Removable selection bias in quasi-experiments

In many experiments and observational studies, random assignment of subjects to treatment and control groups is not possible. These "quasi-experiments'' can potentially lead to what is called selection bias, where the effect of the treatment is confounded with pre-existing differences in the treated and control sequence groups. Some quasiexperimental designs are immune to certain specific selection biases, and it has been widely suggested that this immunity is linked to the absence of an interaction between the selection mechanism and time (this interaction is commonly called selection by maturation interaction). This article shows that immunity to selection bias is not well characterized in terms of selection-by-time interaction, and in particular some good designs can be immune to certain types of selection bias even in the presence of such interaction. An example concerning fraudulent billing is discussed in detail.