As researchers increasingly use web-based surveys, the ease of dropping out in the online setting is a growing issue in ensuring data quality. One theory is that dropout or attrition occurs in phases that can be generalized to phases of high dropout and phases of stable use. In order to detect these phases, several methods are explored. First, existing methods and user-specified thresholds are applied to survey data where significant changes in the dropout rate between two questions is interpreted as the start or end of a high dropout phase. Next, survey dropout is considered as a time-to-event outcome and tests within change-point hazard models are introduced. Performance of these change-point hazard models is compared. Finally, all methods are applied to survey data on patient cancer screening preferences, testing the null hypothesis of no phases of attrition (no change-points) against the alternative hypothesis that distinct attrition phases exist (at least one change-point).

The focus of this thesis was to investigate which baseball metrics are most conducive to run creation and prevention. Stepwise regression and Liu estimation were used to formulate two models for the dependent variables and also used for cross validation. Finally, the predicted values were fed into the Pythagorean Expectation formula to predict a team’s most important goal: winning. Each model fit strongly and collinearity amongst offensive predictors was considered using variance inflation factors. Hits, walks, and home runs allowed, infield putouts, errors, defense-independent earned run average ratio, defensive efficiency ratio, saves, runners left on base, shutouts, and walks per nine innings were significant defensive predictors. Doubles, home runs, walks, batting average, and runners left on base were significant offensive regressors. Both models produced error rates below 3% for run prediction and together they did an excellent job of estimating a team’s per-season win ratio.

Background Over the past three decades, social relationships have been shown to have important effects upon health. However, many different definitions and aspects of social relationships have been considered in the various studies, making comparison of findings difficult. Furthermore, the effects of social relationships upon different health outcomes have rarely been investigated within the same cohort of older people. In addition, there is a paucity of information concerning the effects of social relationships upon health of older Australians. Aim This thesis aims to investigate the effects of the structural aspects of social relationships – that is, social networks – on health among older Australians. The three specific health outcomes considered in this thesis were disability, residential care use and death. The specific aims of the thesis were to: 1. Develop a measurement model of social networks. 2. Examine the effects of total and specific social networks upon disability. 3. Determine the effects of total and specific social networks upon use of residential care. 4. Investigate the effects of total and specific social networks upon survival. An additional aim was to determine if there were threshold effects of social networks on the three specific health outcomes. Methods The study drew on six waves of data from 1477 participants in the Australian Longitudinal Study of Ageing. A range of statistical techniques, including binary and multinomial logistic regression and survival analysis, were used in the analysis of the data. Propensity score adjustment was used to control for the effects of a broad range of covariates that encompassed sociodemographic, health, psychological and lifestyle characteristics of participants. Results A measurement model with social networks for children, relatives, friends and confidants was validated using confirmatory factor analysis. A variable that measured total social networks was also derived. Better social networks with relatives were protective against developing mobility disability over the nine year follow-up period (odds ratio (OR) 0.77; 95% confidence interval (95%CI) 0.62 – 0.96). A similar result was found for Nagi disability (OR 0.76; 95%CI 0.62 – 0.93). Other specific social networks did not have significant effects on either measure of disability. There were no significant effects of social networks on use of low-level residential care overall. There was a significant effect of social networks with confidants and total social networks, such that participants in the upper category of social networks with confidants appeared to be protected against use of high-level residential care (OR 0.53; 95%CI 0.35 – 0.81) compared to participants in the lower category of confidants social networks. Similarly, participants in the upper category for total social networks appeared to be protected against use of high-level residential care (OR 0.68; 95%CI 0.46 – 0.99). In terms of mortality, better social networks with confidants and with friends appeared to be protective against death during the decade following the Wave 1 interview. The hazard ratio (HR) for participants in the upper category for confidants was 0.74 (95%CI 0.63-0.88) compared to participants in the lower category. For friends networks, the analogous HR was 0.75 (95%CI 0.63-0.89). Better total social networks also appeared to be protective against death over the 10 years of follow-up (HR 0.83; 95%CI 0.70- 0.99). There were few significant effects of social networks with children on the three health outcomes considered. There was little evidence of threshold effects of the specific social networks on the health outcomes. Discussion There are important and differing effects of specific social networks on the three health outcomes of disability, residential care and mortality that were considered in this thesis. Policymakers may need to reconsider whether specific kinds of social relationships, beyond spouses and children, have been given adequate weight in current policy frameworks that address the health of older people. / http://proxy.library.adelaide.edu.au/login?url= http://library.adelaide.edu.au/cgi-bin/Pwebrecon.cgi?BBID=1321011 / Thesis (Ph.D.) -- University of Adelaide, School of Mathematical Sciences, 2008

