A method for assessment of optimal choice of parametric model in least squares collocation

Abstract

The process of choosing a suitable parametric model prior least squares collocation suffers from a high degree of arbitrariness. Although, the congruency test (wherein〖σ^2〗_0= σ^2)gives an overall impression on the validity of the model, but further testing is always required, even when the congruency test points to the contrary, since it is possible that effects from different modeling errors cancel each other out, in the computation of(σ^2 ) ̂. Besides, the procedure for estimating the variance components from the parametric model is computationally tasking. A semi automated assessment procedure is therefore herein presented by considering some rudimentary statistics of residuals followed by a student’s t- test on the mean of residuals. The model is simpler as it eliminates the need for variance component estimation and faster to implement compared to when strict reliance is placed on the congruency test in parametric model selection