The logarithm of the relative risk function in a proportional hazards model involving one or more possibly time-dependent covariates is treated as a specified sum of a constant term, main effects, and selected interaction terms. Maximum partial likelihood estimation is used, where the maximization is taken over a suitably chosen finite-dimensional estimation space, whose dimension increases with the sample size and which is constructed from linear spaces of functions of one covariate and their tensor products. The $L_2$ rate of convergence for the estimate and its ANOVA components is obtained. An adaptive numerical implementation is discussed, whose performance is compared to (full likelihood) hazard regression both with and without the restriction to proportional hazards.

In this work we study additive dynamic regression models for longitudinal data. These models provide a flexible and nonparametric method for investigating the time-dynamics of longitudinal data. The methodology is aimed at data where measurements are recorded at random time points. We model the conditional mean of responses given the full internal history and possibly time-varying covariates. We derive the asymptotic distribution for a new nonparametric least squares estimator of the cumulative time-varying regression functions. Based on the asymptotic results, confidence bands may be computed and inference about time-varying coefficients may be drawn. We propose two estimators of the cumulative regression function. One estimator that involves smoothing and one that does not. The latter, however, has twice the variance as the smoothing based estimator. Goodness of fit of the model is considered using martingale residuals. Finally, we also discuss how partly-conditional mean models in which the mean of the response is regressed onto selected time-varying covariates may be analysed in the same framework. We apply the methods to longitudinal data on height development for cystic fibrosis patients.

Estimation of finite population totals in the presence of auxiliary information is considered. A class of estimators based on local polynomial regression is proposed. Like generalized regression estimators, these estimators are weighted linear combinations of study variables, in which the weights are calibrated to known control totals, but the assumptions on the superpopulation model are considerably weaker. The estimators are shown to be asymptotically design-unbiased and consistent under mild assumptions. A variance approximation based on Taylor linearization is suggested and shown to be consistent for the design mean squared error of the estimators. The estimators are robust in the sense of asymptotically attaining the Godambe–Joshi lower bound to the anticipated variance. Simulation experiments indicate that the estimators are more efficient than regression estimators when the model regression function is incorrectly specified, while being approximately as efficient when the parametric specification is correct.

We consider the problem of testing monotonicity of the regression function in a nonparametric regression model. We introduce test statistics that are functionals of a certain natural $U$-process. We study the limiting distribution of these test statistics through strong approximation methods and the extreme value theory for Gaussian processes. We show that the tests are consistent against general alternatives.

For the problem of estimating a regression function, $\mu$ say, subject to shape constraints, like monotonicity or convexity, it is argued that the divergence of the maximum likelihood estimator provides a useful measure of the effective dimension of the model. Inequalities are derived for the expected mean squared error of the maximum likelihood estimator and the expected residual sum of squares. These generalize equalities from the case of linear regression. As an application, it is shown that the maximum likelihood estimator of the error variance $\sigma^2$ is asymptotically normal with mean $\sigma^2$ and variance $2\sigma_2/n$. For monotone regression, it is shown that the maximum likelihood estimator of $\mu$ attains the optimal rate of convergence, and a bias correction to the maximum likelihood estimator of $\sigma^2$ is derived.

Gaussian mixtures provide a convenient method of density estimation that lies somewhere between parametric models and kernel density estimators.When the number of components of the mixture is allowed to increase as sample size increases, the model is called a mixture sieve.We establish a bound on the rate of convergence in Hellinger distance for density estimation using the Gaussian mixture sieve assuming that the true density is itself a mixture of Gaussians; the underlying mixing measure of the true density is not necessarily assumed to have finite support. Computing the rate involves some delicate calculations since the size of the sieve—as measured by bracketing entropy—and the saturation rate, cannot be found using standard methods.When the mixing measure has compact support, using $k_n \sim n^{2/3}/(\log n)^{1/3}$ components in the mixture yields a rate of order $(\log n)^{(1+\eta)/6}/n^{1/6}$ for every $\eta > 0$. The rates depend heavilyon the tail behavior of the true density.The sensitivity to the tail behavior is dimin- ished byusing a robust sieve which includes a long-tailed component in the mixture.In the compact case,we obtain an improved rate of $(\log n/n)^{1/4}$. In the noncompact case, a spectrum of interesting rates arise depending on the thickness of the tails of the mixing measure.

Two coupled Gibbs sampler chains, both with invariant probability density $p$, are run in parallel so that the chains are negatively correlated.We define an asymptotically unbiased estimator of the $\pi$-expectation $E(f(\mathbf(X))$ which achieves significant variance reduction with respect to the usual Gibbs sampler at comparable computational cost. The variance of the estimator based on the new algorithm is always smaller than the variance of a single Gibbs sampler chain, if $\pi$ is attractive and $f$ is monotone nondecreasing in all components of $\mathbf{X}$. For nonattractive targets $\pi$, our results are not complete: The new antithetic algorithm outperforms the standard Gibbs sampler when $\pi$ is a multivariate normal density or the Ising model. More generally, nonrigorous arguments and numerical experiments support the usefulness of the antithetically coupled Gibbs samplers also for other nonattractive models. In our experiments the variance is reduced to at least a third and the efficiency also improves significantly.

An extension of the class of GS-LCI normal models introduced by Andersson and Madsen is defined and studied. The models are defined in terms of symmetry restrictions given by a finite group and conditional independence restrictions given by an acyclic directed graph. Maximum likelihood estimation of the parameters in the models is discussed.

We study the geometry of the parameter space for Bayesian directed graphical models with hidden variables that have a tree structure and where all the nodes are binary.We show that the conditional independence statements implicit in such models can be expressed in terms of polynomial relationships among the central moments.This algebraic structure will enable us to identify the inequality constraints on the space of the manifest variables that are induced by the conditional independence assumptions as well as determine the degree of unidentifiability of the parameters associated with the hidden variables. By understanding the geometry of the sample space under this class of models we shall propose and discuss simple diagnostic methods.

Multivariate Totally Positive $(MTP_2)$ binary distributions have been studied in many fields, such as statistical mechanics, computer storage and latent variable models. We show that $MTP_2$ is equivalent to the requirement that the parameters of a saturated log-linear model belong to a convex cone, and we provide a Fisher-scoring algorithm for maximum likelihood estimation.We also show that the asymptotic distribution of the log-likelihood ratio is a mixture of chi-squares (a distribution known as chi-bar-squared in the literature on order restricted inference); for this we derive tight bounds which turn out to have very simple forms. A potential application of this method is for Item Response Theory (IRT) models, which are used in educational assessment to analyse the responses of a group of subjects to a collection of questions (items): an important issue within IRT is whether the joint distribution of the manifest variables is compatible with a single latent variable representation satisfying local independence and monotonicity which, in turn, imply that the joint distribution of item responses is $MTP_2$.

The S-functionals of multivariate location and scatter, includingthe MVE-functionals, are known to be uniquely defined only at unimodal elliptically symmetric distributions. The goal of this paper is to establish the uniqueness of these functionals under broader classes of symmetric distributions. We also discuss some implications of the uniqueness of the functionals and give examples of striclty unimodal and symmetric distributions for which the MVE-functional is not uniquely defined. The uniqueness results for the S-functionals are obtained by embedding them within a more general class of functionals which we call the M-functionals with auxiliary scale. The uniqueness results of this paper are then obtained for this class of multivariate functionals. Besides the S-functionals, the class of multivariate M-functionals with auxiliary scale include the constrained M-functionals recently introduced by Kent and Tyler, as well as a new multivariate generalization of Yohai’s MM-functionals.