Abstract
Purpose: :
To describe time–series methods for analysing before v. after treatment effects in individual subjects. Specifically, to (1) describe method for evaluating serial dependency in response measures with high dimensionality such as multichannel EEGs. (2) To estimate the interval length between recordings needed to avoid the serial dependencies. (3) To investigate the magnitude of confounding time effects relative to stimulus effects.
Methods: :
EEG recordings were mainly acquired from studies by Delorme et al (2004). Tests were performed using recordings when subject was vigilant but not stimulated (control) or was stimulated (test). Before v. after tests were made with EEGs from different days. Variance of sample mean was estimated from spectral density function of each sample (Priestly, 2004).
Results: :
(1) In simulations, whereas ordinary t–tests produce 30% false positive rates, time–series statistics produce false–positives at about 5% (significance declared at p less than 0.05). (2) Decorrelated component scores from the EEG recordings could be modeled as an autoregressive AR(1) process in normal observers. (3) Serial dependency in component scores decreased as interval between recordings increased to 320 msec. At 2s inter–recording intervals, there was little if any evidence of serial dependency in the EEGs from 14 subjects examined. (4) Passage of time per se (1 day to over a month) produced significant differences in several subjects. Presence of stimulus–evoked activity near 100 msec–latency was highly significant in 12 of 14 subjects.
Conclusions: :
Before v. after treatment design is feasible for single–case studies but one must deal with confounding effects associated with time.
Keywords: electrophysiology: clinical • clinical research methodology • electroretinography: clinical