The effects of the Norwegian Coordination Reform on the use of rehabilitation services: panel data analyses of service use, 2010 to 2013

Background In 2012 the Norwegian Coordination Reform was implemented. The main motivation was to encourage municipalities to expand local, primary health care services. From 2012 to 2014, under the Municipal Co-Financing regime, municipalities were obliged to cover 20 % of the costs of health services provided at the specialist (hospital) level. Importantly, use of rehabilitation services in private institutions was not part of the cost-sharing mechanism of Municipal Co-Financing. Rehabilitation services may be seen as quite similar in nature whether they be provided by municipalities, hospitals or private institutions. Thus, with rehabilitation patients readily “transferrable” between levels, the question is whether the reform brought with it a sought after shift towards more municipal rehabilitation and less specialist rehabilitation. Methods Data from the Norwegian Patient Register and from Statistics Norway/KOSTRA were utilized to gauge annual expenditures and inputs in specialist, municipal and private institution rehabilitation services respectively. Fixed effects and first difference regression analyses for the period 2010–2013 were carried out to account for certain time-invariant traits of municipalities and/or hospital regions, and results were adjusted for contemporaneous trends in local needs. Results Expenditures in specialist rehabilitation services declined sharply (typically by 8–10 %) from 2011 (pre-reform) to 2012 (post-reform), while expenditures in private rehabilitation services rose markedly in the same period (typically by 42–44 %). The results do not suggest any general expansion of municipal rehabilitation services. Conclusions The results of the analyses suggest that municipalities shift away from the use of specialist rehabilitation services and towards the use of rehabilitation services in private institutions since the latter becomes relatively cheaper (free-of charge) than both municipal and specialist services in post-reform periods (as specialist services come at a cost to municipalities post-reform). While the main goal of the reform has not materialized the results nevertheless suggest that incentives (of cost-shifting) do play a significant role in rehabilitation service use. Electronic supplementary material The online version of this article (doi:10.1186/s12913-016-1564-6) contains supplementary material, which is available to authorized users.


Norwegian Coordination Reform and rehabilitation services
Implemented in 2012, the main goal of the Norwegian Coordination Reform (CR) [1] 1 was to reduce overall health care costs and facilitate the coordination of the primary and specialist levels of the Norwegian health care system. Coordination difficulties often follow financial and regulatory divisions between specialist and primary health care [2]. This is also the case in Norway, where the provision of specialist and primary health care services is at different organizational levels and financed through different budgets, a condition particularly pronounced in rehabilitation services. The subsequent division of responsibilities between the specialist and primary health care level within rehabilitation has been described as unclear [3], in the sense that both levels maintain much the same rehabilitation capabilities and competencies. Moreover, as both municipalities-charged with primary health care responsibilities-and specialist health care hospital regions are reasonably well suited to provide such services, the incentives to shift costs onto the other part perhaps constitutes the core "coordination problem" in this area.
In this way, the case of rehabilitation services constitutes something of a litmus test for the CR in terms of the prospects for simple coordination. For instance, a finding that the reform has not had the expected effect on these services, i.e. a decline in specialist rehabilitation and an increase in primary rehabilitation, would highlight a serious shortcoming of the reform. This is not necessarily because the reform was tailored specifically for rehabilitation services, but rather because this is where the potential for simple coordination should be particularly evident.
An important feature of the CR was Municipal Co-Financing (MCF). Under this provision, lasting from 2012 to 2014, municipalities were obligated to cover 20 % of the costs associated with a given set of medical Diagnostic-Related Groups (DRGs), including hospital rehabilitation services. The overall motivation for introducing cofinancing was that it would provide municipalities with an incentive to expand (primary) health care provided at the municipal level, in the hope that this in turn would lessen the need for more expensive specialist health care, and thereby reduce overall health costs and improve health outcomes [3,4]. In addition, because it adheres to the Lowest Effective Level (LEL) of care principle, and because municipalities may naturally integrate local medical services with other municipal social and care services, any growth at this level would be particularly beneficial [3,4].
