Patterns in Residuals in the Two-Way Layout
作者:
Cuthbert Daniel,
期刊:
Technometrics
(Taylor Available online 1978)
卷期:
Volume 20,
issue 4
页码: 385-395
ISSN:0040-1706
年代: 1978
DOI:10.1080/00401706.1978.10489692
出版商: Taylor & Francis Group
关键词: Residuals;Non-additivity;Interactions;Analysis of variance;Factorial representation;Two-way layout;Pattern recognition
数据来源: Taylor
摘要:
Non-additivity may be present in only a few cells in an unreplicated, fixed effects,R×C. two-way table of data. When this is the case, it is wasteful and confusing to summarize the interaction as a mean square with (R− 1)(C− 1) degrees of freedom, or as set ofRCresiduals. When the disturbances of additivity are inG<R/2 rows and inH<C/2 columns, the expected values of the resulting residuals are in recognizable patterns which may suffice to spot and to estimate the disturbances,Dij. The largest residuals nearly always indicate the largest disturbances, but others of the same sign may be concealed in the same rows (or columns). The main result is the pair of equations:Dij=yij− yij′− yi′j+ yi′j′=rij− rij′− ri′j+ ri′j′. where yijis the mean of the (C−H) observations in row i but not in columns with disturbances: yi′jis the mean of the (R–G) observations in columns j but not in rows with disturbances: yi′j′is the mean of the (R−G)(C−H) observations not in the disturbed rows or columns: ther's are the usual residuals under the assumption of additive row- and column-effects. An example is given.
点击下载:
PDF (905KB)
返 回