and Intensity dimensions exist and are independent. This independence is consistent with the fact that there are not only states that increased or decreased both Participation and Intensity over time but also states that increased generosity in one dimension at the expense of the other.

Overall, in the United States, the main trend has been toward an increase in generosity in both dimensions (29 states), with increases in Intensity being much more pronounced. The rebalancing of the LTSS market toward home care may help explain the expansions in Medicaid home care Intensity, as more users with high levels of care needs may move to or stay in the community. For example, over the sample period, additional states initiated Personal Care Plans and started to provide personal care under different waivers and demonstrations. This may explain the increase in Intensity, as personal care services are usually provided over a longer period of time than home health services—that is, personal care tends to be more intensive. The larger numbers of waiver programs may also translate into more generous eligibility criteria and, consequently, higher Participation. In most states, home care policy generosity has continued to increase after 2012, as a result of the ACA provisions. Ongoing efforts to redirect LTSS toward HCBS and expansions in Medicaid eligibility resulted in further increases in Medicaid home care Intensity and Participation.

This study also shows that using the comprehensive latent measures or observed indicators gives different impressions of states’ Medicaid home care policy generosity. The choice of one indicator to capture generosity in a particular dimension is very important and may have a large impact on results.

Our measurement strategy requires a rich data set with many indicators. In addition, results based on latent constructs may sometimes be difficult to communicate, because latent variables have no inherent scale. Yet, even when using observed indicators is preferred, factor analysis can help motivate their selection by revealing which indicators have the largest loadings, that is, the best at capturing the underlying latent Medicaid home care policy dimension.

In possible extensions to our approach, factor analysis has a variety of potential applications in this field. For instance, it can be used to model home care policy in other countries where such policy is also decentralized, such as Canada or Switzerland (e.g., Gonçalves & Weaver, 2017; Stabile, Laporte, & Coyte, 2006). It may also be applied to measure other policies, such as nursing home policy. Future research aiming to explain variations in Medicaid home care policy generosity across states and over time may rely on structural equation models (SEMs) to measure such policy as latent variables. For example, changes in participants’ case mix may help explain variations in Medicaid home care Intensity. It is possible that by expanding generosity in the Participation dimension, some states start to serve participants with lower level of needs, bringing down the average intensity of services. Such reductions in intensity would not correspond to lower generosity of the state in question. Similarly, programs such as Money Follows the Person, in which the goal is to move nursing home residents into the community with home health care and other supports, may increase the average intensity of home care in a way that does not correspond to higher intentional generosity (e.g., Robison, Porter, Shugrue, Kleppinger, & Lambert, 2015).

Although progress has been made, there are still important gaps in home care and HCBS data, which limit the ability to evaluate policy and ultimately design evidence-based policy (Newquist, DeLiema, & Wilber, 2015). Our article has two main implications for the field. The first is conceptual. Policy and research discussions often refer to HCBS expansion as a unified concept. Yet, separating and contrasting the Participation and Intensity dimensions of state Medicaid home care policy is important, as expansion along these two dimensions represents different strategies and potentially different implications for beneficiaries. Increasing the Participation and/or Intensity of Medicaid home care programs is an important policy decision made by states that entails trade-offs. It is important for state policy makers to be aware of the decisions that other states have made about this trade-off and to assess the two dimensions explicitly. It is also important for researchers to monitor Medicaid home care Participation and Intensity across the U.S. states and over time. For example, such analyses may be relevant to inform whether the rebalancing of LTSS toward HCBS achieves its objectives. Second, our results have implications for measurement and future empirical research. Our approach improves measurement of states’ generosity of Medicaid-funded home care, a dominant part of the home care industry, demonstrating that appropriate measurement makes a difference. After several decades of expansion of HCBS by individual state Medicaid programs, there is still remarkably little evidence of the effects of these expansions on costs, quality, and outcomes. As researchers embark on filling this gap, mismeasurement of the policies could lead to incorrect conclusions. The measurement strategy presented in this study is a useful tool for these critical analyses and their input into evidence-based policy.

Footnotes

Appendix

Table A2.

Sensitivity Check: Exclusion of States With Managed LTSS Programs—State Rankings by Increasing Generosity in Medicaid Home Care Participation and Intensity in 2010.

