Submitted Article The Impact of Financial Incentives on the Composition of Long-term Care in Norway

Transcription

1 Applied Economic Perspectives and Policy (2012) volume 34, number 2, pp doi: /aepp/pps021 Submitted Article The Impact of Financial Incentives on the Composition of Long-term Care in Norway Henning Øien*, Martin Karlsson, and Tor Iversen Henning Øien is a PhD candidate in the Department of Health Management and Health Economics, University of Oslo, Norway; Martin Karlsson is Assistant Professor, Technische Universität Darmstadt, Germany, and is Research Fellow, Health Economics Research Programme, University of Oslo; and Tor Iversen is Professor in Health Economics, Department of Health Management and Health Economics, and Director of Research, Health Economics Research Programme, University of Oslo. *Correspondence to be sent to: henning.oien@medisin.uio.no. Submitted 15 October 2011; accepted 10 February Abstract In Norway, municipalities have economic incentives for choosing residential care in nursing homes for high-income clients and home-based care for low-income clients. Using a three-year panel, , on 427 municipalities we provide an analysis of the effect of the Norwegian long-term care (LTC) financing system on the composition of LTC services at the municipality level. Our main result is that the composition of service is determined by local government revenue and local need for LTC services. We cannot identify an effect of average income among older people in a municipality regarding the balance between home-based care and nursing home care. Hence, the results do not provide evidence of a service distortion. Key words: care setting, financial incentives, long-term care, local authorities. JEL codes: H51, I18. Introduction How to best fund and provide long-term care (LTC) services is a policy issue that receives ever-increasing attention from policy-makers and scholars. The apparent reason is that the aging global population is expected to increase demand and expenditures for LTC, thereby raising concerns of the sector s financial sustainability and its ability to provide high-quality services to meet the increasing demand. A much-discussed issue in the funding debate is the presence of unfortunate incentives in LTC funding schemes that may distort the optimal balance of services. According to Wittenberg, # The Author(s) Published by Oxford University Press, on behalf of Agricultural and Applied Economics Association. All rights reserved. For permissions, please journals.permissions@oup.com. Martin Karlsson was not an editor of this paper and did not take part in any decisions regarding the manuscript. 258

2 The Impact of Financial Incentives on the Composition of Long-term Care in Norway Sandhu, and Knapp (2003), perverse incentives is a pervasive issue in LTC financing. Perverse incentives for both suppliers and users of LTC services may arise when one has a system in which the responsibility for funding is divided among several parties. This can create incentives for cost shifting, and assessing the need and provision of services may not be based on the full resource cost. This can potentially create large inefficiencies. The purpose of this paper is to describe and analyze the effects of such a system. We analyze the determinants of LTC service composition in Norway and explore whether the incentives created by the financial system have unintended consequences. Norway s LTC system is mainly publicly provided and financed, with the funding and provision responsibilities decentralized to the lowest government level, the municipalities. However, the financial system distorts the incentives of the municipalities and the users because the system differentiates between home-based and institutional services in two ways: 1) The national insurance scheme (NIS) covers a larger part of the costs when the services are provided as home-based services; and 2) the users of institutional care pay a user charge that is progressively dependent on their current incomes. Institutional care is comprised predominately of nursing homes that provide around-the-clock health and care services. The service is strictly regulated and includes board and lodging, as well as physician, physiotherapy, and dental services. Also included in the service are pharmaceuticals and user fees for specialized health care services. The user charge is approximately 80% of the useŕs income, constrained upwards by the actual cost of the service. Except for the user charge, municipalities have complete funding responsibility for the institutional service. Home-based services are home nursing and help provided to dependent persons living in their own homes or in community housing. Home nursing is provided free of charge, while home help is subject to a user fee that can be dependent on income. The user fee is constrained by the actual cost of the service and by requiring that the user have adequate disposable income for other expenses. Residents in community housing pay rent, which is often equal to the actual costs of the housing. Health care and medication expenses are covered by the national insurance scheme; that is, the users have a co-payment and the residual is covered by national insurance. This system, however, encourages cost shifting at the boundary between institutional and home-based care. That is, the decisionmaker (the municipality) has incentives to reduce its own costs by choosing a service composition that shifts costs to other public agencies (the NIS/ central government) or the users themselves (Norton, Lindrooth, and Dickey 1999). A stylized description follows: A municipality will increase its gross income, controlled for the health status of the recipient, by providing institutional services to high-income individuals, and home-based care to low-income individuals. For the individuals (and their heirs) in need of care, the opposite will be financially beneficial. The result could be a service composition that is cost inefficient and less adapted to the needs of the recipients. In a sector that is expected to face increasing demand, it is important to analyze the effect of funding systems that can cause inefficiencies. This paper does exactly that. To test the hypothesis that the financial system could lead to cost shifting, we examine the determinants of the local LTC service composition. Specifically, we are interested in testing whether the income levels of actual and potential recipients affect LTC service composition. Income level could 259

