Generalization of Results beyond the Experimental Setting: Statistical versus Logical Considerations
The conventional distinction between fixed and random variables in analysis of variance is discussed. It is suggested that certain “incidental” conditions prevailing when experimentation occurs, e.g., time(s) and place(s) of experimentation, can be classified as unbalanced independent variables. Given this classification of these incidental conditions, generalization of experimental results beyond situations exactly duplicating that set of conditions attendant to the experiment, requires either statistical or logical provision. As statistical provision for the entire complex of incidental conditions prevailing during an experiment would be infeasible, logical considerations must be employed in generalization of results beyond the experimental context. It is suggested that to consider incidental conditions of the experimental situation as being unbalanced independent variables should lead to vigilance with regard to those incidental conditions possibly contributing to the obtained experimental outcome. Inasmuch as generalization of obtained results beyond the experimental setting is contingent upon logical considerations, it is suggested that the validity of such logical provision should always be verified by E.