In this dissertation, we have developed and combined several statistical techniques in Bayesian factor analysis (BAYFA) and mixture of factor analyzers (MFA) to overcome the shortcoming of these existing methods. Information Criteria are brought into the context of the BAYFA model as a decision rule for choosing the number of factors m along with the Press and Shigemasu method, Gibbs Sampling and Iterated Conditional Modes deterministic optimization. Because of sensitivity of BAYFA on the prior information of the factor pattern structure, the prior factor pattern structure is learned directly from the given sample observations data adaptively using Sparse Root algorithm. Clustering and dimensionality reduction have long been considered two of the fundamental problems in unsupervised learning or statistical pattern recognition. In this dissertation, we shall introduce a novel statistical learning technique by focusing our attention on MFA from the perspective of a method for model-based density estimation to cluster the high-dimensional data and at the same time carry out factor analysis to reduce the curse of dimensionality simultaneously in an expert data mining system. The typical EM algorithm can get trapped in one of the many local maxima therefore, it is slow to converge and can never converge to global optima, and highly dependent upon initial values. We extend the EM algorithm proposed by cite{Gahramani1997} for the MFA using intelligent initialization techniques, K-means and regularized Mahalabonis distance and introduce the new Genetic Expectation Algorithm (GEM) into MFA in order to overcome the shortcomings of typical EM algorithm. Another shortcoming of EM algorithm for MFA is assuming the variance of the error vector and the number of factors is the same for each mixture. We propose Two Stage GEM algorithm for MFA to relax this constraint and obtain different numbers of factors for each population. In this dissertation, our approach will integrate statistical modeling procedures based on the information criteria as a fitness function to determine the number of mixture clusters and at the same time to choose the number factors that can be extracted from the data.

The Multi-Agency Radiological Site Survey Investigation Manual (MARSSIM) is a regulatory guidance document regarding compliance evaluation of radiologically contaminated soils and buildings (USNRC, 2000). Compliance is determined by comparing radiological measurements to established limits using a combination of hypothesis testing and scanning measurements. Scanning allows investigators to identify localized pockets of contamination missed during sampling and allows investigators to assess radiological exposure at different spatial scales. Scale is important in radiological dose assessment as regulatory limits can vary with the size of the contaminated area and sites are often evaluated at more than one scale (USNRC, 2000). Unfortunately, scanning is not possible in the subsurface and direct application of MARSSIM breaks down. This dissertation develops a subsurface decision framework called the Geospatial Extension to MARSSIM (GEM) to provide multi-scale subsurface decision support in the absence of scanning technologies. Based on geostatistical simulations of radiological activity, the GEM recasts the decision rule as a multi-scale, geospatial decision rule called the regulatory limit rule (RLR). The RLR requires simultaneous compliance with all scales and depths of interest at every location throughout the site. The RLR is accompanied by a compliance test called the stochastic conceptual site model (SCSM). For those sites that fail compliance, a remedial design strategy is developed called the Multi-scale Remedial Design Model (MrDM) that spatially indicates volumes requiring remedial action. The MrDM is accompanied by a sample design strategy known as the Multi-scale Remedial Sample Design Model (MrsDM) that refines this remedial action volume through careful placement of new sample locations. Finally, a new sample design called “check and cover” is presented that can support early sampling efforts by directly using prior knowledge about where contamination may exist. This dissertation demonstrates how these tools are used within an environmental investigation and situates the GEM within existing regulatory methods with an emphasis on the Environmental Protection Agency’s Triad method which recognizes and encourages the use of advanced decision methods. The GEM is implemented within the Spatial Analysis and Decision Assistance (SADA) software and applied to a hypothetical radiologically contaminated site.

This study looks at alternative models of follow-up album success in order to determine if there is a relationship between the decrease in Metascore ratings (assigned by Metacritic.com) between the first and second album for a musician or band and the 1) music genre or 2) the number of years between the first and second album release. The results support the dominant thought, which suggests that neither belonging to a certain genre of music nor waiting more or less time to drop the second album makes an artist more susceptible to the Sophomore Slump. This finding is important because it forces us to identify other potential causes for the observed disappointing performance of a generally favorable musician’s second album.

