Barycenter and maximum likelihood

We refine recent existence and uniqueness results, for the barycenter of points at infinity of Hadamard manifolds, to measures on the sphere at infinity of symmetric spaces of non compact type and, more specifically, to measures concentrated on single orbits. The barycenter will be interpreted as the maximum likelihood estimate (MLE) of generalized Cauchy distributions on Furstenberg boundaries. As a spin-off, a new proof of the general Knight-Meyer characterization theorem will be given.     

Nonlinearity effects in multidimensional scaling

When multidimensional scaling of n cases is derived from dissimilarities that are functions of p basic continuous variables, the question arises of how to relate the values of the variables to the configuration of n points. We provide a methodology based on nonlinear biplots that expresses nonlinearity in two ways: (i) each variable is represented by a nonlinear trajectory and (ii) each trajectory is calibrated by an irregular scale. Methods for computing, calibrating and interpreting these trajectories are given and exemplified. Not only are the tools of immediate practical utility but the methodology established assists in a critical appraisal of the consequences of using nonlinear measures in a variety of multidimensional scaling methods.     

Two level minimization in multidimensional scaling

Multidimensional scaling with city block norm in embedding space is considered. Construction of the corresponding algorithm is reduced to minimization of a piecewise quadratic function. The two level algorithm is developed combining combinatorial minimization at upper level with local minimization at lower level. Results of experimental investigation of the efficiency of the proposed algorithm are presented as well as examples of its application to visualization of multidimensional data.     

Exact maximum likelihood estimator of the structure of a stratified population

An exact expression for the extreme values of the integer vector that maximize the function

Approximate maximum likelihood estimation in linear regression

The application of the ML method in linear regression requires a parametric form for the error density. When this is not available, the density may be parameterized by its cumulants ( _i ) and the ML then applied. Results are obtained when the standardized cumulants ( _i ) satisfy _i = _i+2/ ₂ ^(i+2)/2 =O(v ⁱ ) asv 0 fori>0.Research financed in part by the Research Center of the Athens University of Economics and Business.     

Maximum likelihood estimation for birth and multidimensional Brownian process

The individuals of a population reproduce according to a linear birth process and simultaneously and independently diffuse inp-dimensional Euclidean space according to a Brownian process with dependent coordinates. The properties of the maximum likelihood estimators of the drift vector and the matrix of diffusion coefficients are investigated under two sampling schemes involving incomplete observations.     

Uniform LAN condition of planar Gibbsian point processes and optimality of maximum likelihood estimators of soft-core potential functions

Summary A parametric model of planar point patterns in a bounded region is constructed using grand canonical Gibbsian point processes with soft-core potential functions. A simple and explicit condition that this model becomes a uniform locally asymptotic normal (ULAN) family will be given. From this result we can conclude that the maximum likelihood estimator of the potential function is asymptotically efficient for a wide class of loss functions.     

The asymptotic equivalence of Bayes and maximum likelihood estimation

Let (X, $\text{A}$) be a measurable space, Θ ? $\text{R}$ an open interval and P_Ω ∥ $\text{A}$, Ω ? Θ, a family of probability measures fulfilling certain regularity conditions. Let $\text{Ω}{}_{\mathrm{n}}$ be the maximum likelihood estimate for the sample size n. Let λ be a prior distribution on Θ and let $\text{R}{}_{\mathrm{<mtext>n,</mtext><mtext>x</mtext>}}$ be the posterior distribution for the sample size n given $\text{x}\text{? X}{}^{\mathrm{n}}$. $\text{L:}\text{Θ}\text{× Θ →}\text{R}$ denotes a loss function fulfilling certain regularity conditions and T_n denotes the Bayes estimate relative to λ and L for the sample size n. It is proved that for every compact K ? Θ there exists c_K ≥ 0 such that

This theorem improves results of Bickel and Yahav [3], and Ibragimov and Has'minskii [4], as far as the speed of convergence is concerned.     

Minimum divergence estimators,maximum likelihood and exponential families

The dual representation formula of the divergence between two distributions in a parametric model is presented. Resulting estimators do not make use of any grouping or smoothing. For smooth divergences they all coincide with the MLE on any regular exponential family.     

The use of generalized inverses in restricted maximum likelihood

The calculus of generalized inverses and related concepts in matrix algebra is applied to the general restricted maximum likelihood problem. Some new results on g-inverses, Kronecker products, and matrix differentials are presented. For the restricted maximum likelihood problem we obtain generalizations of the well-known results of Aitchison and Silvey [1]. We use the approach recently developed by Heijmans and Magnus [13, 14] to allow for non-i.i.d. observations. A nonlinear seemingly unrelated regressions model with possibly singular covariance matrix and linear restrictions (NLSURSR) is analyzed, and the linear expenditure system (LES) is discussed as a special case.     

