On Reliability and Stability of Psychophysiological Indicators for Assessing Operator Functional States

A reliable adaptive system should consistently react to variations in operator functional states, and to accomplish this task the input must be reliable. Time, individual differences, tasks and environments are all sources of variation. Anytime we use an indicator to decide whether to trigger an automatic procedure, we need to know how much it is affected by one of these facton. This chapter will distinguish between two different types or reliability: between-subjects reliability, which is of course a key issue for system effectiveness with every user, and within-subject reliability, which is critical for assuring stability of the system over time, over different occasions, and with different set of stimuli supposedly evoking the same pattern. Accorlingly, two approaches are presented that may be useful to achieve these goals: the Generalizability Theory, which is an alternative approach to reliability assessment, and the Bootstrap resampling technique, which is a computationally intensive procedure designed to assess statistical accuracy.