3 Applied Economic Perspectives and Policy have an effect on service composition, since a high level of income reduces the cost of institutional care relative to home care for municipalities. On the other hand, an increased income increases the recipient s cost of institutional care relative to home care. Hence, the direction of the total effect will depend on the extent to which municipalities take into account the cost for the recipient. We are also interested in how local revenue may affect service composition; this is a way to explore whether affluent municipalities have different LTC compositions than poorer municipalities, despite the intergovernmental grant system aimed at equalizing local service levels. The analysis is conducted using publicly available municipality-level data that contains information on local revenue and the LTC service sector, along with a data set on personal income and taxes i. This provides information on structural characteristics of the municipalities that may affect costs, the number of recipients of institutional and home-based care, and the share of users who have limited and severe dependency levels. The income measure we use is average income of people aged 67 years and older at the municipality level. This measure is chosen partly because of data availability. Hopefully, the measure will capture the variation in income of recipients across municipalities, as more than 90% of institutional care recipients are above 67 years of age. We start with an analysis of the institutional share, that is, institutional recipients as a share of total recipients, as our measure of the service composition. In a regression of the institutional share on the average income level of people aged 67 years and above, as well as municipal revenue, controlling for need and cost of institutional care relative to home-based care, we find that the institutional share is not correlated with the income level of older people, and is positively correlated with municipal revenue. This result indicates that personal income does not distort the institutional share. The positive relationship between municipal revenue and the institutional share suggests that affluent areas have higher institutional coverage. Most likely, the political feasibility of distorting services according to profitability is higher among similar groups of recipients. Therefore, we also regress the abovementioned model using two other dependent variables: the institutional share for the oldest people (those aged 80 years and above who receive care services); and institutional recipients as a share of all recipients receiving around-the-clock care (this also includes older people in community housing receiving 24-hour care services). This is because institutional care is a 24-hour care service predominately for dependent older people. For all of our measures of institutional share, income is insignificant, while municipal revenue per capita is positive and significant. The need and cost factors show the expected signs. Thus, we find no evidence that the income level of older people affects the institutional share. Rather, institutional share is determined by need and local revenue. Richer municipalities are able to provide higher institutional coverage than their poor counterparts. The rest of the paper is organized as follows: to explain the perverse incentives present in the financial system, as well as their possible consequences, we begin the next section with a short description of the governance and division of funding responsibilities in the Norwegian LTC sector. Then, we describe the data and its sources before moving on to i 260

4 The Impact of Financial Incentives on the Composition of Long-term Care in Norway describe the econometrical approach in the following section. The last two sections describe and discuss the results, then offer concluding remarks. The Norwegian context The LTC sector and its funding In Norway, the LTC sector is an integrated part of the extensive social welfare system. An essential feature of this system is that health care and social services are provided, either free or at highly subsidized prices, to every citizen in need, irrespective of socioeconomic status 1. Among the provided services are all health and social services formally characterized as LTC services. Another feature is that the responsibility of financing and providing LTC services is decentralized to the lowest government level the municipalities; their responsibilities are regulated in The Municipal Health Care Act and The Social Service Act (Karlsson, Iversen, and Øien 2012). These acts formalize universal access to LTC services, the extent of public funding and provisions, and the division of responsibility among government levels. The financing regulations differentiate between institutional and home-based services. For institutional services, municipalities have the right to demand an income-dependent user charge, which is approximately 80% of the useŕs income. The user charge covers board and lodging, physician, nursing, and dental services, and pharmaceuticals. Except for the user charge, the municipalities are completely responsible for funding the service. Institutional care consists of nursing homes and residential housing, both of which have staff present at all hours and are primarily used for long-term stays by individuals with severe physical or cognitive impairments. The difference between the two is that residential housing is regulated as a social service and has only the requirement of 24-hour care services, whereas nursing homes are regarded as medical institutions with around-the-clock skilled nursing and care services, and are also used for short-term rehabilitation stays. Home-based services include home nursing and home help provided to dependent persons living in their own homes or in community housing. Home nursing is free of charge, while home help, which assists with instrumental activities of daily living, is subject to a minor user charge that depends on use and may also be dependent on income. Recipients of LTC services living in their own homes or in community housing pay similar user fees for physician services and pharmaceuticals, and cover their own housing expenses just as do non-institutionalized persons. Community houses, on the other hand, are dwellings adapted for persons in need of LTC. The houses range from independent units associated with a few home-based services to collective housing with 24-hour services. Perverse incentives This LTC funding system creates unfortunate incentives that can influence the type of care chosen by municipalities and can lead to a 1 In welfare regime theory, this system/model is often referred to as the social democratic model, the universal model, the encompassing model, or the Nordic model (Karlsson, Iversen, and Øien 2012). The latter refers to the stylized fact that public welfare systems in the Nordic European countries share many of the same attributes. 261

