Abstract
The problems of detecting influential observations and collinearity in multiple linear regression are discussed. The commonly used diagnostics of influential observations and collinearity are critically appraised and summarized as a guide for data analysts in the selection and use of diagnostic methodologies. Several significant results are obtained and show that some diagnostic procedures are superior to others due to greater detection ability and protection against spurious indications. The methods of identifying extreme observations is generalized and unified. A graphical procedure is developed for the detection of extreme observations. Additionally two new modes of influence are presented and diagnostic procedures are developed to detect deviant observations in these modes of influence. Several illustrative examples are presented.
Dunn, Mark Raymond (1982). Regression diagnostics. Doctoral dissertation, Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -147530.