Abstract
dc:description.abstractA small sheep experiment (nobs=32) planned to use a randomized complete block design (RCBD) treatment assignment of two binary factors. Complications creating the RCBD blocks prompted the researchers to discard the original blocks from the initial analysis plan and to rearrange their experimental units into new groups using linear covariate adjustment. We compare the blocks from the experiment's initial analysis plan and the groups from the researcher's linear covariate adjustment to groups formed by potential matching methods. We evaluate these three analysis approaches on the original sheep dataset and on simulated sheep datasets. We find that the groups created using matching methods produce less precise estimates and that further, those estimates may be biased. Additionally, the matching methods may alter the experiment's size and thus, its overall power. When small RCBD experiments have complications forming the desired blocks, we recommend the joint use of well-established preliminary testing and post-stratification procedures. This acts as a more formalized version of the sheep researchers' use of linear covariate adjustment and implicit model selection.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (PhD)
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Mathematical Sciences
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Sotherden, Elaine
- Contributors dc:contributor
-
- William C Bridges, Jr, Committee Chair
- Patrick Gerard
- Brook Russell
- Matthew J Saltzman
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://open.clemson.edu/all_dissertations/2279