Preprint #96-19
Generalization of the TLS Approach in the
Errors-in-Variables Problem
by
Yasuo Amemiya
Abstract
The total least squares (TLS) method is an appropriate estimation
procedure for the errors-in-variables model with a single linear relationship
and with the error covariance matrix either known or known
up to a multiple. This type of the errors-in-variables model can be
extended in a number of ways keeping the applicability of the basic TLS
approach in some form. The extensions considered here include models
with a nonlinear relationship, with multiple relationships, and with
an unknown error covariance matrix. Proper estimation procedures
for the extended models can be considered generalizations of the basic
TLS methods. For the models with a known error covariance matrix
and with either a nonlinear relationship or multiple linear relationships,
the estimation procedures are discussed from the TLS point of view.
For the factor analysis model with multiple linear relationships and an
unknown diagonal error covariance matrix, it is described how a proper
estimation method can be carried out using an iterated version of the
weighted TLS method with iteratively estimated weights. Also, the
nonlinear factor analysis problem, an area of current statistical
research interest, is discussed. An estimation procedure which can be
considered a generalization of the TLS method is presented for the nonlinear
factor analysis model with multiple nonlinear relationships and an unknown
diagonal error covariance matrix.
Copies of preprints are available from the author upon request. Use
the preprint number (located at the top of the page) and
make the request directly to the author, Iowa State
University,
Department of Statistics, Snedecor Hall, Ames, IA 50011-1210.