A clusterwise regression method for the prediction of the disposal income in municipalities

The paper illustrates a clusterwise regression procedure applied to the prediction of per capita disposal income (PCDI) in Italian municipalities. The municipal prediction is derived from the provincial PCDI taking into account the discrepancy between municipality and province in some indicators like per capita taxable income, per capita bank deposits, employment rate, etc. The relation between PCDI and indicators is shaped by a regression model. A single regression model doesn't fit very well all territorial units, but different regression models do it in groups of them. The aim of clusterwise regression is just that: detecting clusters where the correspondent regression models explain the data better than an overall regression model does. The application of the procedure to a real case shows that a significative reduction of the regression standard error can be achieved.