Matrismodellen vs Two-part regressionsmodeller - effekter på Region Skånes resursfördelning
Operations researchProgrammingMatrismodellStatisticsResursfördelningRegressionActuarial mathematicsAktuariematematikProgrammeringOperationsanalysStatistikMathematics and Statistics
An important task for Region Skåne is to allocate resources to the health care districts. From 1999 to 2002 Region Skåne used needs-based resource allocation as a model for allocating resources. In a needs-based resource allocation individuals with the same socioeconomic and demographic characteristics are assumed to have the same level of need and are therefore allocated the same amount of resources. During the period of needs-based resource allocation a matrix model was used as a method. In the matrix model individuals were divided into cells after each combination of the socioeconomic and demographic variables. Mean costs in each cell were then calculated and summed for each health care district. An alternative to the matrix model is to use regression analysis. However, the dependent variable health care cost is characterized by a large fraction of individuals with zero costs and few individuals with very high costs; hence health care cost has a highly skewed distribution. Health care cost is therefore assumed to have a mixed distribution; i.e. it is both discrete and continuous. Two-part models are specially developed for this type of distribution. By applying a two-part model a more precise resource allocation is assumed to be accomplished. In this thesis the matrix model is compared to two different specifications of the two-part model but also with an ordinary multiple regression model. The focus is on how the different models affect the resource allocation. The result shows that the two-part models allocate fewer resources in total than the other models. The conclusion is that the advantage of applying a two-part model is low. This is due to the fact that the analysis of the two-part model is complicated, both theoretically and practically.