A general minimum distance estimation procedure is presented for nonstationary time series models that have an evolutionary spectral representation. The asymptotic properties of the estimate are derived under the assumption of possible model misspecification. For autoregressive processes with time varying coefficients, the estimate is compared to the least squares estimate. Furthermore, the behavior of estimates is explained when a stationary model is fitted to a nonstationary process.

We derive minimax rates for estimation in anisotropic smoothness classes. These rates are attained by a coordinatewise thresholded wavelet estimator based on a tensor product basis with separate scale parameter for every dimension. It is shown that this basis is superior to its one-scale multiresolution analog, if different degrees of smoothness in different directions are present.

As an important application we introduce a new adaptive waveletestimator of the time-dependent spectrum of a locally stationary time series. Using this model which was recently developed by Dahlhaus, we show that the resulting estimator attains nearly the rate, which is optimal in Gaussian white noise, simultaneously over a wide range of smoothness classes. Moreover, by our new approach we overcome the difficulty of how to choose the right amount of smoothing, that is, how to adapt to the appropriate resolution, for reconstructing the local structure of the evolutionary spectrum in the time-frequency plane.

A central limit theorem is established for time series regression estimates which include generalized least squares, in the presence of long-range dependence in both errors and stochastic regressors. The setting and results differ significantly from earlier work on regression withlong-range-dependent errors. Spectral singularities are permitted at any frequency. When sufficiently strong spectral singularities in the error and a regressor coincide at the same frequency, least squares need no longer be $n^{1/2}$-consistent, where n is the sample size. However, we show that our class of estimates is $n^{1/2}$-consistent and asymptotically normal. In the generalized least squares case, we show that efficient estimation is still possible when the error autocorrelation is known only up to finitely many parameters. We include a Monte Carlo study of finite-sample performance and provide an extension to nonlinear least squares.

This paper provides limit theorems for multivariate, possibly non-Gaussian stationary processes whose spectral density matrices may have singularities not restricted at the origin, applying those limiting results to the asymptotic theory of parameter estimation and testing for statistical models of long-range dependent processes. The central limit theorems are proved based on the assumption that the innovations of the stationary processes satisfy certain mixing conditions for their conditional moments, and the usual assumptions of exact martingale difference or the (transformed) Gaussianity for the innovation process are dispensed with. For the proofs of convergence of the covariances of quadratic forms, the concept of the multiple Fejér kernel is introduced. For the derivation of the asymptotic properties of the quasi-likelihood estimate and the quasi-likelihood ratio, the bracketing function approach is used instead of conventional regularity conditions on the model spectral density.

Hardy-Weinberg equilibrium and linkage equilibrium are fundamental concepts in population genetics. In practice, testing linkage equilibrium in haplotype data is equivalent to testing independence in a large, sparse, multidimensional contingency table. Testing Hardy-Weinberg and linkage equilibrium simultaneously on multilocus genotype data introduces the additional complications of missing information and symmetry constraints on marginal probabilities. To avoid unreliable large-sample approximations for sparse contingency tables, one can use exact tests like Fisher's classical test that condition on observed marginal totals. Unfortunately, computing p-values for exact tests is often infeasible because of the large number of tables consistent with the marginal totals of an observed table. We develop here Markov chains for sampling from the appropriate conditional distributions for testing genetic equilibrium. These chains compare favorably with a parallel, independent-sampling method that we present. For n haplotype observations on J loci, the Markov chains converge to their stationary distributions in $[(J - 1)n 1n n]/2 + O(n)$ steps and can be an efficient tool for estimating p-values. Our theoretical treatment of these results involves strong stationary stopping times, order statistics, large deviations and the embedding of Poisson processes. We include some general results on the application of strong stationary times to bounding the precision and bias of sample average estimators.

