Experiments and Observational Studies
Experiments and observational studies are two kinds of statistical investigations. Experiments are characterised by tight control over the processes involved. They tend to incur high expenditure per subject (unit), but have greater potential to collect more information in relatively small samples. In contrast, observational studies involve much less control, their expense per unit tends to be lower, but inferences are more difficult to make, and some of the difficulties could not be resolved by increasing the sample size. In many settings, experiments are not feasible, and inferences have to be based on observational studies. Their analysis is more challenging than that of experiments and often has to rely on some unverifiable assumptions.
This chapter presents a framework in which experiments and observational studies are simply ‘studies’ distinguished by the presence or absence, or in general the level, of the designer’s control over the assignment of treatments to units. In several aspects, this approach parallels the sampling theory introduced in Chapter 3. In particular, it makes use of a missing-data formulation of the inferential task.
KeywordsObservational Study Sampling Variance Background Variable Potential Outcome Exclusion Restriction
Unable to display preview. Download preview PDF.