In this article, we analyze the effects of MCF on the use of both specialist and primary level rehabilitation services, along with its effects on the procurement of rehabilitation services in private institutions. Importantly, there has always been in Norway a relatively small private sector offering rehabilitation services. Equally important, the use of private rehabilitation services was, somewhat peculiarly, exempt from the MCF scheme (i.e. the continuation of financing through hospital revenues). The role of private rehabilitation institutions has thus been somewhat politically controversial, and strong incentives for the municipalities to expand their own services could undermine the need for these private sector institutions. In any case, we would expect that at least initially private rehabilitation services would expand under the MCF regime owing to municipal preferences for "free-of-charge" services, insofar as private services may really substitute for services provided by hospitals and/or by municipalities themselves. In the ensuing analyses, we employ regression analysis using annual municipal-level aggregated data from the Norwegian Patient Registry (NPR) [5] (providing data on specialist level and private rehabilitation expenditures) and the Municipality-State-Reporting information system (KOSTRA) [6] (providing data on local health service resources and local demographics) over the period 2010-2013.
The article is organized as follows. To start, we briefly place our research questions within the broader research field dealing with the relationship between the primary and specialist level health care sectors, and describe the overall Norwegian institutional setting, with a particular emphasis on the specifics of the CR and MCF regime. Subsequently, we develop hypotheses concerning the expected demand for rehabilitation services at different levels, both pre-and postreform, as this is responsive to both changes in (i) (demographic) needs, and (ii) to (pre-and postreform) service prices. The following sections then present our data and empirical strategy, and the results from the statistical analyses. In the regression analysis, we trace the changes in demand for different rehabilitation services, employing both fixed effects and first-difference specifications in order to account for local heterogeneity over the period of study. Although we cannot rule out that changes in the use of different rehabilitation services are in fact the result of other confounding factors, we do believe that the introduction of the MCF regime is the most important contemporaneous driving force, and our analyses gives us some leeway in evaluating the impact of confounding trends due to other factors. We discuss our findings and their implications before providing concluding remarks.
The relationship between health care levels, the Norwegian institutional setting, and the CR: implications for rehabilitation services Often described as a decentralized National Health Service (NHS)-type system, the fundamental features of the Norwegian health care system resemble those of the UK NHS, although Norwegian municipalities play a relatively more important role than do their UK counterparts. For example, the responsibility for Long Term Care (LTC) services and community health services lies at the municipal level. Provision of general practitioner (GP) services is also a municipal responsibility, although GPs are usually not municipal employees as such. Rather, they are independent service providers operating in a contractual relationship with the municipalities [7]. 2 For their part, the revenues of the municipalities consist of block grants from the central state, taxes on personal income and wealth, and other specific grants, such as grants to municipalities in rural areas. There is the redistribution of tax income among municipalities to reduce inequalities, and the most important part of the block grant is a per capita component adjusted for centrality and inequalities in the demand for services. Thus, the design of the system serves to adjust for municipal needs and factors affecting the ability of municipalities to deliver services.
For the most part, the financing of somatic specialist health care is through a combination of an adjusted per capita-based block grant and Activity-Based Financing (ABF) based on the DRGs. Even though the Regional Health Authorities (RHAs) are free to choose the model they deem best for financing hospitals in their catchment area, all have opted for the same model as the central state uses to fund the RHAs themselves. Consequently, while the ABF share of the total funding of specialist health care has varied between 40 and 60 % over the last 15 years, in the shorter more recent period under study (2010-2013), it has remained at 40 % throughout.
Introduced in 2012, the CR's main elements were: (i) MCF, whereby the municipalities had to cover 20 % of the running hospital costs for medical DRGs; (ii) a discharge arrangement where municipalities were charged NOK 4000 per day from when a patient was classified as ready for discharge by the hospital; and (iii) new municipal acute bed units (MAUs). The MAUs are intermediate units introduced to reduce use acute admissions to hospitals. By law, both RHAs and municipalities must have a unit responsible for the coordination of rehabilitation services. Municipalities are also required to devise individual plans for each patient/user in need of longterm care and to provide coordinated health services [8]. One of the primary goals of these regulations concerning rehabilitation was to strengthen coordination within both the primary and specialist levels, as well as between the two levels. The integration of health and social care is at the municipal level, and this, in combination with the proximity of local authorities to users, means that municipalities are well suited to provide high-quality "low-tech" rehabilitation services. Moreover, in the white paper first introducing the CR, the then Minister of Health stressed in the foreword the importance of enabling the municipalities to provide rehabilitation services [4]. In other words, there are clear arguments in favor of an increase in the use of municipal rehabilitation, both as far as quality and costs are concerned.