A clone map of part or all of a chromosome is the result of organizing order and overlap information concerning collections of DNA fragments called clone libraries. In this paper the expected amount of information (entropy) needed to create such a map is discussed. A number of different formalizations of the notion of a clone map are considered, and exact or approximate expressions or bounds for the associated entropy are calculated for each formalization. Based on these bounds, comparisons are made for four species of the entropies associated with the mapping of their respective cosmid clone libraries. All the entropies have the same first-order term $N \log_2 N$ (when the clone library size $N \to \infty$) as that obtained by Lehrach et al.

While the additive model is a popular nonparametric regression method, many of its theoretical properties are not well understood, especially when the backfitting algorithm is used for computation of the estimators. This article explores those properties when the additive model is fitted by local polynomial regression. Sufficient conditions guaranteeing the asymptotic existence of unique estimators for the bivariate additive model are given. Asymptotic approximations to the bias and the variance of a homoscedastic bivariate additive model with local polynomial terms of odd and even degree are computed. This model is shown to have the same rate of convergence as that of univariate local polynomial regression.

Discrete choice models are frequently used in statistical and econometric practice. Standard models such as logit models are based on exact knowledge of the form of the link and linear index function. Semiparametric models avoid possible misspecification but often introduce a computational burden especially when optimization over nonparametric and parametric components are to be done iteratively. It is therefore interesting to decide between approaches. Here we propose a test of semiparametric versus parametric single index modelling. Our procedure allows the (linear) index of the semiparametric alternative to be different from that of the parametric hypothesis. The test is proved to be rate-optimal in the sense that it provides (rate) minimal distance between hypothesis and alternative for a given power function.

In a partial linear model, the dependence of a response variate Y on covariates (W, X$ is given by $$Y = W \beta + \eta(X) + \mathscr{E}$$ where $\mathscr{E}$ is independent of $(W, X)$ with densities g and f, respectively. In this paper an asymptotically efficient estimator of $\beta$ is constructed solely under mild smoothness assumptions on the unknown $\eta$, f and g, thereby removing the assumption of finite residual variance on which all least-squares-type estimators available in the literature are based.

When a spatial point process is observed through a bounded window, edge effects hamper the estimation of characteristics such as the empty space function F, the nearest neighbor distance distribution G and the reduced second-order moment function K. Here we propose and study product-limit type estimators of F, G and K based on the analogy with censored survival data: the distance from a fixed point to the nearest point of the process is right-censored by its distance to the boundary of the window. The resulting estimators have a ratio-unbiasedness property that is standard in spatial statistics. We show that the empty space function F of any stationary point process is absolutely continuous, and so is the product-limit estimator of F. The estimators are strongly consistent when there are independent replications or when the sampling window becomes large. We sketch a CLT for independent replications within a fixed observation window and asymptotic theory for independent replications of sparse Poisson processes. In simulations the new estimators are generally more efficient than the "border method" estimator but (for estimators of K), somewhat less efficient than sophisticated edge corrections.

The joint distribution of a d-dimensional random field restricted to a box of size k can be estimated by looking at a realization in a box of size $n \gg k$ and computing the empirical distribution. This is done by sliding a box of size k around in the box of size n and computing frequencies. We show that when $k = k(n)$ grows as a function of n, then the total variation distance between this empirical distribution and the true distribution goes to 0 a.s. as $n \to \infty$ provided $k(n)^d \leq (\log n^d)/(H + \varepsilon)$ (where H is the entropy of the random field) and providing the random field satisfies a condition called quite weak Bernoulli with exponential rate. This class of processes, studied previously, includes the plus state for the Ising model at a variety of parameter values and certain measures of maximal entropy for certain subshifts of finite type. Marton and Shields have proved such results in one dimension and this paper is an attempt to extend their results to some extent to higher dimensions.