5 Applied Economic Perspectives and Policy Figure 1 Municipalities cost of care net of user charge of institutional and home-based services Source: Own calculations from national accounts and Gabrielsen et al. (2010). composition of services that is cost inefficient and less adapted to users needs and preferences. This is because the progressive user charge applied in institutional care creates adverse incentives among municipalities and recipients. Municipalities with a large share of high-income recipients have an incentive to increase reipients reliance on institutional care, while the opposite is true for municipalities with a high share of low-income recipients. High-income recipients pay higher user charges in nursing homes, and thus have an incentive to demand homebased services in sheltered housing, while low-income recipients realize an economic gain through demanding institutional care; these incentives may create a relationship between recipients income levels and the composition of LTC service. To illustrate this incentive structure, in figure 1 we show how average unit costs of institutional and homebased services vary with recipients income. The calculation of net unit costs is done by a making a simple calculation of average costs per user and subtracting the income-dependent user charge. The unit cost in institutional care is calculated by assuming that it is equal to the municipalities expenses divided by the number of users. The assumption relies on institutional recipients having similar care needs; this is likely since 3/4 of institutional recipients have severe dependency levels and 98% are 50 years and older (Gabrielsen et al. 2010). Since cost-shifting is more likely between similar services, we have calculated the unit cost of care in home-based services by multiplying the cost per hour by the average hours of care received by severely dependent persons aged 50 and older. As we see in figure 1, the net unit cost of institutional services falls progressively with income and becomes less expensive than home-based care for the municipality at incomes higher than 278,

6 The Impact of Financial Incentives on the Composition of Long-term Care in Norway kroner (49,591 USD) ii. As mentioned, the direction of the effect will depend on the extent to which the municipality takes the cost of the user into account, as well as the bargaining position of the recipients and their families in the service-allocation process. If the primary concern of municipalities is their own budget constraints, it will be expected that municipalities with affluent recipients provide relatively more institutional care than municipalities with less affluent recipients. On the other hand, local politicians care about the total utility of recipients in their electorate. Thus, if the income of the user (and the user s heirs) net of user charges is of primary concern, we will expect opposite effects on service composition. The theoretical ambiguity deserves an econometric analysis of the issue. Data and descriptive statistics This analysis uses data from 427 municipalities in Norway covering a three-year period The data are obtained from two sources a database named KOSTRA, which is an abbreviation for municipalitystate-reporting, and tax statistics for individual taxpayers. All data are publicly available and can be found on the home page of Statistics Norway iii. KOSTRA includes information on local accounts, demographics, and services. Since 2007, the database has also incorporated some of the information in IPLOS (i.e. statistics on individuals in need of care, aggregated to the municipality level), which is a national information system that covers the nursing and care sector. The data is based on annual reports from the local authorities to Statistics Norway; the response rate varies from 94 to 100%. Taken together, the data sources contain information on the number of recipients of different LTC services, the number of applicants and recipients of LTC with limited and severe dependency levels, municipality revenue from taxes and intergovernmental grants, and the average income of people aged 67 and older. The dependency levels of LTC recipients and applicants are calculated based on 15 indicators describing the ability to perform activities and instrumental activities of daily living (ADL and IADL). 2 The information on functional abilities is gathered from IPLOS. Statistics Norway calculates a dependency score, and each individual s functional abilities are described as either low, medium to high, or severe. The dependency levels are calculated by first dividing the indicators into five groups: 1) social abilities; 2) cognitive abilities; 3) ability to perform housework; 4) self-care; and 5) the ability to take care of one s own health. Each indicator is given a discrete value between 1 and 5, where 5 is the lowest score. Within each group the mean is calculated for each individual; individuals with a group mean of less than 2 are considered to have a limited dependency level, individuals with a group mean between 2 and 3 a medium to high dependency level, while individuals with a group mean above 3 have a severe dependency level. An unweighted average of the group means is used as a common ii Kroner (NOK) is the Norwegian currency. The annual average of daily spot rates of NOK per 1 USD is equal to 5.61 in 2011, see: iii ssb.no/english. 2 The indicators of ADL include, among others, personal hygiene, dressing and undressing, and eating; IADL indicators include housework, meal preparation, taking medication, and shopping. 263