Furthermore, we have argued that rehabilitation services rendered at different levels (specialist, municipal, and private) are "substitutable" in a very simple fashion. Numerous studies show that primary health care interventions, seen as preventive interventions, may have positive effects in terms of better long-term health outcomes and lower overall health care costs [9]. Nevertheless, such studies, and associated recommendations, will typically point to gains from the substitution of simpler primary care interventions for costly and complex treatment at later stages, i.e. gains in terms of better health outcomes and/or lower overall health costs only secured after some time [9][10][11][12][13][14][15][16].
In contrast, when analyzing the use of rehabilitation services at different levels, we look at the substitution of quite similar services across levels, which may or may not have effects in terms of better outcomes in the longer run. In other words, the focus of the analysis is on services that are easily shifted between levels and should be strengthened at the municipal level (given municipal services have certain qualities), but that might not shift in this fashion given the peculiarities of regulation, financing regime, and reform design. This facilitates the analysis of potential incentive effects, such as those stemming from the MCF, and ensures that other more "medical"-related coordination problems (e.g. the correct ratio of heart surgery to exercise) are less of a concern. In sum, both the apparent political significance of the rehabilitation area and certain simplifying aspects of the area itself serve to motivate the importance of the analysis.
Before the CR, use of specialist rehabilitation services was ostensibly "free of charge" for municipalities. With the reform from 2012 onward, and following the introduction of MCF, municipalities were required to cover 20 % of all specialist-level health care costs. The main motivation for this was to provide municipalities with an incentive to expand their own services, including rehabilitation services. Lastly, as noted earlier, nonprofit private rehabilitation institutions play an important role in the institutional configuration of rehabilitation services in Norway. Importantly, the MCF scheme did not cover the use of private rehabilitation, with its financing continuing through hospital block grants.
We now hypothesize about the incentive effects in the pre-and postreform periods. These, of course, do not necessarily reflect the aspirations of the reform. The first issue concerns the likely patterns of service provision demand as we change from the pre-to the postreform period. As the use of private rehabilitation providers is "free of charge" for municipalities, with costs covered through hospital block grants-and given both the use of (postreform) specialist rehabilitation and that the expansion of own capabilities comes at a cost-we expect municipal demands for private rehabilitation services to increase at the expense of their demands for specialist and municipal rehabilitation. In addition, we would expect that the demand for municipal provision would not increase much in absolute terms regardless. Even if there are certain intrinsic qualities regarding municipal services compared with specialist and private services, the former are likely considered "too expensive" vis-à-vis readily available alternatives, i.e. specialist services at what is likely a fraction of the cost (80 % of the specialist service price at any rate), and free private services. 3 Before we turn to a description of the data and the empirical analyses, we briefly consider that the hypotheses developed above focus on municipal incentives. While such a focus is in line with the "official" working mechanism of the reform, we also believe this is appropriate for two reasons. First, municipal incentives change dramatically and clearly as the reform sets in and as described. In contrast, hospitals are only partly able to maximize disposable revenues and/or minimize outlays. However, if hospitals prefer long-term "major activity", they will certainly favor a situation where municipalities purchase specialist services, insofar as specialist DRG prices are accurate. Conversely, they will most disfavor the situation where municipalities demand private rehabilitation, as this will also deplete current disposable funds (that nearly match activity-based refunds-recall that private rehabilitation is not part of the ABF system). Given the limited incentives of hospitals, with their largely refund-based financing system, in the following analyses and interpretations, we focus on municipal incentives. Moreover, it should be clear from the discussion that a finding in accordance with the main hypotheses outlined above (dealing with municipal incentives) is in fact a conservative one (given that hospital incentives should drive patterns in the opposite direction). In addition, the analyses also offer controls for certain invariant characteristics of hospitals as well as municipalities and private service providers (e.g. rigid capacity limits).
