In behavioural cognitive effects studies, baseline (pretest) group differences may confound post test group effect size comparisons. To account for this, stratification methods on baseline measures can be applied when assigning subjects to groups. However, baseline cognitive tests measures are inherently noisy on novels tasks, and much of the variance can be ascribed to random fluctuations due to good/bad test day, novelty of technical skills required for the task etc.
In a current study, we're administering 6 weeks of computer based cognitive training, which primarily requires visuospatial working memory and attention functions. The population is 100 normal subjects between 60 and 75 years of age, gender and education background matched. The design has 4 groups, two active and two control.
Regarding assignment of subjects to groups, we're currently considering the mentioned stratification on baseline. However, since these measures as so noisy I'm inclined to choose a randomised design approach, with (combined) stratification on stationary measures, such as educational background, age or gender.
Any advice, experience, and references on this matter will be highly appreciated - thanks.