Date of Award

Summer 2002

Document Type


Degree Name

Doctor of Philosophy (PhD)


Mathematics and Statistics


Computational and Applied Mathematics

Committee Director

John P. Morgan

Committee Member

N. Rao Chaganty

Committee Member

Dayanand N. Naik

Committee Member

John Stufken


One of the fundamental principles of experimental design is the separation of heterogeneous experimental units into subsets of more homogeneous units or blocks in order to isolate identifiable, unwanted, but unavoidable, variation in measurements made from the units. Given v treatments to compare, and having available b blocks of k experimental units each, the thoughtful statistician asks, “What is the optimal allocation of the treatments to the units?” This is the basic block design problem. Let nij be the number of times treatment i is used in block j and let N be the v x b matrix N = (nij). There is now a considerable body of optimality theory for block design settings where binarity (all nij ∈ {0, 1}), and symmetry or near-symmetry of the concurrence matrix NNT, are simultaneously achievable. Typically the same classes of designs are found to be best using any of the standard optimality criteria. Among these are the balanced incomplete block designs (BIBDs), many species of two-class partially balanced incomplete block designs, and regular graph designs.

However, there are triples (v, b, k) in which binarity precludes near-symmetry. For these combinatorially problematic settings, recent explorations have resulted in new optimality results and insight into the combinatorial issues involved. Of particular interest are the irregular BIRD settings, that is, triples (v, b, k) where the necessary conditions for a BIBD are fulfilled but no such design exists. A thorough study of the smallest such setting, (15,21,5), has produced some surprising optimal designs which will be presented in the first chapter of this document.

An incomplete block design is said to be resolvable if the blocks can be partitioned into classes, or replicates such that each treatment appears in exactly one block of each replicate. Resolvable designs are indispensable in many industrial and agricultural experiments, especially when the entire experiment can not be completed at one time or when there is a risk that the experiment may be prematurely terminated. In chapters two and three we will investigate the classes of resolvable designs having five or fewer replications and two blocks of possibly unequal size per replicate. Theory for identifying the best designs with respect to important optimality criteria will be developed, and with the optimality theory in hand, optimal designs will be identified and constructions provided. We will conclude with a comment on the robustness of resolvable designs to the loss of a replicate.