Ovarian cancer (OC) is the fifth most common cancer among women and has the highest mortality of any cancer of the female reproductive system. The majority (61%) of OC cases are diagnosed at a distant stage. Because diagnoses occur most commonly at a late-stage and prognosis for advanced disease is poor, research focusing on the development of effective OC screening methods to facilitate early detection in high-risk, asymptomatic women is fundamental in reducing OC-specific mortality. Presently, there is no screening modality proven efficacious in reducing OC-mortality. However, transvaginal ultrasonography (TVS) has shown value in early detection of OC. TVS presents with the possibility of false positive results which occur when a women receives an abnormal TVS screening test result that is deemed benign following repeat testing (about 7% of the time). The purpose of this dissertation was to evaluate the impact of false positive TVS screening test results on a variety of psychological and behavioral outcomes using mixed and trajectory statistical modeling. The three specific aims of this dissertation were to 1) compare psychological and behavioral outcomes between women receiving normal and false positive results, 2) identify characteristics of women receiving false positive results associated with increased OC-specific distress and 3) characterize distress trajectories following receipt of false positive results. Analyses included a subset of women participating in an experimental study conducted through the University of Kentucky Ovarian Cancer Screening Program. 750 women completed longitudinal assessments: 375 false positive and 375 normal results. Mixed and group-based trajectory modeling were used to evaluate the specific aims. Results suggest women receiving false positive TVS result experience increased OC-specific distress compared to women receiving normal results. Among those receiving false positives, less education, no history of an abnormal screening test result, less optimism and more social constraint were associated with increased OC-specific distress. Family history was associated with increased distress among women with monitoring informational coping styles. Three distinct trajectories characterize the trajectory of distress over a four-month study period. Although decreasing over time, a notable proportion of women experience sustained high levels of OC-specific distress.

Copy-number variation (CNV) has been implicated in many complex diseases. It is of great interest to detect and locate such regions through genetic association testings. However, the association testings are complicated by the fact that CNVs usually span multiple markers and thus such markers are correlated to each other. To overcome the difficulty, it is desirable to pool information across the markers. In this thesis, we propose a kernel-based method for aggregation of marker-level tests, in which first we obtain a bunch of p-values through association tests for every marker and then the association test involving CNV is based on the statistic of p-values combinations. In addition, we explore several aspects of its implementation. Since p-values among markers are correlated, it is complicated to obtain the null distribution of test statistics for kernel-base aggregation of marker-level tests. To solve the problem, we develop two proper methods that are both demonstrated to preserve the family-wise error rate of the test procedure. They are permutation based and correlation base approaches. Many implementation aspects of kernel-based method are compared through the empirical power studies in a number of simulations constructed from real data involving a pharmacogenomic study of gemcitabine. In addition, more performance comparisons are shown between permutation-based and correlation-based approach. We also apply those two approaches to the real data. The main contribution of the dissertation is the development of marker-level association testing, a comparable and powerful approach to detect phenotype-associated CNVs. Furthermore, the approach is extended to high dimension setting with high efficiency.

The document is my habilitation thesis, which is a prerequisite for obtaining the "habilitation à diriger des recherche (HDR)" in France (https://fr.wikipedia.org/wiki/Habilitation_universitaire#En_France). The thesis is of cumulative form, thus providing an overview of my published works until summer 2013.

A graphical model is simply a representation of the results of an analysis of relationships between sets of variables. It can include the study of the dependence of one variable, or a set of variables on another variable or sets of variables, and can be extended to include variables which could be considered as intermediate to the others. This leads to the concept of representing these chains of relationships by means of a graph; where variables are represented by vertices, and relationships between the variables are represented by edges. These edges can be either directed or undirected, depending upon the type of relationship being represented. The thesis investigates a number of outstanding problems in the area of statistical modelling, with particular emphasis on representing the results in terms of a graph. The thesis will study models for multivariate discrete data and in the case of binary responses, some theoretical results are given on the relationship between two common models. In the more general setting of multivariate discrete responses, a general class of models is studied and an approximation to the maximum likelihood estimates in these models is proposed. This thesis also addresses the problem of measurement errors. An investigation into the effect that measurement error has on sample size calculations is given with respect to a general measurement error specification in both linear and binary regression models. Finally, the thesis presents, in terms of a graphical model, a re-analysis of a set of childhood growth data, collected in South Wales during the 1970s. Within this analysis, a new technique is proposed that allows the calculation of derived variables under the assumption that the joint relationships between the variables are constant at each of the time points.