Local government expenditure decisions: Empirical models for the analysis of spatial data

This thesis is concerned with the English local authority expenditure decisions in the 1980s. I focus on two features of local governments that might affect their levels of taxation and spending. The first is the presence of a non-linear grant distribution system (the Block Grant). I estimate a demand function for local public spending in the presence of the piecewise-linear non-convex budget constraint created by the Block Grant. Since Ordinary Least Squares yield biased estimates - due to the endogeneity of the choice of segment on the kinked budget constraint - I use a two- error Maximum Likelihood procedure. The second is the spatial character of the data collected at the subcentral government level. Firstly, I explore whether local public expenditure exhibits a spatial pattern. I compute spatial statistics which suggest that local governments' expenditures are positively spatially autocorrelated. Application of spatial econometric models shows that, when allowing both for a spatially lagged dependent variable and for spatial correlation in the error term, most of the spatial correlation in the data is captured by correlated shocks. Secondly, I look at the implications of spatial dependence for local taxation and its effects on local election results. I model the relationship between voters and incumbent governments as a principal agent one, where local jurisdictions are subject to spatially correlated cost shocks and imperfectly informed voters decide whether to re-elect the incumbent after looking at their jurisdiction's relative performance. The empirical evidence on a panel data set of the English districts' elections shows that the own tax has a negative impact on the incumbent's re-election chances, which is only partially offset by the positive impact of neighbouring local governments' taxes. However, local election outcomes also appear to be largely affected by national politics issues