Full weighted least squares (full WLS) and robust weighted least squares (robust WLS) are currently the two primary estimation methods designed for structural equation modeling with ordinal observed variables. These methods assume that continuous latent variables were coarsely categorized by the measurement process to yield the observed ordinal variables, and that the model proposed by the researcher pertains to these latent variables rather than to their ordinal manifestations. Previous research has strongly suggested that robust WLS is superior to full WLS when models are correctly specified. Given the realities of applied research, it was critical to examine these methods with misspecified models. This Monte Carlo simulation study examined the performance of full and robust WLS for two-factor, eight-indicator confirmatory factor analytic models that were either correctly specified, overspecified, or misspecified in one of two ways. Seven conditions of five-category indicator distribution shape at four sample sizes were simulated. These design factors were completely crossed for a total of 224 cells. Previously findings of the relative superiority of robust WLS with correctly specified models were replicated, and robust WLS was also found to perform better than full WLS given overspecification or misspecification. Robust WLS parameter estimates were usually more accurate for correct and overspecified models, especially at the smaller sample sizes. In the face of misspecification, full WLS better approximated the correct loading values whereas robust estimates better approximated the correct factor correlation. Robust WLS chi-square values discriminated between correct and misspecified models much better than full WLS values at the two smaller sample sizes. For all four model specifications, robust parameter estimates usually showed lower variability and robust standard errors usually showed lower bias. These findings suggest that robust WLS should likely remain the estimator of choice for applied researchers. Additionally, highly leptokurtic distributions should be avoided when possible. It should also be noted that robust WLS performance was arguably adequate at the sample size of 100 when the indicators were not highly leptokurtic. / text

ABSTRACT ROBUST CONTROL CHARTS &Ccedil / etiny&uuml / rek, Aysun M. Sc., Department of Statistics Supervisor: Dr. BariS S&uuml / r&uuml / c&uuml / Co-Supervisor: Assoc. Prof. Dr. Birdal Senoglu December 2006, 82 pages Control charts are one of the most commonly used tools in statistical process control. A prominent feature of the statistical process control is the Shewhart control chart that depends on the assumption of normality. However, violations of underlying normality assumption are common in practice. For this reason, control charts for symmetric distributions for both long- and short-tailed distributions are constructed by using least squares estimators and the robust estimators -modified maximum likelihood, trim, MAD and wave. In order to evaluate the performance of the charts under the assumed distribution and investigate robustness properties, the probability of plotting outside the control limits is calculated via Monte Carlo simulation technique.

Le contrôle explicite de la convergence des sommes convenablement normalisées de variables aléatoires, ainsi que l'étude du principe de déviations modérées associé à ces sommes constituent les thèmes centraux de cette thèse. Nous étudions principalement deux types de processus. Premièrement, nous nous intéressons aux processus indexés par un arbre binaire, aléatoire ou non. Ces processus ont été introduits dans la littérature afin d'étudier le mécanisme de la division cellulaire. Au chapitre 2, nous étudions les chaînes de Markov bifurcantes. Ces chaînes peuvent être vues comme une adaptation des chaînes de Markov "usuelles'' dans le cas où l'ensemble des indices à une structure binaire. Sous des hypothèses d'ergodicité géométrique uniforme et non-uniforme d'une chaîne de Markov induite, nous fournissons des inégalités de déviations et un principe de déviations modérées pour les chaînes de Markov bifurcantes. Au chapitre 3, nous nous intéressons aux processus bifurcants autorégressifs d'ordre p (). Ces processus sont une adaptation des processus autorégressifs linéaires d'ordre p dans le cas où l'ensemble des indices à une structure binaire. Nous donnons des inégalités de déviations, ainsi qu'un principe de déviations modérées pour les estimateurs des moindres carrés des paramètres "d'autorégression'' de ce modèle. Au chapitre 4, nous traitons des inégalités de déviations pour des chaînes de Markov bifurcantes sur un arbre de Galton-Watson. Ces chaînes sont une généralisation de la notion de chaînes de Markov bifurcantes au cas