Thesis (M.Soc.Sc.)--University of Hong Kong, 1993. / Includes bibliographical references (leaf 33) Also available in print.

Thesis (Ph. D.)--Washington State University, May 2010. / Title from PDF title page (viewed on June 30, 2010). "Department of Mathematics." Includes bibliographical references (p. 61-65).

Previous research suggests that a specific biomarker measured in the blood correlates with cancer status, for a specific type of cancer: higher values of the biomarker are generally found in patients with progressive cancer. The aim of this study is to investigate this relationship using a Linear mixed model. Patients with a progressive disease have on average significantly higher values of the log of the biomarker and patients with partial or complete remission of the disease have on average significantly lower values of the logged biomarker, both compared to patients with a stable disease. Also, patients with a liver tumor have on average higher values of the log of the biomarker, compared to patients without. Including both a random intercept and a random slope in the Linear mixed model, in addition to the fixed effects, results in the best model fit. Due to the non-random sample used, these results are only valid for this specific sample but can be of guidance for the conduction and planning of future studies.

No description available.

In causal inference, a central aim of covariate selection is to provide a subset of covariates, that is sufficient for confounding adjustment. One approach for this is to construct a subset of covariates associated with the outcome. This is sometimes referred to as the outcome approach, which is the subject for this thesis. Apart from confounding, there may exist effect modification. This occurs when a treatment has different effect on the outcome, among different subgroups, defined by effect modifiers. We describe how the outcome approach implemented by regression models, can be used for estimating the ATE, and how sufficient subsets of covariates may be constructed for these models. We also describe a novel method, called the difference LASSO, which results in identification of effect modifiers, rather than determination of sufficient subsets. The method is defined by an algorithm where, in the first step, an incorrectly specified model is fitted. We investigate the bias, arising from this misspecification, analytically and numerically for OLS. The difference LASSO is also compared with a regression estimator. The comparison is done in a simulation study, where the identification of effect modifiers is evaluated. This is done by analyzing the proportion of times a selection procedure results in a set of covariates including only the effect modifiers, or a set where the effect modifiers are included as a subset. The results show that the difference LASSO works relatively well for identification of effect modifiers. Among four designs, a set containing only the true effect modifiers were selected in at least 83:2%. The corresponding result for the regression estimator was 27:9%. However, the difference LASSO builds on biased estimation. Therefore, the method is not plausible for interpretation of treatment effects.

This study compares the performance of the Maximum Likelihood estimator (MLE), estimators based on spacings called Generalized Maximum Spacing estimators (GSEs), and the One Step Minimum Hellinger Distance estimator (OSMHD), on data originating from a circular distribution.  The purpose of the study is to investigate the different estimators’ performance on directional data. More specifically, we compare the estimators’ ability to estimate parameters of the von Mises distribution, which is determined by a location parameter and a scale parameter. For this study, we only look at the scenario in which one of the parameters is unknown. The main part of the study is concerned with estimating the parameters under the condition, in which the data contain outliers, but a small part is also dedicated to estimation at the true model.  When estimating the location parameter under contaminated conditions, the results indicate that some versions of the GSEs tend to outperform the other estimators. It should be noted that these seemingly more robust estimators appear comparatively less optimal at the true model, but this is a tradeoff that must be made on a case by case basis. Under the same contaminated conditions, all included estimators appear to have seemingly greater difficulties estimating the scale parameter. However, for this case, some of the GSEs are able to handle the contamination a bit better than the rest. In addition, there might exist other versions of GSEs, not included in this study, which perform better.

Structured output prediction is a machine learning tasks in which an input object is not just assigned a single class, as in classification, but multiple, interdependent labels. This means that the presence or value of a given label affects the other labels, for instance in text labelling problems, where output labels are applied to each word, and their interdependencies must be modelled. Non-parametric Bayesian (NPB) techniques are probabilistic modelling techniques which have the interesting property of allowing model capacity to grow, in a controllable way, with data complexity, while maintaining the advantages of Bayesian modelling. In this thesis, we develop NPB algorithms to solve structured output problems. We first study a map-reduce implementation of a stochastic inference method designed for the infinite hidden Markov model, applied to a computational linguistics task, part-of-speech tagging. We show that mainstream map-reduce frameworks do not easily support highly iterative algorithms. The main contribution of this thesis consists in a conceptually novel discriminative model, GPstruct. It is motivated by labelling tasks, and combines attractive properties of conditional random fields (CRF), structured support vector machines, and Gaussian process (GP) classifiers. In probabilistic terms, GPstruct combines a CRF likelihood with a GP prior on factors; it can also be described as a Bayesian kernelized CRF. To train this model, we develop a Markov chain Monte Carlo algorithm based on elliptical slice sampling and investigate its properties. We then validate it on real data experiments, and explore two topologies: sequence output with text labelling tasks, and grid output with semantic segmentation of images. The latter case poses scalability issues, which are addressed using likelihood approximations and an ensemble method which allows distributed inference and prediction. The experimental validation demonstrates: (a) the model is flexible and its constituent parts are modular and easy to engineer; (b) predictive performance and, most crucially, the probabilistic calibration of predictions are better than or equal to that of competitor models, and (c) model hyperparameters can be learnt from data.

No description available.

No description available.

No description available.