statistical identical groups. When this is done with a large enough number of participants, any differences among them that might influence their response to the treatment will be distributed evenly.

regression adjustment—a statistical technique for reducing bias in an experiment that can occur when variables other then the one(s) being studied may affect the results in nonrandom ways.

regression discontinuity design—a quasi-experimental analysis that can be used in program evaluation when randomized assignment is not feasible. It is based on the assumption that individuals who fall just above or below a cut-off point on a particular scale are likely to be similar, so that this group can be treated as varying randomly.

selection bias—an unrecognized difference in the characteristics of the subjects of an experiment who do or do not receive the treatment, or who or do not benefit from it, that will affect the results.

shadow value/shadow price—the true value or cost of the results of a particular decision, as calculated when no market price is available; a dollar value attached to an opportunity cost.

worst-case bounds—a statistical analysis in which the outer limit assumptions for an experiment—both the best possible and worst possible outcomes in terms of the data supporting or not supporting the experimental hypothesis-—are examined. This analysis provides a way of assessing the significance of actual error that may occur in any experiment.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement