作者pei16 (^^)
看板NCTU-STAT98G
标题[演讲公告] 12/25 统计所专题演讲
时间Mon Dec 21 22:40:34 2009
※ [本文转录自 NCTU-STAT97G 看板]
作者: pei16 (^^) 看板: NCTU-STAT97G
标题: [演讲公告] 12/25 统计所专题演讲
时间: Mon Dec 21 22:34:46 2009
交通大学、清华大学 统计学研究所 专题演讲
题 目：D-optimal Partially Replicated Two-Level Factorial Designs
主讲人：廖振铎教授（台湾大学农艺所生物统计组）
时 间：98年12月25日(星期五)上午11:10-12:00
(上午10:50-11:10茶会於交大统计所429室举行)
地 点：交大综合一馆427室
Abstract
At the early stages of a factorial experiment, unreplicated fractional
two-level designs are commonly used to identify important or active effects.
Under the situation that there is no prior information available on which
effects might be active, minimum aberration designs may serve as reasonable
choices for gaining more information about a large set of potential effects.
However, the analysis methods for unreplicated data may perform
unsatisfactorily in identifying truly active effects, particularly when the
effect sparsity principle does not hold. This is due mainly to the lack of a
replication-based estimate of the error variance. Therefore, when the prior
information is provided and the set of possibly active effects contains all
the potential effects. We may first find an economical design, not
necessarily a minimum aberration design, for estimating the specified
possibly active effects. If some additional runs remain, then we can consider
running repeated treatment combinations to obtain a realistic estimate of
experimental error, which is used to test whether the specified possibly
active effects are truly active. The partially replicated two-level factorial
designs usually work well regardless of the effect sparsity. In this talk, we
will discuss D-optimal partially replicated designs derived from
parallel-flats designs and Hadamard matrices.
Keywords: Parallel-flats design; Hadamard matrix; orthogonal array;
projection property; pure error.
--
※ 发信站: 批踢踢实业坊(ptt.cc)
◆ From: 140.113.252.129
--
※ 发信站: 批踢踢实业坊(ptt.cc)
◆ From: 140.113.252.129