For most of this century, randomization has been a cornerstone of scientific experimentation, especially when dealing with humans as experimental units. In practice, however, noncompliance is relatively common with human subjects, complicating traditional theories of inference that require adherence to the random treatment assignment. In this paper we present Bayesian inferential methods for causal estimands in the presence of noncompliance, when the binary treatment assignment is random and hence ignorable, but the binary treatment received is not ignorable. We assume that both the treatment assigned and the treatment received are observed. We describe posterior estimation using EM and data augmentation algorithms. Also, we investigate the role of two assumptions often made in econometric instrumental variables analyses, the exclusion restriction and the monotonicity assumption, without which the likelihood functions generally have substantial regions of maxima. We apply our procedures to real and artificial data, thereby demonstrating the technology and showing that our new methods can yield valid inferences that differ in practically important ways from those based on previous methods for analysis in the presence of noncompliance, including intention-to-treat analyses and analyses based on econometric instrumental variables techniques. Finally, we perform a simulation to investigate the operating characteristics of the competing procedures in a simple setting, which indicates relatively dramatic improvements in frequency operating characteristics attainable using our Bayesian procedures.

In this paper, we study linear inverse problems where some generalized moments of an unknown positive measure are observed. We introduce a new construction, called the maximum entropy on the mean method (MEM), which relies on a suitable sequence of finite-dimensional discretized inverse problems. Its advantage is threefold: It allows us to interpret all usual deterministic methods as Bayesian methods; it gives a very convenient way of taking into account prior information; it also leads to new criteria for the existence question concerning the linear inverse problem which will be a starting point for the investigation of superresolution phenomena. The key tool in this work is the large deviations property of some discrete random measure connected with the reconstruction procedure.

We consider a general random effects model for repeated binary measures, assuming a latent linear model with any class of mixing distributions. The latent model is assumed to have the Laird-Ware structure, but the random effects may be from any specified class of multivariate distributions and the error vector may have any specified continuous distribution. Elementwise threshold crossing then gives the observed vector of binary outcomes. Special cases of this model include recently discussed mixed logistic regression and probit models, which have had either parametric (usually Gaussian) or nonparametric mixing distributions. We give sufficient conditions for identifiability of the mixing distribution and fixed effects and for convergence of maximum likelihood estimators for the mixing distribution and fixed effects. As expected, the conditions are much stronger for nonparametric mixing than for Gaussian mixing. We illustrate the conditions by applying them to a practical example.

Consider finite mixture models of the form $g(x; Q) = \int f(x; \theta) dQ(\theta)$, where f is a parametric density and Q is a discrete probability measure. An important and difficult statistical problem concerns the determination of the number of support points (usually known as components) of Q from a sample of observations from g. For an important class of exponential family models we have the following result: if P has more than p components and Q is an appropriately chosen p-component approximation of P, then $g(x; P) - g(x; Q)$ demonstrates a prescribed sign change behavior, as does the corresponding difference in the distribution functions. These strong structural properties have implications for diagnostic plots for the number of components in a finite mixture.

Least squares penalized regression estimates with total variation penalties are considered. It is shown that these estimators are least squares splines with locally data adaptive placed knot points. The definition of these variable knot splines as minimizers of global functionals can be used to study their asymptotic properties. In particular, these results imply that the estimates adapt well to spatially inhomogeneous smoothness. We show rates of convergence in bounded variation function classes and discuss pointwise limiting distributions. An iterative algorithm based on stepwise addition and deletion of knot points is proposed and its consistency proved.

Quantile smoothing splines were introduced by Koenker, Ng and Portnoy as natural and appealing estimates of conditional quantiles of response variables. The natural setting for the problem considers minimization of a weighted combination of a "fit" penalty and a "roughness" penalty over the space of functions whose derivatives have bounded variation. Although this space is not traditional, Shen has shown recently that the quantile smoothing splines do indeed converge at the usual optimal rate $n^{-2/5)$ in various norms. Here, local asymptotic results are obtained by establishing Bahadur representations for local parameters of the splines. These are used to obtain local rates of convergence, to establish uniform convergence rates, to provide local distribution theory for quantile B-splines and to expand the "fit" measure in order to analyze an information criterion for determining the smoothing parameter. Examples of using derivatives of the smoothing splines for estimating jump functions are also presented.