Exploring the spatial pattern of mental health expenditure.

BACKGROUND Recent years have witnessed growing interest in cross-sectional variations in municipality mental health expenditure. However, empirical work to date has not examined the links between such variability and demand and supply factors, particularly in the spatial domain. AIMS OF THE STUDY The aim is to examine whether a local authority's spending decisions in the mental health field respond to neighbouring expenditure decisions. We explore a number of reasons why there might be interdependence between local authorities' decisions, labelling them the demonstrative, market leader, contextual, directive, shared resource and inducement effects. METHODS Exploratory techniques from spatial data analysis are used to test for the existence of spatial structure. Drawing hypotheses from these initial exploratory analyses, we then adopt a reduced form demand and supply model, extended to incorporate possible policy interaction. The analysis of expenditure and cost variations has traditionally been based on regression models under the classical assumption that the observations are independent. But omitting the recognition that observations are interdependent might lead to erroneous statistical conclusions. Hence, we use spatial econometric techniques that explicitly take into account the potential interdependence of data in order to study the sources of spending variation between municipalities. RESULTS The exploratory data analyses reveal the presence of positive significant spatial correlation. Per capita mental health spending distributes in clusters, with the highest concentrations in metropolitan areas such as Greater London, Greater Manchester and Birmingham. The estimated spatial regression models indicate that spatial autocorrelation characterises local expenditure decisions, consistent with some degree of policy interdependence between neighbouring municipalities. Comparing the results from our spatial model with those from a classical ('non-spatial') model suggests that the differences in the regression coefficients could be explained by the evident spatial pattern of the phenomenon, since the omission of the lagged dependent variable induces bias in the OLS estimates. IMPLICATIONS FOR HEALTH POLICIES These results help central and local decision makers understand the factors that influence local spending levels, including variations between municipalities in their achievement of expenditure-related and perhaps other performance targets. The actual patterns of spatial interaction may well be more complex than simple contiguity (the structure assumed here), but there seems little doubt that positive interdependence is an important feature of decision making. IMPLICATIONS FOR FURTHER RESEARCH Statistical interrogation of a panel dataset would allow exploration of both time-series and cross-municipality variation in mental health expenditure. Subsequent analysis would also benefit from more disaggregated data (e.g. at a census ward level) and the accompanying use of spatial multilevel techniques.