States	Participation		States	Intensity
States	All states	States with managed LTSS programs	States	All states	States with managed LTSS programs
Wisconsin	1	—	Alabama	1	1
Arizona	2	—	Georgia	2	2
Georgia	3	1	New Mexico	3	—
Florida	4	—	Oregon	4	3
Wyoming	5	2	Arizona	5	—
Kansas	6	3	Mississippi	6	4
Delaware	7	4	Kentucky	7	5
Indiana	8	5	Arkansas	8	6
Pennsylvania	9	—	Kansas	9	7
Oregon	10	6	Texas	10	—
New Hampshire	11	8	Idaho	11	9
Utah	12	9	Maine	12	8
Maine	13	10	New Hampshire	13	12
North Dakota	14	7	Wyoming	14	10
Tennessee	15	—	Vermont	15	13
Ohio	16	12	Iowa	16	11
Colorado	17	13	Oklahoma	17	14
Illinois	18	11	Michigan	18	—
Mississippi	19	15	Delaware	19	15
Kentucky	20	14	Utah	20	19
Nebraska	21	16	South Dakota	21	16
Hawaii	22	—	North Carolina	22	—
West Virginia	23	17	West Virginia	23	17
Montana	24	18	Missouri	24	20
Virginia	25	20	Minnesota	25	—
South Dakota	26	19	Nebraska	26	18
Maryland	27	21	Ohio	27	21
Nevada	28	23	Rhode Island	28	22
New Mexico	29	—	California	29	—
Vermont	30	22	District of Columbia	30	23
Massachusetts	31	—	New Jersey	31	24
Rhode Island	32	25	Florida	32	—
Louisiana	33	24	Pennsylvania	33	—
South Carolina	34	26	Connecticut	34	26
Washington	35	—	South Carolina	35	25
Texas	36	—	Nevada	36	27
Iowa	37	28	Wisconsin	37	—
New Jersey	38	27	Illinois	38	28
Alabama	39	29	Virginia	39	29
Michigan	40	—	Louisiana	40	30
Oklahoma	41	30	North Dakota	41	31
Minnesota	42	—	Montana	42	32
New York	43	—	New York	43	—
Idaho	44	31	Washington	44	—
Missouri	45	32	Colorado	45	33
Arkansas	46	33	Tennessee	46	—
Connecticut	47	34	Massachusetts	47	—
North Carolina	48	—	Indiana	48	34
Alaska	49	35	Alaska	49	35
California	50	—	Maryland	50	36
District of Columbia	51	36	Hawaii	51	—

Note. 2010 is the last year for which data exist for all states. States ordered by increasing level of generosity. LTSS = long-term services and supports.

Acknowledgements

We are grateful to Peter Kemper for his interest in measuring Medicaid home care policy and Alberto Holly for his methodological suggestions. We appreciate the comments of Eva Cantoni, Jaya Krishnakumar, two anonymous reviewers, and the participants at the 3rd International Conference on Evidence-Based Policy in Long-Term Care. We thank Jeffrey Silverman at CMS for data support.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by the Swiss National Science Foundation (Grant PDFMP1_134899) and the Swiss School of Public Health+. Work was conducted while Judite Gonçalves and France Weaver were at the Geneva School of Economics and Management, University of Geneva.

Notes

ORCID iD

Judite Gonçalves

References

Bureau of Economic Analysis. (2014). Implicit regional price deflator. Retrieved from http://www.bea.gov/iTable/iTableHtml.cfm?reqid=70&step=30&isuri=1&7022=102&7023=8&7024=Non-Industry&7033=-1&7025=0&7026=XX&7027=-1&7001=8102&7028=1&7031=0&7040=-1&7083=Levels&7029=102&7090=70#.V9Z7hbgWFJ4.link

Bentler

P. M.

(1990). Comparative fit indexes in structural models. Psychological Bulletin, 107, 238-246.

Blackburn

Locher

J. L.

Morrisey

M. A.

Becker

D. J.

Kilgore

M. L.

(2016). The effects of state-level expenditures for home- and community-based services on the risk of becoming a long-stay nursing home resident after hip fracture. Osteoporosis International, 27, 953-961.