Municipalities can act on their incentives in three ways. First, although the relationship between the municipality and the GP is contractual (the vast majority of GPs are contractors), the municipality likely has some influence over the GP's decisions. Second, they can prioritize sending patients that come from the hospital to a private rehabilitation hospital, rather than providing municipal services or using public hospital services. Third, they can refer patients who have received care in the new MAUs who are in need of further rehabilitation to private institutions and public hospitals or municipal rehabilitation.
A final point is that in analyzing the changes in the patterns of rehabilitation service provision, we should naturally adjust for possibly confounding changes in local health care needs during the period of study (proxied by local demographics in our study). Thus, a natural, albeit auxiliary, set of hypotheses is that changes in service levels should be less sensitive to needs when services are costly. In other words, we expect that needs variables are more important predictors of specialist rehabilitation use before the reform than after the reform. This is also an effect of the relative prices of the different alternative rehabilitation services available to the municipalities.

Data
In the empirical analyses, we use aggregate data from the Norwegian Patient Registry (NPR) recording (i) annual municipality-level utilization of six specialist rehabilitation DRGs and (ii) annual municipality-level use of private rehabilitation services (24 h stays) over the period 2010-2013. 4 We calculate (i) the annual costs of specialist rehabilitation DRGs per capita in each municipality by annually aggregating individual DRG diagnoses multiplied by the annual DRG weight attached to each DRG type multiplied by the annual DRG price 5 [17] and dividing by the annual average municipal population; (ii)the annual costs of private rehabilitation services per capita in each municipality by annually aggregating individual stays in private rehabilitation institutions multiplied by the average price per stay (NOK 2886 in 2012) and dividing by the annual average municipal population. 6 As a measure of the level of municipal rehabilitation services, we use Statistics Norway/KOSTRA records on annual man-hours for physio-and ergotherapists in municipal employment, and calculate (iii)the annual municipal sum of man-hours worked by physio-and ergotherapists per capita by dividing total man-hours by the annual average municipal population [18]. 7 To compare the level of service costs over time and the cost sensitivities to local needs and resources, we deflate the measure of DRG use using the Norwegian Consumer Price Index [19]. Thus, the costs presented in the tables and analyses are in 2010 NOK (or its log).
In addition, we use data from Statistics Norway for the period 2010-2013 to gauge local levels of rehabilitation needs [20]. Specifically, we use the annual municipal population shares of 67-79 year-olds and those 80 years old and over as measures of local needs. This is because these population groups will reasonably require the most extensive health care, including rehabilitation services. We also employ a measure of reported local crime rates (per 10,000 inhabitants), and the rate of deaths per capita as proxies for local living conditions. This is because lower living conditions will reasonably entail heavier use of health services, including rehabilitation services. Additionally, we extract from the KOSTRA data the annual local population to calculate the per capita rates of the service-level variables (i)-(iii) above. Table 1 provides descriptive statistics for the 409 municipalities with valid values for these variables. 8 The most apparent pattern in Table 1 is the sharp decline in the use of specialist rehabilitation services from an average of NOK 252 per inhabitant in 2010 (prereform) to NOK 200 in 2013 (a 20 % decline). In addition, the use of rehabilitation services in private institutions increases dramatically over the same period, from NOK 18 per inhabitant in 2010 to NOK 29 in 2013, i.e. the average increases by a staggering 61 %.
Conversely, man-hour inputs in municipal rehabilitation services increase only from 3.73 per 10,000 inhabitants in  The empirical question is whether these results hold when we adjust for contemporaneous trends in local needs and look specifically at the patterns within the typical municipality (by way of fixed-effects and firstdifference model specifications).

