On nonparametric classification with missing covariates

General procedures are proposed for nonparametric classification in the presence of missing covariates. Both kernel-based imputation as well as Horvitz-Thompson-type inverse weighting approaches are employed to handle the presence of missing covariates. In the case of imputation, it is a certain regression function which is being imputed (and not the missing values). Using the theory of empirical processes, the performance of the resulting classifiers is assessed by obtaining exponential bounds on the deviations of their conditional errors from that of the Bayes classifier. These bounds, in conjunction with the Borel-Cantelli lemma, immediately provide various strong consistency results.     

Additive rate model with auxiliary covariates

In this paper, we consider an inference method for recurrent event data in which the primary exposure covariate is assessed only in a validation set, while as an auxiliary covariate for the main exposure is available for the full cohort. Additive rate model is considered. The existing estimating equations in the absence of primary exposure are corrected by taking use of the validation data and auxiliary information, which yield consistent and asymptotically normal estimators of the regression parameters. The estimated baseline mean process is shown to converge weakly to a zero-mean Gaussian process. Extensive simulations are conducted to evaluate finite sample performance.     

Generalized partially linear models with missing covariates

In this article we study a semiparametric generalized partially linear model when the covariates are missing at random. We propose combining local linear regression with the local quasilikelihood technique and weighted estimating equation to estimate the parameters and nonparameters when the missing probability is known or unknown. We establish normality of the estimators of the parameter and asymptotic expansion for the estimators of the nonparametric part. We apply the proposed models and methods to a study of the relation between virologic and immunologic responses in AIDS clinical trials, in which virologic response is classified into binary variables. We also give simulation results to illustrate our approach.     

On categorical time series models with covariates

We study the problem of stationarity and ergodicity for autoregressive multinomial logistic time series models which possibly include a latent process and are defined by a GARCH-type recursive equation. We improve considerably upon the existing conditions about stationarity and ergodicity of those models. Proofs are based on theory developed for chains with complete connections. A useful coupling technique is employed for studying ergodicity of infinite order finite-state stochastic processes which generalize finite-state Markov chains. Furthermore, for the case of finite order Markov chains, we discuss ergodicity properties of a model which includes strongly exogenous but not necessarily bounded covariates.     

Nonparametric ANCOVA with two and three covariates

Fully nonparametric analysis of covariance with two and three covariates is considered. The approach is based on an extension of the model of Akritas et al. (Biometrika 87(3) (2000) 507). The model allows for possibly nonlinear covariate effect which can have different shape in different factor level combinations. All types of ordinal data are included in the formulation. In particular, the response distributions are not restricted to comply to any parametric or semiparametric model. In this nonparametric model, hypotheses of no main effect no interaction and no simple effect, which adjust for the covariate values, are defined through a decomposition of the conditional distribution functions of the response given to the factor level combination and covariate values. The test statistics are based on averages over the covariate values of certain Nadaraya–Watson regression quantities. Under their respective null hypotheses, such test statistics are shown to have a central χ² distribution. Small sample corrections are also provided. Simulation results and the analysis of two real datasets are also presented.     

Nonparametric estimation of competing risks models with covariates

In competing risks model, several failure times arise potentially. The smallest failure time and its index only are observed. Without specific assumptions, the joint or even the marginal distribution functions of the underlying failure times are not identifiable (A. Tsiatis, Proc. Natl. Acad. Sci. USA 72 (1975) 20). Nonetheless, if each individual is characterized by a “sufficiently informative” set of covariates, these distributions are identifiable under some conditions of regularity (J.J. Heckman and B. Honoré, Biometrika 76 (1989) 325). In this paper, nonparametric kernel estimators of the joint distribution function of failure times conditional on the covariates are proposed. Their weak and strong consistency are discussed.     

A note on Left-Spherically Distributed test with covariates

The Left-Spherically Distributed linear scores test of Läuter et al. (1998) (Läuter, J., Glimm, E., Kropf, S., 1998. Multivariate tests based on left-spherically distributed linear scores. Annals of Statistics 26, 1972-1988) is extended to account for nuisance parameters, particularly for covariates that are assumed to explain (part of) the response variables but are not under test. An R code is available on the someMTP package in CRAN.     

Sparse random graphs with clustering

In 2007, we introduced a general model of sparse random graphs with (conditional) independence between the edges. The aim of this article is to present an extension of this model in which the edges are far from independent, and to prove several results about this extension. The basic idea is to construct the random graph by adding not only edges but also other small graphs. In other words, we first construct an inhomogeneous random hypergraph with (conditionally) independent hyperedges, and then replace each hyperedge by a (perhaps complete) graph. Although flexible enough to produce graphs with significant dependence between edges, this model is nonetheless mathematically tractable. Indeed, we find the critical point where a giant component emerges in full generality, in terms of the norm of a certain integral operator, and relate the size of the giant component to the survival probability of a certain (non‐Poisson) multi‐type branching process. While our main focus is the phase transition, we also study the degree distribution and the numbers of small subgraphs. We illustrate the model with a simple special case that produces graphs with power‐law degree sequences with a wide range of degree exponents and clustering coefficients. © 2010 Wiley Periodicals, Inc. Random Struct. Alg., 38, 269–323, 2011     

Covariate-adaptive designs with missing covariates in clinical trials

Many covariate-adaptive randomization procedures have been proposed and implemented to balance important covariates in clinical trials. These methods are usually based on fully observed covariates. In practice, the covariates of a patient are often partially missing. We propose a novel covariate-adaptive design to deal with missing covariates and study its properties. For the proposed design, we show that as the number of patients increases, the overall imbalance, observed margin imbalance and fully observed stratum imbalance are bounded in probability. Under certain covariate-dependent missing mechanism, the proposed design can balance missing covariates as if the covariates are observed. Finally, we explore our methods and theoretical findings through simulations.     

