Nian-Sheng Tang

Model Selection Criteria for Missing-Data Problems Using the EM Algorithm

We consider novel methods for the computation of model selection criteria in missing-data problems based on the output of the EM algorithm. The methodology is very general and can be applied to numerous situations involving incomplete data within an EM framework, from covariates missing at random in arbitrary regression models to nonignorably missing longitudinal responses and/or covariates. Toward this goal, we develop a class of information criteria for missing-data problems, called ICH, Q, which yields the Akaike information criterion and the Bayesian information criterion as special cases. The computation of ICH, Q requires an analytic approximation to a complicated function, called the H-function, along with output from the EM algorithm used in obtaining maximum likelihood estimates. The approximation to the H-function leads to a large class of information criteria, called ICH̃(k), Q. Theoretical properties of ICH̃(k), Q, including consistency, are investigated in detail. To eliminate the analytic approximation to the H-function, a computationally simpler approximation to ICH, Q, called ICQ, is proposed, the computation of which depends solely on the Q-function of the EM algorithm. Advantages and disadvantages of ICH̃(k), Q and ICQ are discussed and examined in detail in the context of missing-data problems. Extensive simulations are given to demonstrate the methodology and examine the small-sample and large-sample performance of ICH̃(k), Q and ICQ in missing-data problems. An AIDS data set also is presented to illustrate the proposed methodology.

Bayesian Methods for Analyzing Structural Equation Models With Covariates, Interaction, and Quadratic Latent Variables

The analysis of interaction among latent variables has received much attention. This article introduces a Bayesian approach to analyze a general structural equation model that accommodates the general nonlinear terms of latent variables and covariates. This approach produces a Bayesian estimate that has the same statistical optimal properties as a maximum likelihood estimate. Other advantages over the traditional approaches are discussed. More important, we demonstrate through examples how to use the freely available software WinBUGS to obtain Bayesian results for estimation and model comparison. Simulation studies are conducted to assess the empirical performances of the approach for situations with various sample sizes and prior inputs.

A Statistical Analysis of Brain Morphology Using Wild Bootstrapping

Methods for the analysis of brain morphology, including voxel-based morphology and surface-based morphometries, have been used to detect associations between brain structure and covariates of interest, such as diagnosis, severity of disease, age, IQ, and genotype. The statistical analysis of morphometric measures usually involves two statistical procedures: 1) invoking a statistical model at each voxel (or point) on the surface of the brain or brain subregion, followed by mapping test statistics (e.g., t test) or their associated p values at each of those voxels; 2) correction for the multiple statistical tests conducted across all voxels on the surface of the brain region under investigation. We propose the use of new statistical methods for each of these procedures. We first use a heteroscedastic linear model to test the associations between the morphological measures at each voxel on the surface of the specified subregion (e.g., cortical or subcortical surfaces) and the covariates of interest. Moreover, we develop a robust test procedure that is based on a resampling method, called wild bootstrapping. This procedure assesses the statistical significance of the associations between a measure of given brain structure and the covariates of interest. The value of this robust test procedure lies in its computationally simplicity and in its applicability to a wide range of imaging data, including data from both anatomical and functional magnetic resonance imaging (fMRI). Simulation studies demonstrate that this robust test procedure can accurately control the family-wise error rate. We demonstrate the application of this robust test procedure to the detection of statistically significant differences in the morphology of the hippocampus over time across gender groups in a large sample of healthy subjects.

FRATS: Functional Regression Analysis of DTI Tract Statistics

Diffusion tensor imaging (DTI) provides important information on the structure of white matter fiber bundles as well as detailed tissue properties along these fiber bundles in vivo. This paper presents a functional regression framework, called FRATS, for the analysis of multiple diffusion properties along fiber bundle as functions in an infinite dimensional space and their association with a set of covariates of interest, such as age, diagnostic status and gender, in real applications. The functional regression framework consists of four integrated components: the local polynomial kernel method for smoothing multiple diffusion properties along individual fiber bundles, a functional linear model for characterizing the association between fiber bundle diffusion properties and a set of covariates, a global test statistic for testing hypotheses of interest, and a resampling method for approximating the p-value of the global test statistic. The proposed methodology is applied to characterizing the development of five diffusion properties including fractional anisotropy, mean diffusivity, and the three eigenvalues of diffusion tensor along the splenium of the corpus callosum tract and the right internal capsule tract in a clinical study of neurodevelopment. Significant age and gestational age effects on the five diffusion properties were found in both tracts. The resulting analysis pipeline can be used for understanding normal brain development, the neural bases of neuropsychiatric disorders, and the joint effects of environmental and genetic factors on white matter fiber bundles.

Local Influence Analysis of Nonlinear Structural Equation Models.