Model specification
As we wish to trace the changes in the use of different rehabilitation services, we estimate variants of the following panel (fixed effects) model: That is, we estimate a set of k = 3 equations where y is a particular measure of rehabilitation service use (k = specialist, private, municipal) in municipality i in year t (t = 2010, 2011, 2012, 2013); and needs is a vector of c = 4 municipal needs variables (c = two measures of shares of elderly, death rates and crime rates), with γ a vector of associated parameters. The inclusion of indicators for N = 409 municipalities, i.e. the Mun indicator variables, means that we effectively control for any traits of municipalities-or the relationship between the individual municipality and the health region-that are stable over the relevant period (such as stable capacity constraints, say). In addition, we estimate the equations in terms of logs of the variables, so that we may interpret effects as elasticities, and so as to get at a more reasonable and normal scale for measured changes in outcomes in particular: As can be seen in Table 1, outcome levels (i.e. rehabilitation service use) are (naturally) heavily skewed to the right, while this is also so, although less pronounced, for the needs variables. 9 The tendency for distributions to be right-skewed also applies to relative changes in variables (not shown), and a logtransformation (of change rates) seems to remedy such problems considerably: As can be seen in Additional file 1: Table S1, minimum and maximum values of logtransformed changes seem to be distanced quite symmetrically around their mean values.
Much of the "distributional concerns" discussed above also have to do with the fact that municipalities are predominantly small in Norway 10 and population rates of use of rehabilitation services are in themselves relatively small. This causes the variables (the dependent variables in particular) to be poorly measured as "substantive changes" in quite many of the observations. Specifically, variable values in most cases, i.e. in the smaller localities, will reflect large changes due to quite random movements of very few individuals (in and out of age groups, in and out of municipal employment). Thus, when reporting descriptive statistics and when estimating the equations, we weight observations (it, municipality-years) by the average annual municipal population, and interpretations are in terms of effects for the "average sized municipality" (around 12,000 inhabitants in 2012).
Our primary interest lies in the estimates of the α T coefficients, i.e. the effects associated with the Year T indicator variables (T = 2011, 2012, 2013, with 2010 the reference category). Although we are not able to take account of all other possible factors that coincide with the reform, other than changes in municipal needs and time-invariant municipal characteristics, we can think of no other systematic factor that should drive service use in a particular direction during the period under study. Accordingly, we interpret the α T as "reform effects".

Results
We present results from the estimation of variants of equation (1) in Table 2. First, in models without covariates, i.e. models (A), we estimate significantly different service levels in reform years (2012 and 2013) as compared to the reference year (2010). For example, as compared to levels in 2010, average use of specialist services decreases by 14 % by 2012 and by 20 % by 2013. 11 Conversely, the use of rehabilitation services in private institutions increases sharply in the same manner, while there is less pronounced increase in the use of municipal services. Joint tests of the differences between 2011 and 2012 levels, which may be interpreted as the adjustment going into the reform, reveal that effects in all service sectors are substantial and highly significant. 12 On the other hand, levels do not change substantively or significantly in the pre-reform period from 2010 to 2011, save for municipal rehabilitation services. 13 Moreover, joint tests of differences between levels in 2012 and 2013, i.e. within the reform, are smaller than changes from 2011 to 2012, suggesting that there is little adjustment both before and after the reform, but some adjustment going into the reform. 14 In the next column (B) we add needs levels (Needs c ) in the individual municipality-year, since service use should arguably be affected by changes in needs. We observe that needs-in the form of shares of elderly people-are positively, but not significantly, related to use of specialist and municipal service levels, but negatively related to use of rehabilitation services in private institutions. Likely, this reflects that specialist rehabilitation, more than rehabilitation in private institutions, is to some extent more attuned to the needs of the elderly. Regardless, estimated α T largely retain their previous values, suggesting that there are pronounced "reform effects".
Next, in models (C) we apply instrumental variables (IV) estimation, instrumenting the two elderly shares variables and the death rate variable by their lagged values from three years back. The idea is that a straightforward ordinary least squares (OLS) specification, as in model variants (A) and (B), will give biased estimates of needs effects, since there is most likely some reverse causality in the relationship between needs and service use. Specifically, a particular pattern of service use, induced by the CR, say, might arguably influence the survival of inhabitants and demographics as such (young vs. elderly) in the individual municipality.
