Incorporating Knowledge that Data will be Missed in the Design of Human Factors Experiments

Abstract

Experiments with humans invariably have missing data; i.e., equipment fails, subjects may become ill, the weather turns unseasonable. This prior knowledge is seldom incorporated explicitly into the design of experiments.

This paper examines the influence of missing data on the quality of empirical research. A simple approximation function for the expected information content of an experiment is presented. Decision rules for choosing between complete randomization and blocking of subjects are presented. The results are applied to a sequential study of carbon monoxide and alcohol effects on drivers.