où l'ensemble des indices est un arbre de Galton-Watson binaire. Elles permettent dans le cas de la division cellulaire de prendre en compte la mort des cellules. Les hypothèses principales que nous faisons dans ce chapitre sont : l'ergodicité géométrique uniforme d'une chaîne de Markov induite et la non-extinction du processus de Galton-Watson associé. Au chapitre 5, nous nous intéressons aux modèles autorégressifs linéaires d'ordre 1 ayant des résidus corrélés. Plus particulièrement, nous nous concentrons sur la statistique de Durbin-Watson. La statistique de Durbin-Watson est à la base des tests de Durbin-Watson, qui permettent de détecter l'autocorrélation résiduelle dans des modèles autorégressifs d'ordre 1. Nous fournissons un principe de déviations modérées pour cette statistique. Les preuves du principe de déviations modérées des chapitres 2, 3 et 4 reposent essentiellement sur le principe de déviations modérées des martingales. Les inégalités de déviations sont établies principalement grâce à l'inégalité d'Azuma-Bennet-Hoeffding et l'utilisation de la structure binaire des processus. Le chapitre 5 est né de l'importance qu'a l'ergodicité explicite des chaînes de Markov au chapitre 3. L'ergodicité géométrique explicite des processus de Markov à temps discret et continu ayant été très bien étudiée dans la littérature, nous nous sommes penchés sur l'ergodicité sous-exponentielle des processus de Markov à temps continu. Nous fournissons alors des taux explicites pour la convergence sous exponentielle d'un processus de Markov à temps continu vers sa mesure de probabilité d'équilibre. Les hypothèses principales que nous utilisons sont : l'existence d'une fonction de Lyapunov et d'une condition de minoration. Les preuves reposent en grande partie sur la construction du couplage et le contrôle explicite de la queue du temps de couplage. / The explicit control of the convergence of properly normalized sums of random variables, as well as the study of moderate deviation principle associated with these sums constitute the main subjects of this thesis. We mostly study two sort of processes. First, we are interested in processes labelled by binary tree, random or not. These processes have been introduced in the literature in order to study mechanism of the cell division. In Chapter 2, we study bifurcating Markov chains. These chains may be seen as an adaptation of "usual'' Markov chains in case the index set has a binary structure. Under uniform and non-uniform geometric ergodicity assumptions of an embedded Markov chain, we provide deviation inequalities and a moderate deviation principle for the bifurcating Markov chains. In chapter 3, we are interested in p-order bifurcating autoregressive processes (). These processes are an adaptation of $p$-order linear autoregressive processes in case the index set has a binary structure. We provide deviation inequalities, as well as an moderate deviation principle for the least squares estimators of autoregressive parameters of this model. In Chapter 4, we dealt with deviation deviation inequalities for bifurcating Markov chains on Galton-Watson tree. These chains are a generalization of the notion of bifurcating Markov chains in case the index set is a binary Galton-Watson tree. They allow, in case of cell division, to take into account cell's death. The main hypothesis that we do in this chapter are : uniform geometric ergodicity of an embedded Markov chain and the non-extinction of the associated Galton-Watson process. In Chapter 5, we are interested in first-order linear autoregressive models with correlated errors. More specifically, we focus on the Durbin-Watson statistic. The Durbin-Watson statistic is at the base of Durbin-Watson tests, which allow to detect serial correlation in the first-order autoregressive models. We provide a moderate deviation principle for this statistic. The proofs of moderate deviation principle of Chapter 2, 3 and 4 are essentially based on moderate deviation for martingales. To establish deviation inequalities, we use most the Azuma-Bennet-Hoeffding inequality and the binary structure of processes. Chapter 6 was born from the importance that explicit ergodicity of Markov chains has in Chapter 2. Since explicit geometric ergodicity of discrete and continuous time Markov processes has been well studied in the literature, we focused on the sub-exponential ergodicity of continuous time Markov Processes. We thus provide explicit rates for the sub-exponential convergence of a continuous time Markov process to its stationary distribution. The main hypothesis that we use are : existence of a Lyapunov fonction and of a minorization condition. The proofs are largely based on the coupling construction and the explicit control of the tail of the coupling time.