7 Applied Economic Perspectives and Policy score to indicate dependency level. This unweighted average is categorized in the same way as the group means: an individual with a mean of less than 2 is said to have a limited dependency level, while an individual with a mean higher than 3 is severely dependent (Gabrielsen et al. 2010). The period of analysis is constrained by the implementation of IPLOS in The numbers from 2010 were dropped because information on personal income has not yet been reported and a substantial number of municipalities have not submitted their reports on the service sector for 2010 in a timely fashion. In addition, we dropped from the sample three municipalities that did not have any information on either LTC recipients and/or municipality revenue. 3 Missing observations from a single year are replaced by the mean of the observations from other years within a municipality. As these variables show little variation over time, mean substitution within a municipality is the best guess. Observations that still have missing values are given the population mean. The advantage of population mean substitution is that it keeps the sample mean intact. However, the major drawback is that it deflates the variance and covariance of the variables. All regressions in this paper are repeated with complete cases as a robustness check. The results for the complete cases are given in table 2 in the online appendix. This gives us a balanced panel data set containing 1,281 observations. The descriptive statistics of the variables included in the analysis are shown in table 1. Muncipality revenues are deflated using a price index for municipal consumption and the income variable is deflated by the consumer price index. This is because the compositions of goods purchased by the municipality are different from those of goods purchased by consumers. Life expectancy is measured as the combined average life expectancy of men and women at birth. The density variable measures the number of people per 1,000km 2 in densely populated areas. Densely populated areas are defined as areas with a population of at least 2,000 inhabitants, and where the distance between households is not over 50 meters. Descriptions of the other variables are given in table 1 and further discussion about them appears in the methods section. As we see in the table, the within variation in most of the variables is considerably lower than the between variation. This will constrain our choices of estimation methods, as we will see in the methods section. Determinants of the LTC service composition Empirical strategy To empirically test the possible relationship between recipients income and service composition we hypothesize that an outcome variable, SC it, describing the LTC service composition in municipality i in year t, is determined by the following linear relationship: SC it = b 0 + b 1 INC it + b 2 Y it + X it g + 1 it (1) where INC it is an income measure of potential recipients in municipality i in year t, Y it is unconditional income per capita in municipality i in year t, 3 These municipalities are: Bykle municipality number 941, Torsken municipality number 1928 and Utsira municipality number

8 The Impact of Financial Incentives on the Composition of Long-term Care in Norway Table 1. Summary statistics Variable Mean Standard deviation Overall Between Within Institutional share (IS) Institutional share, population years, (IS80) Average disposable income, population years, 100,000 kr, (INC) Unconditional municipal revenue per capita 100,000 kr, (Y) Percentage of recipients severe dependency level, (SDL) Percentage of recipients limited dependency level, (LDL) Percentage of population (sh6779) Percentage of population (sh8089) Percentage of population (sh90a) Average distance to center, hours (distance) Life expectancy at birth (LE) Percentage living alone population years, (single) Population density in densely populated areas per 1,000 km2, (density) Notes: Number of municipalities: 427. and X it is a vector of control variables that describe the cost differences of providing different LTC services among municipalities. The control variables are described in detail below; 1it is an error term describing all unobserved variations in the dependent variable. The income variable is expected to capture the effect of the progressive user charge that reduces (increases) the price of institutional care relative to home-based care with increasing income for the municipality (recipient). As argued above, the direction of the effect is uncertain. The variable Y it is municipal revenue from sources other than user charges local taxes on income and wealth and intergovernmental block grants. This is used to capture the tendency of high-income municipalities to provide relatively more institutional care compared to poorer municipalities (Borge and Marianne 2005); this will also control for the possible relationship between recipients income and municipal revenue. Higher recipient income will most likely be positively associated with revenue from income taxes; the relationship is uncertain since the effect will be countered by the block grant, which is based partly on tax and spending equalization. The control variables included in vector X it describe differences among municipalities in cost and need of LTC services. The main dependent variable used in the analysis is the institutional share, IS it, defined as institutional recipients as a share of total recipients; 265