Borck

Schmitz

Doty

Drabek

(2016, September). Transition rates from the community to nursing home care among older adult Medicaid enrollees, 2006-2009 (Report for the U.S. Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation). Cambridge, MA: Mathematica Policy Research. Retrieved from https://aspe.hhs.gov/basic-report/transition-rates-community-nursing-home-care-among-older-adult-medicaid-enrollees-2006-2009

Brown

T. A.

(2006). Confirmatory factor analysis for applied research. New York, NY: Guilford Press.

Burr

J. A.

Mutchler

J. E.

Warren

J. P.

(2005). State commitment to home- and community-based services. Journal of Aging and Social Policy, 17, 1-18.

Cronbach

L. J.

(1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297-334.

Eiken

(2016, September). Medicaid long-term services and supports beneficiaries in 2012. Centers for Medicare & Medicaid Services. Retrieved from https://www.medicaid.gov/medicaid/ltss/downloads/ltss-beneficiaries-2012.pdf

Eiken

(2017a, September). Medicaid long-term services and supports beneficiaries in 2013. Centers for Medicare & Medicaid Services. Retrieved from https://www.medicaid.gov/medicaid/ltss/downloads/reports-and-evaluations/ltss-beneficiaries-2013.pdf

10.

Eiken

(2017b, September). Medicaid section 1915(c) waiver data based on the CMS 372 Report, 2013-2014. Centers for Medicare & Medicaid Services. Retrieved from https://www.medicaid.gov/medicaid/ltss/downloads/reports-and-evaluations/cms-372-report-2014.pdf

11.

Eiken

Sredl

Burwell

Woodward

(2017, April). Medicaid expenditures for long-term services and supports (LTSS) in FY 2015. Centers for Medicare & Medicaid Services. Retrieved from https://www.medicaid.gov/medicaid/ltss/downloads/reports-and-evaluations/ltssexpendituresffy2015final.pdf

12.

Gardner

Gilleskie

D. B.

(2012). The effects of state Medicaid policies on the dynamic savings patterns and Medicaid enrollment of the elderly. The Journal of Human Resources, 47, 1082-1127.

13.

Gonçalves

Weaver

(2017). Effects of formal home care on hospitalizations and doctor visits. International Journal of Health Economics and Management, 17, 203-233.

14.

Harrington

Laplante

Kaye

H. S.

(2012). Medicaid home- and community-based services: Impact of the Affordable Care Act. Journal of Aging and Social Policy, 24, 169-187.

15.

L.-T.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6, 1-55.

16.

Jöreskog

K. G.

(1969). A general approach to confirmatory maximum likelihood factor analysis. Psychometrika, 34, 183-202.

17.

Kemper

Weaver

Short

P. F.

Shea

Kang

(2008). Meeting the need for personal care among the elderly: Does Medicaid home care spending matter? Health Services Research, 43, 344-362.

18.

Kenney

Rajan

(2000). Understanding dual enrollees’ use of Medicare home health services: The effects of differences in Medicaid home care programs. Medical Care, 38, 90-98.

19.

Kitchener

Harrington

(2007a). Medicaid home- and community-based services for the elderly: Trends in programs and policies. Journal of Applied Gerontology, 26, 305-324.

20.

Kitchener

Harrington

(2007b). Medicaid state plan personal care services: Trends in programs and policies. Journal of Aging and Social Policy, 19, 9-26.

21.

Konetzka

R. T.

(2014). The hidden costs of rebalancing long-term care. Health Services Research, 49, 771-777.

22.

Konetzka

R. T.

Karon

S. L.

Potter

D. E. B.

(2012). Users of Medicaid home- and community-based services are especially vulnerable to costly avoidable hospital admissions. Health Affairs, 31, 1167-1175.

23.

Miller

N. A.

(2011). Relations among home- and community-based services investment and nursing home rates of use for working-age and older adults: A state-level analysis. American Journal of Public Health, 101, 1735-1741.

24.

Mor

Zinn

Gozalo

Feng

Intrator

Grabowski

D. C.

(2007). Prospects for transferring nursing home residents to the community. Health Affairs, 26, 1762-1771.

25.

Moyo

Huang

T.-Y.

