Norm restricted maximum likelihood estimators for binary regression models with parametric link

Parametric link transformation families have shown to be useful in the analysis of binary regression data since they avoid th? problem of link misspecifaction. Inference for these models are commonly based on likelihood methods. Duffy and Santner (1988, 1989) however showed that ordinary logistic maximum likelihood estimators (MLE) have poor mean square error (MSE) behavior in small samples compared to alternative norm restricted estimators. This paper extends these alternative norm restricted estimators to binary regression models with any specified parametric link family. These extended norm restricted MLE's are strongly consistent and efficient under regularity conditions. Finally a simulation study shows that an empiric version of norm restricted MLE's exhibit superior MSE behavior in small samples compared to MLE's with fixed known link.     

General saddlepoint approximations to distributions under an elliptical population

General saddlepoint approximations are derived for the distributions of statistics under an elliptical population. The technique is applied to obtain the tail probabilities of latent roots of a sample covariance matrix. It is shown that the method based on normalizing transformations by Tsuchiya and Konishi (1997) is efficient for the sample correlation coefficient in an elliptical sample.     

A Transformation Characterizing the Normal Distribution

In this article we propose a new transformation of random variables (RVs) which characterizes the normal distribution. It allows us to transform n i.i.d. normal RVs whose mean and variance are unknown into new n ? 2 i.i.d. new normal variables with zero mean while maintaining the same unknown variance. This belongs to the class of transformations designed to reduce the number of unknown parameters or remove them altogether.

Some historical remarks concerning methods for removing parameters in the normal distribution are given and two possible applications of the new transformation are described.     

Double samble tests— a distribution free procedure

A method for constructing two-stage (double samble) tests is presented which does not require the evaluation of complicated bivariate distribution function. The procedure results from a modification of Fisher's method for combining independent tests of significance and is distribution free in the way it combines the test results from the two sampies. However, the one sample test statistics for the two samples are assumed to have continuous distributions and may be parametric. A rule is also given or the selection of a particular test out of a family of possible two-stage tests which can be generated by this method. Specific examples are given and comparisons are made with two double sample tests which have previously been presented in the literature.     

Model Selection When a Key Parameter is Constrained to be in an Interval

This article considers model selection procedures based on choosing the model with the largest maximized log-likelihood minus a penalty, when key parameters are restricted to be in a closed interval. Its main emphasis is how these penalties might be chosen in small samples to give good properties of the resultant procedure. We illustrate two model selection problems in the context of Box–Cox transformations and their application to the linear regression model. Simulation results for both problems indicate that the new procedure clearly dominates existing procedures in terms of having higher probabilities of correctly selecting the true model.     

The starship for point estimates and confidence intervals on a mean and for percentiles

The starship procedure for transformations to normality described by Owen (1988) is implemented using the Johnson (1949) System of transformations, the Slifker-Shapiro (1980) technique for choosing a transformation, and the Shapiro-Wilk (1965) test for normality. This procedure was applied to obtain maximum likelihood point estimates of a mean, to obtain confidence intervals on a mean, and to estimate percentiles of a distribution based on a sample. Simulations of three distributions show that the starship has many desirable properties, and can be compared very favorably with the bootstrap procedures of Efron (1987).     

Variance Reduction Using Nonlinear Controls and Transformations

Nonlinear regression-adjusted control variables are investigated for improving variance reduction in statistical and system simulations. To this end, simple control variables are piecewise sectioned and then transformed using linear and nonlinear transformations. Optimal parameters of these transformations are selected using linear or nonlinear least-squares regression algorithms. As an example, piecewise power-transformed variables are used in the estimation of the mean for the twovariable Anderson-Darling goodness-of-fit statistic W ₂ ². Substantial variance reduction over straightforward controls is obtained. These parametric transformations are compared against optimal, additive nonparametric transformations obtained by using the ACE algorithm and are shown, in comparison to the results from ACE, to be nearly optimal.     

Robust and diagnostic regression analyses

Graphical methods of diagnostic regression analysis are applied to three examples in which least squares and robust regression analyses give substantially different results. The diagnostic tools lead to the identification of data deficiencies and model inadequacies. The analyses serve as a reminder that robust regressions depend upon the linear model and upon the scale in whicli the response is analysed. The robust analysis may also be sensitive to gross errors in one or more explanatory variables