Robust estimation for ordinal regression

Ordinal regression is used for modelling an ordinal response variable as a function of some explanatory variables. The classical technique for estimating the unknown parameters of this model is Maximum Likelihood (ML). The lack of robustness of this estimator is formally shown by deriving its breakdown point and its influence function. To robustify the procedure, a weighting step is added to the Maximum Likelihood estimator, yielding an estimator with bounded influence function. We also show that the loss in efficiency due to the weighting step remains limited. A diagnostic plot based on the Weighted Maximum Likelihood estimator allows to detect outliers of different types in a single plot.     

A monte carlo study of variable selection and inferences in a two stage random coefficient linear regression model with unbalanced data

A growth curve analysis is often applied to estimate patterns of changes in a given characteristic of different individuals. It is also used to find out if the variations in the growth rates among individuals are due to effects of certain covariates. In this paper, a random coefficient linear regression model, as a special case of the growth curve analysis, is generalized to accommodate the situation where the set of influential covariates is not known a priori. Two different approaches for seleaing influential covariates (a weighted stepwise selection procedure and a modified version of Rao and Wu’s selection criterion) for the random slope coefficient of a linear regression model with unbalanced data are proposed. Performances of these methods are evaluated by means of Monte-Carlo simulation. In addition, several methods (Maximum Likelihood, Restricted Maximum Likelihood, Pseudo Maximum Likelihood and Method of Moments) for estimating the parameters of the selected model are compared Proposed variable selection schemes and estimators are appliedtotheactualindustrial problem which motivated this investigation.     

Modified maximum likelihood estimation for rayleigh distribution

For defining a Modified Maximum Likelihood Estimate of the scale parameter of Rayleigh distribution, a hyperbolic approximation is used instead of linear approximation for a function which appears in the Maximum Likelihood equation. This estimate is shown to perform better, in the sense of accuracy and simplicity of calculation, than the one based on linear approximation for the same function. Also the estimate of the scale parameter obtained is shown to be asymptotically unbiased. Numerical computation for random samples of different sizes from Rayleigh distribution, using type I1 censoring is done and is shown to be better than that obtained by Lee et al. (1980)     

Maximum likelihood estimation of variance components of heteroscedastic random anova model

the estimation of variance components of heteroscedastic random model is discussed in this paper. Maximum Likelihood (ML) is described for one-way heteroscedastic random models. The proportionality condition that cell variance is proportional to the cell sample size, is used to eliminate the efffect of heteroscedasticity. The algebraic expressions of the estimators are obtained for the model. It is seen that the algebraic expressions of the estimators depend mainly on the inverse of the variance-covariance matrix of the observation vector. So, the variance-covariance matrix is obtained and the formulae for the inversions are given. A Monte Carlo study is conducted. Five different variance patterns with different numbers of cells are considered in this study. For each variance pattern, 1000 Monte Carlo samples are drawn. Then the Monte Carlo biases and Monte Carlo MSE’s of the estimators of variance components are calculated. In respect of both bias and MSE, the Maximum Likelihood (ML) estimators of variance components are found to be sufficiently good.     

An evaluation by simulation of alternative estimators for the marginal Fisher distribution

The Fisher distribution is a standard model for directional data (or spherical data). In some cases though, only the co-latitudes can be observed, resulting in a sample of observations from the corresponding marginal distribution. This paper reports on an extensive simulation to compare and evaluate the robustness of 11 test-statistics corresponding to various estimators of the parameters of this distribution. The estimators include Maximum Likelihood and Moment-type estimators, as well as sample means and variances based on approximations to the marginal Fisher distribution. Of the test-statistics considered, the Likelihood-Ratio statistic was the only one whose sampling distribution remained close to its asymptotic distribution for all parameter values and sample sizes considered. In general, the other statistics were close to their approximate distributions only when ksin2?0, was fairly large. The paper includes details on the computational methods for finding the Maximum Likelihood and Moment estimators, and concludes with some practical advice on the choice of estimation procedure.     

