ABSTRACT
This chapter presents design features that produce such inadvertent manipulations, though the problem is by no means confined to questions of proper design. It deals with the very simple situation in which only the variance of a dependent variable, say X3, is manipulated in one’s sample design. The chapter shows that an overselection of extreme cases on the dependent variable will tend to confound their effects with whatever independent variables have been singled out for investigation. The more general point is that a reasonably complete recursive system will contain a number of variables, most of which will be taken as dependent in one of the equations in the system. The chapter focuses on matching designs in which there is selection on a combination of dependent and independent variables. It also deals with the confounding likely to occur whenever one uses aggregated data based on geographic proximity.