Testing separability in marked multidimensional point processes with covariates

In modeling marked point processes, it is convenient to assume a separable or multiplicative form for the conditional intensity, as this assumption typically allows one to estimate each component of the model individually. Tests have been proposed in the simple marked point process case, to investigate whether the mark distribution is separable from the spatial–temporal characteristics of the point process. Here, we extend these tests to the case of a marked point process with covariates, and where one is interested in testing the separability of each of the covariates, as well as the mark and the coordinates of the point process. The extension is not at all trivial, and covariates must be treated in a fundamentally different way than marks and coordinates of the process, especially when the covariates are not uniformly distributed. An application is given to point process models for forecasting wildfire hazard in Los Angeles County, California, and solutions are proposed to the problem of how to proceed when the separability hypothesis is rejected.     

Robust estimation for partially linear models with large-dimensional covariates

We are concerned with robust estimation procedures to estimate the parameters in partially linear models with large-dimensional covariates. To enhance the interpretability, we suggest implementing a nonconcave regularization method in the robust estimation procedure to select important covariates from the linear component. We establish the consistency for both the linear and the nonlinear components when the covariate dimension diverges at the rate of o(n1/2), where n is the sample size. We show that the robust estimate of linear component performs asymptotically as well as its oracle counterpart which assumes the baseline function and the unimportant covariates were known a priori. With a consistent estimator of the linear component, we estimate the nonparametric component by a robust local linear regression. It is proved that the robust estimate of nonlinear component performs asymptotically as well as if the linear component were known in advance.Comprehensive simulation studies are carried out and an application is presented to examine the fnite-sample performance of the proposed procedures.     

Mixtures of factor analyzers: an extension with covariates

This paper examines the analysis of an extended finite mixture of factor analyzers (MFA) where both the continuous latent variable (common factor) and the categorical latent variable (component label) are assumed to be influenced by the effects of fixed observed covariates. A polytomous logistic regression model is used to link the categorical latent variable to its corresponding covariate, while a traditional linear model with normal noise is used to model the effect of the covariate on the continuous latent variable. The proposed model turns out be in various ways an extension of many existing related models, and as such offers the potential to address some of the issues not fully handled by those previous models. A detailed derivation of an EM algorithm is proposed for parameter estimation, and latent variable estimates are obtained as by-products of the overall estimation procedure.     

Sparse potentials with fractional Hausdorff dimension

We construct non-random bounded discrete half-line Schrödinger operators which have purely singular continuous spectral measures with fractional Hausdorff dimension (in some interval of energies). To do this we use suitable sparse potentials. Our results also apply to whole line operators, as well as to certain random operators. In the latter case we prove and compute an exact dimension of the spectral measures.     

Sparse geometric graphs with small dilation

Given a set S of n points in , and an integer k such that 0k<n, we show that a geometric graph with vertex set S, at most n−1+k edges, maximum degree five, and dilation O(n/(k+1)) can be computed in time O(nlogn). For any k, we also construct planar n-point sets for which any geometric graph with n−1+k edges has dilation Ω(n/(k+1)); a slightly weaker statement holds if the points of S are required to be in convex position.     

Weighted estimating equations for additive hazards models with missing covariates

Annals of the Institute of Statistical Mathematics - This paper presents simple weighted and fully augmented weighted estimators for the additive hazards model with missing covariates when they are...     

Semiparametric efficient inferences for lifetime regression model with time-dependent covariates

Through a threshold equation, we propose a time-transformed accelerated failure time (AFT) model with time-dependent covariate history in survival analysis. This model contains a general class of semiparametric lifetime regression models, including AFT with identical time-scale and a wide spectrum of Cox’s hazard regression models and their frailty variants. We first construct the semiparametric efficient statistical inferences on the AFT model with identical time-scale. The theoretical semiparametric Fisher information bound is explicitly derived under right-censored data setting. And the overidentified estimating equation (OEE) approach based on two martingale processes is shown to achieve this semiparametric efficiency bound. Extensions of the semiparametric efficient statistical inferences to the time-transformed AFT versions are also discussed. We also conclude that most log-rank estimating equations would suffer severe information loss primarily caused by wiggling pattern of the baseline hazard function, while the OEE approach can alleviate the damaging effects. A simulated biological life history example is numerically studied.     

Principled sure independence screening for Cox models with ultra-high-dimensional covariates

It is rather challenging for current variable selectors to handle situations where the number of covariates under consideration is ultra-high. Consider a motivating clinical trial of the drug bortezomib for the treatment of multiple myeloma, where overall survival and expression levels of 44760 probesets were measured for each of 80 patients with the goal of identifying genes that predict survival after treatment. This dataset defies analysis even with regularized regression. Some remedies have been proposed for the linear model and for generalized linear models, but there are few solutions in the survival setting and, to our knowledge, no theoretical support. Furthermore, existing strategies often involve tuning parameters that are difficult to interpret. In this paper we propose and theoretically justify a principled method for reducing dimensionality in the analysis of censored data by selecting only the important covariates. Our procedure involves a tuning parameter that has a simple interpretation as the desired false positive rate of this selection. We present simulation results and apply the proposed procedure to analyze the aforementioned myeloma study.