Wpływ agregacji i współliniowości na wyniki regresji wielokrotnej w analizie przestrzeni

Abstract

Simulated data are used to test the impact of aggregation and of multicollinearity on the results of the multiple regression analysis. Aggregation increases the level of the multiple correlation coefficient. Furthermore, contextual (or ecological) effects can conceal the true relationship between dependent and independent variables. Redundant variables due to multicollinearity are usually eliminated by stepwise regression procedures. However, the variables thus selected differ from one model to another: the actual underlying relation is not dislosed. Various ways of coping with aggregation and multicollinearity are proposed as conclusions.