Thus, observed correlations between needs and service use may not reflect a true causal effect from the former to the latter. Moreover, since the mentioned needs variables are likely endogenous in this way, not only will estimates for needs effects be biased, effect estimates for other explanatory variables that are correlated with needs-for example year effects-will also be biased. Our argument for the IV specification is that previous needs levels in part determine present needs (death rates and demographics), but not current service use, which is plausibly determined systematically by present needs only. Results for IV estimations are presented in column (C) of Table 2. 15 From the estimations of model variant (C), as compared to (A) and (B), we observe that the positive CR effects, i.e. year effects, are somewhat more pronounced in the case of private rehabilitation services-with α 2012 estimtates increasing from around 0.26 to 0.32; that the negative CR effect in specialist services is somewhat attenuated-with the estimated α 2012 dropping from −0.15 to−0.11; and that the positive effect in municipal rehabilitation services shifts to a negative, albeit not significant, effect. In addition, the negative effects of shares of elderly mentioned above become more pronounced and highly significant. Despite this however, joint tests of the α T estimates reveal that the pattern of pre-and postreform non-adjustment and reform adjustment (from 2011 to 2012) stays much the same as in model specifications (A) and (B). 16 These results are in line with the hypothesized pattern in the previous discussion: Municipalities shift use towards services that become relatively less costly. For ease of exposition, we illustrate the results from the analyses in model (C), i.e. from joint tests of α T estimates, in Fig. 1. Here, annual predicted changes in the use of the different rehabilitation service types are plotted along time (year). The pattern clearly suggests that municipalities adjust their use of rehabilitation services towards those that become cheaper (i.e. towards private services going into the MCF reform in 2012) and away from those that get costlier (i.e. away from specialist services in the same period).
As a final robustness check, we estimate effects in a first-differenced specification: The motivation for this is that error terms, ε it , might be serially correlated, and under such circumstances first-differencing, rather than a fixed effects specification, is a better choice [21,22]. 17 Results are shown in Table 3. Bearing in mind that the α T effects are now estimates of changes directly, or differences in changes, rather than levels, one observes in model variant (D) much the same pattern as in the fixed effects analyses in model specification (1). Specifically, there is little and insignificant change in specialist service levels from 2010 to 2011, as witnessed by the intercept estimate  Table 1. Overall, in the case of in specialist and private rehabilitation service use, the mentioned pattern of reform adjustment and pre-and post-reform non-adjustment also prevails in the first-differenced specification. 18 Lastly, in model variant (E) we allow for time-specific needs effects, in order to test the hypothesis that municipal use of specialist rehabilitation services should be less sensitive to needs as services become costlier. 19 As can be seen in column (E) in the analysis of specialist rehabilitation services, the needs effects are all positive, and significantly so for the crime rate variable. 20 In addition, as shown in Additional file 3: Table S3, the accompanying interaction terms are by and large negative, and significantly so for the interaction between crime rate variable and the indicator variable for 2012. This is broadly in line with the hypothesis that municipal use of (specialist) rehabilitation services should be less sensitive to needs as services become costlier (as they do in reform periods). Moreover, the pattern of reform adjustment and pre-and post-reform non-adjustment, evaluated by joint tests of the α 0 and the α T estimates, is apparent also in the time-varying needs effects specification (E). Specifically, for specialist rehabilitation services there is no significant pre-reform (2010-2011) adjustment (p = 0.37), a significant adjustment going into the reform and no (p = 0.00), and no postreform adjustment (p = 0.23). The pattern is the same for use of private rehabilitation services, but with a significant decrease pre-reform (cf. the negative and significant, p = 0.02, α 0 ), and there are no significant year-toyear changes for the use of municipal rehabilitation services. 21

Discussion
While there was a pronounced focus on improving and expanding rehabilitation services in the years preceding the reform, its primary goal was to provide municipalities with an incentive to expand their own services. Nonetheless, the CR does not appear to have led to a general increase in the provision of municipal rehabilitation services.
