Parameter orthogonalization for the Siler mortality model

Correlation and, in general, close relationships between parameters can cause problems in the estimation of a model and the consequent fluctuation in the trend of its coefficients. We show the connections existing between parameters in the Siler model, one of the most widely used in demography to approximate mortality over the entire life span, and propose a method to reduce them. Parameter orthogonalization via the Gram-Schmidt-Fisher scoring algorithm seems a promising technique for limiting identification issues and numerical instabilities often encountered when maximizing the likelihood.