Simoni-Wastila

Harrington

(2018). Exploratory and confirmatory factor analyses of delirium symptoms in a sample of nursing home residents. Journal of Applied Gerontology, 37, 228-255. doi:10.1177/0733464816633859

26.

Muramatsu

Campbell

R. T.

(2002). State expenditures on home- and community-based services and use of formal and informal personal assistance: A multilevel analysis. Journal of Health and Social Behavior, 43, 107-124.

27.

Muramatsu

Hoyem

R. L.

Yin

Campbell

R. T.

(2008). Place of death among older Americans: Does state spending on home- and community-based services promote home death? Medical Care, 46, 829-838.

28.

Muramatsu

Yin

Campbell

R. T.

Hoyem

R. L.

Jacob

M. A.

Ross

C. O.

(2007). Risk of nursing home admission among older Americans: Does states’ spending on home- and community-based services matter? Journal of Gerontology: Social Sciences, 62B, 169-178.

29.

Muramatsu

Yin

Hedeker

(2010). Functional declines, social support, and mental health in the elderly: Does living in a state supportive of home- and community-based services make a difference? Social Science and Medicine, 70, 1050-1058.

30.

Newquist

D. D.

DeLiema

Wilber

K. H.

(2015). Beware of data gaps in home care research: The streetlight effect and its implications for policy making on long-term services and supports. Medical Care Research and Review, 72, 622-640.

31.

Harrington

Musumeci

(2011, October). State options that expand access to Medicaid home- and community-based services. The Henry Kaiser Family Foundation. Retrieved from https://kaiserfamilyfoundation.files.wordpress.com/2013/01/8241.pdf

32.

Harrington

Musumeci

Ubri

(2016, October). Medicaid home- and community-based services programs: 2013 data update. The Henry Kaiser Family Foundation. Retrieved from http://files.kff.org/attachment/Report-Medicaid-Home-and-Community-Based-Services-Programs-2013-Data-Update

33.

Pezzin

L. E.

Kasper

J. D.

(2002). Medicaid enrollment among elderly Medicare beneficiaries: Individual determinants, effects of state policy, and impact on service use. Health Services Research, 37, 827-847.

34.

Rahman

Tyler

Thomas

K. S.

Grabowski

D. C.

Mor

(2015). Higher Medicare SNF care utilization by dual-eligible beneficiaries: Can Medicaid long-term care policies be the answer? Health Services Research, 50, 161-179.

35.

Reaves

E. L.

(2013, July). Long term services and supports (LTSS): Medicaid’s role and options for states. The Henry Kaiser Family Foundation. Retrieved from https://www.nhpf.org/uploads/Handouts/Reaves-slides_07-19-13.pdf

36.

Rice

J. B.

Kasper

J. D.

Pezzin

L. E.

(2009). A comparative analysis of Medicaid long-term care policies and their effects on elderly dual enrollees. Health Economics, 18, 275-290.

37.

Robison

Porter

Shugrue

Kleppinger

Lambert

(2015). Connecticut’s “Money follows the person” yields positive results for transitioning people out of institutions. Health Affairs, 34, 1628-1636.

38.

Saucier

Kasten

Burwell

Gold

(2012, July). The growth of Managed Long-Term Services and Supports (MLTSS) programs: A 2012 update. Centers for Medicare & Medicaid Services. Retrieved from https://www.medicaid.gov/medicaid-chip-program-information/by-topics/delivery-systems/downloads/mltssp_white_paper_combined.pdf

39.

Spearman

(1904). “General intelligence,” objectively determined and measured. The American Journal of Psychology, 15, 201-292.

40.

Stabile

Laporte

Coyte

P. C.

(2006). Household responses to public home care programs. Journal of Health Economics, 25, 674-701.

41.

Thomas

K. S.

Mor

(2013). The relationship between older Americans Act Title III state expenditures and prevalence of low-care nursing home residents. Health Services Research, 48, 1215-1226.

42.

Wenzlow

Eiken

Sredl

(2016, June). Improving the balance: The evolution of Medicaid expenditures for long-term services and supports (LTSS), FY 1981-2014. Centers for Medicare & Medicaid Services. Retrieved from https://www.medicaid.gov/medicaid/ltss/downloads/evolution-ltss-expenditures.pdf