9 Applied Economic Perspectives and Policy IS it includes all residents of municipality i that receive an LTC service financed by municipality i. We model IS it to be linearly dependent on the explanatory variables, and pool the data across time and space, which gives us the following empirical model to estimate: IS it = b 0 + b 1 INC it + b 2 Y it + b 3 SDL it + b 4 LDL it + b 5 sh6779 it + b 6 sh8089 it + b 7 sh90a it + b 7 LE it + b 8 Single it + b 9 distance it + b 10 density it + g t + 1 it (i = 1, 2,..., 427; t = 1, 2, 3) where INC is the average income net of taxes from pensions, interest, and wage employment for individuals 67 years and older, and Y it is unconditional revenue per capita measured as the sum of local taxes and block grants. The correlation between INC and Y is The variables are significantly correlated, but too low to make us concerned about collinearity. SDL and LDL are the percentage of recipients with high and limited dependency levels, respectively, sh6779, sh8089, and sh90a are the percentage of the population in the respective age groups 67 to 79, 80 to 89, and 90 years and above. Further, distance it measures the average traveling distance in hours to the center of the municipality, density is the number of residents per 10,000 square kilometers in densely populated areas, and gives us a measure of the population density that will capture the economies of scale of institutional services. Single is the share of people 67 years and above who live alone, gt are dummies for each year t, and 1it is a stochastic disturbance that is assumed to be distributed independently of the explanatory variables. We apply cluster-robust standard errors that allow for general correlation over time for a given municipality. To also account for heteroskedasticity related to the size of the population of the municipalities, we have weighted the above regression model with the square root of the respective municipality population. The chosen estimation method is pooled ordinary least squares (POLS), chosen for the necessity of exploiting variation both within and among municipalities. As seen in table 1, the variation over time for a given municipality is generally low. It is not surprising that variables relating to service composition show little variation over time for a given municipality. For instance, changes in the institutional share require investment decisions to be determined by slow bureaucratic and political processes. Hence, the analysis may be unsuitable for fixed effects estimation, as it would imply a considerable loss of efficiency. The estimation method requires correcting for correlation between error terms in different time periods within municipalities. The sluggishness of the institutional share will most likely cause equicorrelation, i.e. E( 1 it 1 is ) = 0. This causes the POLS estimates of the standard errors to be biased (Woolridge 2003). To correct for this we apply cluster-robust standard errors that allow for general heteroskedasticity and correlation over time for a given municipality (Woolridge 2003). Crucially, POLS relies on the assumption that municipality-fixed effects that are captured by the error term are uncorrelated with the explanatory variables. To answer this concern we have included control variables describing time-invariant characteristics of the municipalities that could contribute to the explanation of the local institutional share. Another problem of using a linear model is that our dependent variable is a share that varies only in the interval [0,1]. The problem is that POLS 266

10 The Impact of Financial Incentives on the Composition of Long-term Care in Norway could predict values of the dependent variable outside the appropriate range. However, when running the model we find that POLS predicts that the dependent variable will take on values from to Another possible method would be to use a logistic transformation of the dependent variable. A problem with this approach is that the magnitudes of the estimated coefficients are not readily interpretable, as all coefficients are measured relative to the marginal effect of the error term, i.e. b k = Y / Y X k 1. Therefore we have chosen to use a linear model and base inference on the large sample properties of POLS. In table 1 in the online appendix we have included the results of a model that uses the logistic transformation of the dependent variables, and the results are similar. Our primary goal is to find out whether the income of potential and actual recipients affects institutional share. Thus, the main parameter of interest is b 1, which determines the effect of average income of older people on the institutional share. Increasing the average income of older people should indirectly, through the progressive user charge, decrease the cost of institutional care relative to home-based care for the municipality and increase the cost for the user. This means that, all else being equal, a b 1 significantly different from zero implies a relationship between LTC service composition, measured as the institutional share, and the average income of older people. If the local government budget is the primary concern for determining the service composition, b 1 would be positive. If the local authorities care more about the finances of the user (who is part of their electorate), b 1 is expected to be negative. The interpretation is that municipalities with an older population that has high average income will have a cost advantage in providing institutional care. It is important to acknowledge that the average income of older people is not an optimal indicator of the income distribution of LTC service recipients. This problem will be further discussed in the concluding remarks section. Another interesting variable is municipal revenue, Y. After controlling for cost and need, the coefficient in front of Y will tell us whether local economic conditions influence IS. Borge and Haraldsvik (2005) finds a positive relationship between local revenue and institutional share. The reason given is that institutional care is such a costly service that the local revenue constrains the magnitude of IS. This indicates that high income municipalities have a better ability to provide institutional care for their residents, and that the block grants allocated through an income and expenditure adjustment system do not equalize IS across municipalities. This is contrary to national policy LTC services should be allocated according to need. This conclusion depends on Y being independent of local preferences. Although there are national policies on minimum service provision in the LTC sector, the recent increase in community housing that to a certain degree can be adapted to fulfill the same needs as institutional care indicates that local preferences can be a determinant of IS. Thus, if there is a relationship between local revenue and preferences, this will introduce a bias to the estimation of the revenue coefficient. Intuitively, a relationship between local preferences and the optimal division of local private and public income is likely to exist. However, local discretion in revenue raising is very limited in Norway, at least when viewed in relation to the extensive (and expensive) responsibility, defined by law, to provide social and health 267

11 Applied Economic Perspectives and Policy services. This is the reason why studies analyzing local spending behavior have treated unconditional income as an exogenous variable; see Kalseth and Rattsø (1995), Borge and Rattsø (1995), Aaberge et al. (2010), and Aaberge and Langørgen (2003). Since 1979, every Norwegian municipality has charged the maximum allowable tax rate (Borge and Rattsø 1995), and the grant allocations system is designed by the central government to be dependent on structural factors outside the municipalities direct control (Karlsson, Iversen and Øien 2012). This indicates that municipal revenue can be treated as an exogenous variable. Control variables The included control variables can be divided into two groups those affecting need of institutional care relative to home-based care, and those affecting the relative cost between the two care modes. An obvious determinant of municipalities need for institutional care relative to home-based care is the health status of residents in need of LTC services. Low health status is associated with a lower ability to take care of oneself, resulting in a need for health and care services; the lower the health status, the more intensive the care needed. Institutional care offers 24-hour health and care services, indicating that municipalities with a relatively high share of residents in poor health will have, ceteris paribus, a higher need for institutional care relative to home-based care. The variables included to control for health status are age composition (sh6779, sh8089, and sh90a), life expectancy at birth (LE), percentage of older people 67 years and above who live alone (single), and the percentage of recipients with severe and limited dependency levels (SDL and LDL). Age composition and life expectancy are included as proxies for the health status of municipality population, since old age is often associated with deteriorating health. They are also included to control for the fact that institutional care is predominately a care service for dependent older people, while home-based care has a larger share of younger users, so that having an older population in itself could result in a higher institutional share for a municipality. Moreover, it has been suggested in the literature that the amount of resources devoted to an individual depends on his or her remaining life expectancy (Breyer, Lorenz, and Niebel 2011). The share of older people living alone is used to control for the observation that people living alone have a higher probability of needing institutional services (Nihtilä and Martikainen 2008). The dependency levels of LTC recipients and applicants are calculated based on 15 indicators describing the ability to perform activities and instrumental activities of daily living (ADL and IADL), as described in the data section. The proportion of recipients with limited and severe dependency levels are included in the analysis to control for health status of potential and actual LTC recipients. Including ADL measures is crucial for controlling for the need for institutional care across municipalities, as performing ADL are the best indicators of health status resulting in need for nursing home care (Norton 2000). The anticipation is that the share of severely dependent residents is positively associated with institutional share. The share of residents with a limited dependency level is expected to show a negative relationship with institutional share, as there has been a trend toward deinstitutionalization and the idea that persons with physical disabilities should live at home as 268