Bootstrap prediction intervals for autoregressive conditional duration models

In the recent past, the autoregressive conditional duration (ACD) models have gained popularity in modelling the durations between successive events. The aim of this paper is to propose a simple and distribution free re-sampling procedure for developing the forecast intervals of linear ACD Models. We use the conditional least squares method to estimate the parameters of the ACD Model instead of the conditional Maximum Likelihood Estimation or Quasi-Maximum Likelihood Estimation and show that they are consistent for large samples. The properties of the proposed procedure are illustrated by a simulation study and an application to two real data sets.     

DYNAMIC FACTOR MODELS

This paper introduces nonlinear dynamic factor models for various applications related to risk analysis. Traditional factor models represent the dynamics of processes driven by movements of latent variables, called the factors. Our approach extends this setup by introducing factors defined as random dynamic parameters and stochastic autocorrelated simulators. This class of factor models can represent processes with time varying conditional mean, variance, skewness and excess kurtosis. Applications discussed in the paper include dynamic risk analysis, such as risk in price variations (models with stochastic mean and volatility), extreme risks (models with stochastic tails), risk on asset liquidity (stochastic volatility duration models), and moral hazard in insurance analysis.

We propose estimation procedures for models with the marginal density of the series and factor dynamics parameterized by distinct subsets of parameters. Such a partitioning of the parameter vector found in many applications allows to simplify considerably statistical inference. We develop a two- stage Maximum Likelihood method, called the Finite Memory Maximum Likelihood, which is easy to implement in the presence of multiple factors. We also discuss simulation based estimation, testing, prediction and filtering.     

On Testing Sample Selection Bias Under the Multicollinearity Problem

This paper reviews and extends the literature on the finite sample behavior of tests for sample selection bias. Monte Carlo results show that, when the “multicollinearity problem” identified by Nawata (1993) is severe, (i) the t-test based on the Heckman-Greene variance estimator can be unreliable, (ii) the Likelihood Ratio test remains powerful, and (iii) nonnormality can be interpreted as severe sample selection bias by Maximum Likelihood methods, leading to negative Wald statistics. We also confirm previous findings (Leung and Yu, 1996) that the standard regression-based t-test (Heckman, 1979) and the asymptotically efficient Lagrange Multiplier test (Melino, 1982), are robust to nonnormality but have very little power.     

Inference Under Right Censoring in a Discrete Setup

In this article, we consider the right random censoring scheme in a discrete setup when the lifetime and censoring variables are independent and have geometric distributions with means 1/θ₁ and 1/θ₂, respectively. We first obtain the Maximum Likelihood and Method of Moment estimators of the unknown parameters. We also find the Bayes and Posterior Regret Gamma Minimax estimators of the parameters for the two cases when the prior distributions are dependent and independent, assuming a squared error loss function. We then discuss the Proportional Hazard model, and obtain Maximum Likelihood estimators of the unknown parameters and derive the Bayes estimators assuming squared error loss using Markov Chain Monte Carlo methods.     

Two sided acceptance sampling plans based on mml estimators

Alternative estimators that are robust to non-normality in the symmetric thick tailed situation are shown to yield much better results tl)an do [Xbar] and s. The particular estimators suggested in this paper are the Modified Maximum Likelihood estimators of Tiku (1967).     

Computation of maximum likelihood estimates for μ and β from a grouped sample of a normal population. A comparison of algorithms

Computation of Maximum Likelihood Estimates for μ and β from a grouped sample of a normal population is a numerical problem which occurs frequently in applied statistics. We investigate the performance of several algorithms for this problem (including Newton-Raphson, Method of Scoring, Expectation-Maximization and others) using simulated data. One of the main results is that Method of Scoring is best both in number of iterations and CPU time though no second-order derivatives are used.     

A Bivariate Gamma Model for a Latent Degradation Process

A model is presented in this article based on a bivariate gamma process in which, the first component is latent and determines the failure time and the second represents the marker. This process is a more realistic model for a degradation process. After introducing the model, we obtain failure and survival probability distributions and discuss parametric and predictive inferences based on the Maximum Likelihood method and in a Bayesian setup.     