Optimal tuning parameter estimation in maximum penalized likelihood method

In maximum penalized or regularized methods, it is important to select a tuning parameter appropriately. This paper proposes a direct plug-in method for tuning parameter selection. The tuning parameters selected using a generalized information criterion (Konishi and Kitagawa, Biometrika, 83, 875–890, 1996) and cross-validation (Stone, Journal of the Royal Statistical Society, Series B, 58, 267–288, 1974) are shown to be asymptotically equivalent to those selected using the proposed method, from the perspective of estimation of an optimal tuning parameter. Because of its directness, the proposed method is superior to the two selection methods mentioned above in terms of computational cost. Some numerical examples which contain the penalized spline generalized linear model regressions are provided.     

Parallel branch and bound for multidimensional scaling with city-block distances

Multidimensional scaling is a technique for exploratory analysis of multidimensional data. The essential part of the technique is minimization of a multimodal function with unfavorable properties like invariants and non-differentiability. Recently a branch and bound algorithm for multidimensional scaling with city-block distances has been proposed for solution of medium-size problems exactly. The algorithm exploits piecewise quadratic structure of the objective function. In this paper a parallel version of the branch and bound algorithm for multidimensional scaling with city-block distances has been proposed and investigated. Parallel computing enabled solution of larger problems what was not feasible with the sequential version.     

Branch and bound algorithm for multidimensional scaling with city-block metric

A two level global optimization algorithm for multidimensional scaling (MDS) with city-block metric is proposed. The piecewise quadratic structure of the objective function is employed. At the upper level a combinatorial global optimization problem is solved by means of branch and bound method, where an objective function is defined as the minimum of a quadratic programming problem. The later is solved at the lower level by a standard quadratic programming algorithm. The proposed algorithm has been applied for auxiliary and practical problems whose global optimization counterpart was of dimensionality up to 24.     

Experimental comparison of least-squares and maximum likelihood methods in factor analysis

Three popular methods to estimate the unknown parameters in the factor analysis model, simple (SLS) and weighted (WLS) least-squares methods and the maximum likelihood method (ML), are compared by a Monte Carlo study. The experiments were conducted with 200 replications for every combination of levels of the following three conditions: method (3 levels), sample size (3 levels) and uniquenesses (2 levels). It was found that SLS performed most favorably when the sample size is relatively small and unique variances are relatively large. WLS and ML proved to be rather alike.     

maxLik: A package for maximum likelihood estimation in R

This paper describes the package maxLik for the statistical environment R. The package is essentially a unified wrapper interface to various optimization routines, offering easy access to likelihood-specific features like standard errors or information matrix equality (BHHH method). More advanced features of the optimization algorithms, such as forcing the value of a particular parameter to be fixed, are also supported.     

Nonparametric maximum likelihood density estimation and simulation-based minimum distance estimators

Indirect inference estimators (i.e., simulation-based minimum distance estimators) in a parametric model that are based on auxiliary nonparametric maximum likelihood density estimators are shown to be asymptotically normal. If the parametricmodel is correctly specified, it is furthermore shown that the asymptotic variance-covariance matrix equals the inverse of the Fisher-information matrix. These results are based on uniform-in-parameters convergence rates and a uniform-inparameters Donsker-type theorem for nonparametric maximum likelihood density estimators.     

Limit theorems for maximum likelihood estimators in the Curie-Weiss-Potts model

The Curie-Weiss-Potts model, a model in statistical mechanics, is parametrized by the inverse temperature β and the external magnetic field h. This paper studies the asymptotic behavior of the maximum likelihood estimator of the parameter β when h = 0 and the asymptotic behavior of the maximum likelihood estimator of the parameter h when β is known and the true value of h is 0. The limits of these maximum likelihood estimators reflect the phase transition in the model; i.e., different limits depending on whether β < β_c, β = β_c or β > β_c, where β_c ε (0, ∞) is the critical inverse temperature of the model.     

Probabilities of large and moderate deviations for maximum likelihood estimates

One proves theorems on the probabilities of large and moderate deviations for maximum likelihood estimates while for less restrictive assumptions one establishes coarse asymptotics for probabilities of large deviations in the regular case. Similar results are obtained in the case of observations in white Gaussian noise.Translated from Zapiski Nauchnykh Seminarov Leningradskogo Otdeleniya Matematicheskogo Instituta im. V. A. Steklova AN SSSR, Vol. 108, pp. 154–169, 1981.The author expresses his deep gratitude to Io Ao Ibragimov fort he formulation of the problem, for his constant care and for useful advice.