The Effect of Experimental Mortality on the Internal and External Validity of the Randomized Comparative Experiment

Behavioral researchers are often faced with the problem of missing observations. Experimental mortali ty, the loss of Ss from treatment groups, can devastate the most carefuUy designed and controlled experimental research. Standard textbooks in research methods or experimental design give no help to the researcher who has experienced substantial experimental mortality. This paper proposes a method for empirically testing whether the experimental mortality is random or non-random both between and within treatment groups in the analysis of variance sense. The results of this test allow the researcher to determine whether the results of his study are susceptible to internal and/or external invalidity. Thus, proper interpretation of the experimental re sults can be made even when substantial numbers of the observations are missing.