Of course, we should note some possible limitations of our study. First, our measure of municipal rehabilitation use is rather rough (see note 7). Future research should attempt to acquire better data to address this point, as the measure arguably does not encompass the full breadth of municipal rehabilitation services. Second, as we do not have information about the source of the patients admitted to private rehabilitation institutions, we cannot rule out the possibility that much of the increase stems from patients arriving from hospitals. Certainly, the length of stay for patients in hospitals decreased after the CR [23], although the waiting time for somatic patients appears to have changed less [24]. Nonetheless, a reduction in the mean LOS may also have led to a greater share of patients arriving at private rehabilitation institutions from public hospitals, but we do not know to what extent this has happened. Another mechanism introduced in the CR was discharge arrangements where municipalities were charged a fee per day for patients, in the case that they were declared ready for discharge but could not be received by the municipalities. One way for municipalities to handle this, for example if they did not have the capacity to receive patients, was to use private rehabilitation institutions. Some of the patients provided with acute care in the new MAUs rather than in a hospital would probably also require rehabilitation services. Unfortunately, again, our data does not allow us to identify where patients come from. Other elements of the CR, in particular the new discharge arrangement, may play an important role here. Nevertheless, municipalities appear to have vastly increased their use of private providers of rehabilitation services rather than expanding their own services. This is most likely because of the relative increase in the price for specialist rehabilitation services in the wake of the CR. For example, with the MCF, the municipalities would need to cover 20 % of the costs of such services themselves. On the other hand, the use of private providers of rehabilitation services has remained free of charge throughout, which may explain the observed strong decline in the use of specialist rehabilitation services as the reform set in. Thus, the reform aspirations that would point toward local service expansion have not occurred. Instead, local authorities have perhaps naturally shifted their attention toward readily available and cheaper alternatives (private rehabilitation services).
Increased cost sharing under the MCF regime might have created stronger incentives for expanding municipal services. Nevertheless, under such an arrangement, municipalities would have had to handle increased risks. Demand for services will fluctuate from year to year, and for smaller municipalities, i.e. for smaller catchment areas, this will often be difficult to manage [3]. In other words, there is a trade-off between the incentives placed on agents (municipalities) and the amount of risk entailed. In the Norwegian context, where the municipalities are often small and therefore particularly susceptible to financial risk, this is a very real problem. Consequently, one way to enhance the potential for incentive schemes such as MCF-but without disproportionately increasing financial risks to an unacceptable level-would be the consolidation of local governments. However, at present the prospects for substantial local government structural reform in Norway do not look good.
Reforms that rely on incentivizing mechanisms also need to address the institutional configuration of the system. On the one hand, much of the literature [25], along with evaluations of the recent Danish structural reform [26], indicates that the potential for success from incentivizing schemes such as the MCF is low. There are few indications of a general substitution effect. Nevertheless, rehabilitation services are arguably an area where the potential could be greater. We find no effect in terms of reform aims, i.e. expansion of local government services, although there are indications that municipalities responded to financial incentives present in the reform. With private rehabilitation services exempt from the MCF and ABF altogether, the reform created incentives for municipalities to rely on private providers, and indeed, we observed a strong increase in the use of private rehabilitation services. On this basis, we could argue that providing municipalities with stronger incentives to expand their own services might have worked better to achieve reform goals. Counterfactually, the inclusion of private rehabilitation services in the MCF and/or larger local governments might have led to different patterns of rehabilitation service use. As argued, it seems that the institutional configuration, i.e. a strong preference for "small local government," provides little room for implementing cost-sharing mechanisms sufficiently strong enough to have any substantial effect.

Conclusions
Our findings suggest that municipalities shifted away from the use of specialist rehabilitation services and toward the use of rehabilitation services in private institutions because the latter are relatively cheaper (i.e. free-of charge) than either municipal or specialist services in the postreform period. In addition, there is little evidence of any substantial expansion of the municipal rehabilitation service sector following from the CR and the MCF mechanism. Since ours is a temporal field study, and neither an experiment, nor one in which a separate control group is available, we cannot rule out that other unobserved confounding factors contemporaneous to the reform are what is driving the patterns of rehabilitation service use. However, to our knowledge there are few other systematic factors correlated with the initiation of the reform, but not stemming form the reform