By regarding the latent random vectors as hypothetical missing data and based on the conditional expectation of the complete-data log-likelihood function in the EM algorithm, we investigate assessment of local influence of various perturbation schemes in a nonlinear structural equation model. The basic building blocks of local influence analysis are computed via observations of the latent variables generated by the Metropolis-Hastings algorithm, while the diagnostic measures are obtained via the conformal normal curvature. Seven perturbation schemes, including some perturbation schemes on latent vectors, are investigated. The proposed procedure is illustrated by a simulation study and a real example.

Bayesian estimation of semiparametric nonlinear dynamic factor analysis models using the Dirichlet process prior

Parameters in time series and other dynamic models often show complex range restrictions and their distributions may deviate substantially from multivariate normal or other standard parametric distributions. We use the truncated Dirichlet process (DP) as a non-parametric prior for such dynamic parameters in a novel nonlinear Bayesian dynamic factor analysis model. This is equivalent to specifying the prior distribution to be a mixture distribution composed of an unknown number of discrete point masses (or clusters). The stick-breaking prior and the blocked Gibbs sampler are used to enable efficient simulation of posterior samples. Using a series of empirical and simulation examples, we illustrate the flexibility of the proposed approach in approximating distributions of very diverse shapes.

Bayesian analysis of structural equation models with mixed exponential family and ordered categorical data.

Structural equation models are very popular for studying relationships among observed and latent variables. However, the existing theory and computer packages are developed mainly under the assumption of normality, and hence cannot be satisfactorily applied to non-normal and ordered categorical data that are common in behavioural, social and psychological research. In this paper, we develop a Bayesian approach to the analysis of structural equation models in which the manifest variables are ordered categorical and/or from an exponential family. In this framework, models with a mixture of binomial, ordered categorical and normal variables can be analysed. Bayesian estimates of the unknown parameters are obtained by a computational procedure that combines the Gibbs sampler and the Metropolis-Hastings algorithm. Some goodness-of-fit statistics are proposed to evaluate the fit of the posited model. The methodology is illustrated by results obtained from a simulation study and analysis of a real data set about non-adherence of hypertension patients in a medical treatment scheme.

Testing the equality of proportions for correlated otolaryngologic data

In otolaryngologic (or ophthalmologic) studies, each subject usually contributes information for each of two ears (or eyes), and the values from the two ears (or eyes) are generally highly correlated. Statistical procedures that fail to take into account the correlation between responses from two ears could lead to incorrect results. On the other hand, asymptotic procedures that overlook small sample designs, sparse data structures, or the discrete nature of data could yield unacceptably high type I error rates even when the intraclass correlation is taken into consideration. In this article, we investigate eight procedures for testing the equality of proportions in such correlated data. These test procedures will be implemented via the asymptotic and approximate unconditional methods. Our empirical results show that tests based on the approximate unconditional method usually produce empirical type I error rates closer to the pre-chosen nominal level than their asymptotic tests. Amongst these, the approximate unconditional score test performs satisfactorily in general situations and is hence recommended. A data set from an otolaryngologic study is used to illustrate our proposed methods.

Tests of Noninferiority via Rate Difference for Three-Arm Clinical Trials with Placebo

Abstract In assessing a noninferiority trial, the investigator intends to show efficacy by demonstrating that a new experimental drug/treatment is not worse than a known active control/reference by a small predefined margin. If it is ethically justifiable, it may be advisable to include an additional placebo group for internal validation purpose. This constitutes the well-known three-arm clinical trial with placebo. In this paper, we study two asymptotic statistical methods for testing of noninferiority in three-arm clinical trials with placebo for binary outcomes based on rate difference. They are sample-based estimation method and restricted maximum likelihood estimation method, respectively. We investigate the performance of the proposed test procedures under different sample size allocation settings via a simulation study. Both methods perform satisfactorily under moderate to large sample settings. However, the restricted maximum likelihood estimation method usually possesses slightly smaller actual type I error rates, which are relatively close to the prespecified nominal level, while the sample-based method can be expressed in a simple closed-form format. Real examples from a pharmacological study of patients with functional dyspepsia and a placebo- controlled trail of subjects with acute migraine are used to demonstrate our methodologies.

Empirical likelihood inference for mean functionals with nonignorably missing response data

An empirical likelihood (EL) approach to inference on mean functionals with nonignorably missing response data is developed. The nonignorably missing mechanism is specified by an exponential tilting model. Several maximum EL estimators (MELEs) for the response mean functional are proposed under different scenarios. We systematically investigate asymptotic properties of the proposed MELEs for the response mean functional. With the use of auxiliary information, MELEs are statistically more efficient. Confidence intervals (CIs) for the response mean are constructed on the basis of the EL method and the normal approximation (NA) method. Simulation studies are presented to evaluate the finite sample performance of our proposed MELEs and CIs. A real earnings data from the New York Social Indicators Survey is used to illustrate our proposed EL method. Empirical results show that our proposed EL method is robust.