WHEN:
Thursday, January 20, 2005 at 2:30 PM
WHERE:
Durham Science Center, Room 255
WHAT:
Faculty Candidate
will give a talk on
ABSTRACT:
Crop researchers performing germplasm screenings are often unable to
replicate their plots due to scarcity of seed and the large numbers of
genotypes being evaluated. The incorporation of known "check" varieties is a
common method of overcoming the difficulties associated with unreplicated
trials. However, the researcher who chooses to use an unreplicated design must
decide on the number of checks to include in the study (i.e., check plot
density) and whether to regard treatment as a fixed or random effect. In this
simulation study, we explore the effects of check plot density on the
efficiency of the resulting analysis. We also explore the effect of analyzing
treatments as random versus fixed. Our study considers ten different designs
with check densities ranging from 5% to 50%. The designs and analyses are then
compared on the basis of the correlation of the actual treatment effects with
each of the following: "observed" yield, least square means (LSMEANs) for
fixed treatments, and best linear unbiased predictors (BLUPs) for random
treatments. Finally, we observe the frequency with which the analysis ranks
the top 10% of the treatments within the top 15% of the LSMEANs or BLUPs.
Back to the Mathematics Colloquium Page