M. Malyutov
REAGAN Estimates of a Quasilinear
Regression
We study
Design and Analysis of non-linear in parameters smooth Quasilinear
Regression Model (abbreviated as Q-model) which admits in every iteration
almost the same type of analysis as linear regression for simultaneous
iterative estimation of both the mean and variance of observations. Q-models
are more flexible in applications than Linear models.
We prove local convergence of iterations θs, s ∈ Z,
in probability and the asymptotic normality of θˆ = lims→∞ θs in iteration
procedure called REweighted Algorithm of GAuss-Newton and illustrate its broad use by many
appropriate applications.
КЛЮЧЕВЫЕ СЛОВА: Gauss-Newton algorithm, iterations, convergence in probability, asymptotic normality, maximum likelihood, exponential families