Multivariate analysis of data gathered from generally balanced designs
Nelder (1965a, b, 1968) introduced the notion of general balance for designed experiments and obtained combined-strata treatment effects. This was then generalized by Houtman & Speed (1983) to all univariate block designs for some treatment structure. This thesis extends the theory to multivariate designed experiments, including the estimation of combined-strata treatment effects, and their confidence regions from canonical variate analysis. A wide range of designs types are worked through, with their MANOVA table, and a numerical example given in each case. Using a Monte Carlo simulation for a balanced incomplete block design (BIBD), the combined strata confidence regions are found to significantly outperform those from intra-block information alone
Thesis, Dissertation, English, 1996
1996