Search results

This article suggests criteria for suitable research designs for use with large data bases. The advantages and disadvantages of several types of quasi-experimental designs are compared. Among the more interesting designs are those trying to minimize the possibilities for selection differences between treatment and control groups by considering assignment to treatment in creative ways. When the probability of receiving a treatment is largely outside an individual's control, possible selection bias should be reduced. These designs often use data on populations rather than on the particular recipients of an intervention. Such analysis of impact on a population, rather than of effect on recipients, makes it comparatively difficult to find that an intervention makes a difference. Examples are taken from our research with the Manitoba Health Services Commission data.

This article discusses methods of organizing and checking administrative data banks to increase their usefulness for research and evaluation. Examples are drawn from the studies using the Manitoba Health Services Commission data bank. The various uses of both manual and computerized checks are explained. A problem profile is presented for the Manitoba data bank; this profile summarizes different specific problems, incor porating new threats to internal validity associated with the use of data banks. The improvements in data collection represented by data banks are discussed in terms of developing multiple control group designs.