An EM algorithm for multivariate Poisson distribution and related models

Multivariate extensions of the Poisson distribution are plausible models for multivariate discrete data. The lack of estimation and inferential procedures reduces the applicability of such models. In this paper, an EM algorithm for Maximum Likelihood estimation of the parameters of the Multivariate Poisson distribution is described. The algorithm is based on the multivariate reduction technique that generates the Multivariate Poisson distribution. Illustrative examples are also provided. Extension to other models, generated via multivariate reduction, is discussed.     

ESTIMATION OF PARAMETERS IN THE MARGINAL FISHER DISTRIBUTION1

The Fisher distribution is frequently used as a model for the probability distribution of directional data, which may be specified either in terms of unit vectors or angular co-ordinates (co-latitude and azimuth). If, in practical situations, only the co-latitudes can be observed, the available data must be regarded as a sample from the corresponding marginal distribution. This paper discusses the estimation by Maximum Likelihood (ML) and the Method of Moments of the two parameters of this marginal Fisher distribution. The moment estimators are generally simpler to compute than the ML estimators, and have high asymptotic efficiency.     

BOOK REVIEWS

Book reviewed in this article:
Shenton, L. R. and Bowman, K. O. Maximum Likelihood Estimation in Small Samples.
Bierman, G. J. Factorization Methods for Discrete Sequential Estimation
Kendall, Sir Maurice and Plackett, R. L. (Eds). Studies in the History of Statistics and Probability.     

Efficient von Mises–Fisher concentration parameter estimation using Taylor series

A simple method is outlined for constructing a Taylor series for the Maximum Likelihood Estimate of the von Mises–Fisher concentration parameter based around an initial heuristic estimate. While existing treatments require multiple computationally intensive calculations of a Bessel ratio, this method provides accurate results using only one such calculation. The accuracy of the method is tested extensively, and the reuse of the Taylor series for multiple calculations is explored.     

Errors in variables: consistent adjusted least squares (cals) estimation

A single equation errors-in-variables model is considered. Exact restrictions on the parameters in the model are assumed to be available such that the model is just-identified. A Consistent Adjusted Least Squares (CALS) estimator for this model is proposed and its asymptotic distribution is given. Special cases are given as illustrations. CALS is identical to the Method of Moments (MM), and to Maximum Likelihood (ML) under the structural interpretation. Under the functional interpretation it is identical to ML in cases where the latter method is consistent.     

Estimation of nonlinear CUB models via numerical optimization and EM algorithm

The analysis of human perceptions is often carried out by resorting to surveys and questionnaires, where respondents are asked to express ratings about the objects being evaluated. A class of mixture models, called CUB (Combination of Uniform and shifted Binomial), has been recently proposed in this context. This article focuses on a model of this class, the Nonlinear CUB, and investigates some computational issues concerning parameter estimation, which is performed by Maximum Likelihood. More specifically, we consider two main approaches to optimize the log-likelihood: the classical numerical methods of optimization and the EM algorithm. The classical numerical methods comprise the widely used algorithms Nelder–Mead, Newton–Raphson, Broyden–Fletcher–Goldfarb–Shanno (BFGS), Berndt–Hall–Hall–Hausman (BHHH), Simulated Annealing, Conjugate Gradients and usually have the advantage of a fast convergence. On the other hand, the EM algorithm deserves consideration for some optimality properties in the case of mixture models, but it is slower. This article has a twofold aim: first we show how to obtain explicit formulas for the implementation of the EM algorithm in nonlinear CUB models and we formally derive the asymptotic variance–covariance matrix of the Maximum Likelihood estimator; second, we discuss and compare the performance of the two above mentioned approaches to the log-likelihood maximization.     

Estimation and identification considerations for the multiconsequence intervention model

A General Multiconsequence Intervention Model class that describes the simultaneous occurrence of a change in the process mean and covariance structure is introduced. When the covariance change is negligible, this model class reduces to intervention models described by Box and Tiao (1975). Maximum Likelihood Estimators for the parameters of the multiconsequence model class are developed for various important modeling situations that result from different a priori information about the form of the mean shift function form and the model parameters. As a consequence of these estimation results, an identification procedure for determining an appropriate dynamic mean shift form is suggested. The necessary hypothesis tests and corresponding confidence intervals.