12 The Impact of Financial Incentives on the Composition of Long-term Care in Norway long as possible (Karlsson, Iversen, and Øien 2012), indicating a lower institutional share, all else being equal. The cost factors we have included describe characteristics of the municipalities that influence the cost of institutional care relative to home-based care. In municipalities with a dispersed settlement pattern, home-based care is likely to be more expensive due to large travel distances. Provision of institutional services is also expected to be based on economies of scale because of indivisible goods such as food services and the legally-required staffing. To control for the settlement pattern, we have included the average travel distance, in hours, to the municipality center and the population density in densely populated areas within the municipality. A higher density leads to lower coordination costs of home-based services and also decreasing unit costs of institutional services if the number of users is positively correlated with our density variable. Thus, the effect of the variable is uncertain. These factors are included in the analysis since there is a probability that these municipality characteristics can be correlated with the local economic environment, captured by Y and INC. We have also included dummies for the first two years to control for the general trend of deinstitutionalizing the LTC sector and the possible real change in income and municipal revenue. Cost shifting The financial system provides municipalities the opportunity to choose a service composition that shifts part of the cost to the central government or to the user. This means that the service composition is prone to costshifting behavior. Cost shifting is here understood to be, as in Norton, Lindrooth, and Dickey (1999), the opportunity of the decision-maker to choose services that reduce their own costs by inducing another public agency or the user to pay for similar services. However, the political feasibility of cost-shifting depends on the services to be substituted that to a certain degree they can meet the same needs. The general political perception of the Norwegian welfare state is that health and social services should be provided according to need, and not based on local cost considerations. It is not likely that local authorities will provide a few hours of home-based care services to persons (in their own electorate) in need of 24-hour care. Thus, we should expect that potential cost-shifting behavior is more pronounced among similar services. To investigate this, we also estimate the above model with the institutional share for recipients who are 80 years and older, IS80a, and the ratio of institutional recipients to the sum of recipients 67 years and older receiving 24-hour care in community housing and institutional recipients, Hi. The first specification is included because institutional care is predominately a service for dependent older people. In calculating H it we exclude all home-based services except 24-hour care for people aged 67 years and above. This specification is included because municipalities could choose to provide 24-hour care either in nursing homes (institutional care) or community houses. Results The results of the three regressions describing LTC service composition are presented in table 2. The first column provides the results for the 269

13 Applied Economic Perspectives and Policy Table 2. Regression results linear model (1) (2) (3) IS IS80 Hi INC (0.0151) (0.0104) (0.0655) Y 0.230*** (0.0850) 0.214*** (0.0609) (0.193) SDL * ( ) ( ) ** ( ) LDL ( ) ( ) * ( ) sh ( ) ( ) (0.0107) sh ** ( ) ( ) (0.0173) sh90a (0.0132) (0.0119) (0.0525) LE *** ( ) ( ) (0.0139) single ( ) ( ) ( ) distance (0.0414) (0.0346) (0.144) density 0.184*** (0.0460) 0.106*** (0.0394) 0.416** (0.175) t *** ( ) ( ) (0.0185) t ( ) ( ) ( ) _cons *** (0.345) (0.301) (1.150) N adj. R Notes: Standard errors in parentheses. *p, 0.10, ** p, 0.05, *** p, institutional share, IS. We see that the coefficient of the income variable is positive. The estimated effect is minor and insignificant; a b 1 equal to means that a 1,000 kroner (178 USD) increase in income is predicted to increase the institutional share by percentage points. This suggests that the average income of older people does not influence institutional coverage. Unconditional income per capita is positive and significant, and the estimated effect is substantial. A 10,000 (1,782 USD) kroner increase in revenue per capita is predicted, on average, to increase institutional share by 2.30 percentage points, which is 11.3% of the institutional share s mean. Thus, it appears that high income municipalities have the ability to provide higher institutional coverage. The coefficients in front of the dependency level variables show the expected signs, and municipalities with a high share of recipients with severe dependency levels have a higher institutional share. This is in line with the national policy that LTC services should be allocated according to need. The predicted effect, though, is rather small. A 1% increase of recipients with a severe dependency level will, on average, increase the institutional share by percentage points. Life expectancy has a more substantial effect; an increase of life expectancy by one year increases, on average, the institutional share by 1.23 percentage points. This indicates that an aging population increases the need for 24-hour care services, and that old age in itself has an impact on the need for institutional services. The time dummy coefficients for the first and second year are positive, which reflects a downward trend of the institutional share. This is in line with the policy of decreasing reliance on institutional care in the LTC sector. There is a positive relationship between density and institutional shares, indicating that the density variable captures increasing returns to scale in the provision of institutional services for given travel distances. 270

14 The Impact of Financial Incentives on the Composition of Long-term Care in Norway Results for the institutional share of recipients aged 80 years and above (IS80), and the institutional share of recipients receiving 24-hour care (Hi) are presented in columns 2 and 3, respectively. The results for the model with IS80a as the dependent variable, apart from the need variables, are qualitatively the same as the results from our first specification. Thus, leaving out younger recipients, there is no indication that municipalities are adapting the service composition of elderly care according to the income level of older people. All of the need variables become insignificant in this specification. In our model, the institutional share for recipients 80 years and above is solely determined by municipal revenue and cost factors. In our last specification, the coefficients in front of SDL and LDL are significant, while the municipal revenue coefficient and the coefficients in front of the need variables become insignificant. Concluding remarks In this paper we have used a simple model to explain the LTC service composition among municipalities in Norway. This was undertaken to explore factors determining service composition in an environment with perverse incentives. Using institutional share as the dependent variable, with the underlying hypothesis that it is influenced by the income level of potential and actual recipients (captured by the average income of older people), local revenue, need, and cost factors, we find no evidence that the progressive user charge creates a relationship between the income level of older people and institutional share. The institutional share is determined by need, cost factors, and municipal revenue. Moreover, municipal revenue has a positive effect on institutional share, indicating that highrevenue municipalities, despite the income and expenditure adjustment system, provide higher institutional coverage relative to less affluent municipalities. Using the same set of explanatory variables we estimate a second model using institutional share for recipients aged 80 years and older as the dependent variable, and a third model using the number of nursing home recipients as a share of all users of 24-hour care services. The first model takes into account that nursing home care is not a political viable service for younger persons in need of LTC; nursing home care is predominately a service for dependent older people. Thus, it could be politically easier to adapt IS80a according to local cost considerations than altering the institutional coverage for younger recipients. However, the results of this model are qualitatively equal to the first model, so there are no indications that municipalities alter the institutional coverage for the oldest of the old to save costs. The institutional share of the oldest old is predominately driven by municipal revenue per capita and factors that describe cost differences. In the last model, we investigate whether the effects are different for services that, at least to a certain degree, can fulfill the same need. In this specification as well, there is no indication of cost shifting. The coefficients in front of the variables controlling for relative need are significant in this specification; however, the magnitude of the effect is predicted to be very small. It seems that municipalities dual regard for the preferences of the user and their own budget balances out. When interpreting and drawing 271

15 Applied Economic Perspectives and Policy conclusions from the results, it should be kept in mind that this paper is based on limited data. First, using the average income of older people as a proxy for the income level of recipients is problematic. Using average income of older people as our income measure is constrained by the available data and is defended by the fact that LTC services are predominately relevant for older people. Of course this assumes that average income for older people will capture the variation in income for actual and potential LTC recipients. However, it could be the case that the level of average income can be driven by a few rich individuals, while actual recipients could have an income well below average. In this case, our income measure will not capture the variation in recipient income the income variable that creates variation in the cost of nursing home care facing municipalities. Second, we are not able to control for municipality fixed effects because of limits within variation in our short panel. This can cause, as is well known, biased estimates if the municipality s fixed effects are correlated with our included regressors. As discussed, the variables used in this analysis show little variation over time and fixed-effects regression would mean a substantial loss of efficiency. Hopefully, the variables that are constant over time within municipalities capture the potential municipality fixed effects. Third, we do not have information on access to family care. For instance, it would be interesting to examine whether variation in the share of LTC recipients living near their children or other close relatives causes differences in the service composition. However, we do not believe that including other variables describing family care alongside the share of elderly living alone would fundamentally change our results. It is a general norm that elderly couples in need of LTC care for each other, and that they prefer formal care in cases of severe dependency (Ingebretsen and Eriksen 2004). This is an evolution of the welfare state that secures extensive public responsibility for primary health care and social services. The Social Care Act, introduced in 1964, removed all legal obligations for families to take care of their dependent elderly. The formal public LTC services are dominant, and family care is likely to work as a complement rather than a substitute to formal care (Ingebretsen and Eriksen 2004). Although a formal analysis would be better, this indicates that family care does not fundamentally alter the distribution of formal public services across municipalities. In future studies we hope to gain access to data at the level of the individual recipient to analyze whether the results stand up to more accurate data. Even though the results give no indication of local government costshunting, there is no reason for funding schemes to differentiate among similar services, thereby causing unfortunate incentives. It is highly probable that LTC financing schemes with unfortunate incentives that face higher demand and tight public budgets due to an aging population would lead to a distorted service composition. As stated in Wittenberg, Sandhu, and Knapp (2003), the chosen LTC services should be based on overall cost effectiveness, need, and user preference. Clearly, the progressive user charge in nursing home care distorts the preferences of the user, as nursing home care becomes increasingly more expensive, relative to community housing, with increasing income. Thus, all else being equal, community housing is relatively more attractive to users, an attractiveness that increases with personal income. These preferences can lead to increased reliance on 24-hour care in community housing where the income level of 272