Statistics Seminars at the University of Newcastle

Statistics seminars for 2011

Semester 1
Date	Presenter	Title	Abstract
2 May	Mr Kim Colyvas The University of Newcastle	Statistical Support Service - Past, Present and Future	[Click to show or hide]
		The Statistical Support Service has been operating since 2003 and provides statistical consulting services to University of Newcastle researchers. The first part of the talk will cover the development of the service, aspects of its operations, how researchers and research students are helped and challenges for the future. The second part will focus on perspectives relating to the teaching of statistics. The consulting process allows our consultants to sample the statistical maturity of research students and this maturity is in part a reflection of the statistical education they have received. We will discuss how this has influenced our teaching materials and some perspectives on the teaching of statistics for RHD students and undergraduate students.
30 May	Prof Dominique Haughton Bentley University and Université Toulouse I	Living Standards Analytics - Three Illustrative Examples	[Click to show or hide]
		This talk will outline several current main directions of interest in the analysis of household survey data, focusing on three illustrative examples. In the first example we demonstrate how Kohonen maps - otherwise known as self-organizing maps – can be used to address the issue of comparing the living standards of Vietnamese provinces. In the past single indicators such as the GDP per capita or the poverty rate have been used to rank provinces. However it is recognized that the living standard of a province is a multi-dimensional concept, and the problem arises of how to rank provinces on the basis of several indicators. Attempts to use composite indicators such as the Human Development Index (HDI) carry some limitations. The Kohonen map methodology makes it possible to map Vietnamese provinces on a two-dimensional grid, on the basis of a number of living standards indicators, and to give an interpretation of the dimensions on the grid. Our results shed some light onto some intriguing past rankings based on single or composite indicators, and at the same time demonstrate the usefulness of the Kohonen map methodology in the social sciences. The second example discusses a methodology to obtain small area estimates in the context of the Vietnam Living Standards Surveys. After very briefly introducing these surveys, we recall main concepts in small area estimation, and go on to discuss random effects in small area regression models, presenting two multilevel models for small area estimation at the commune level in Vietnam, one which applies to all communes and one to rural communes only. Our models for estimating the commune- level mean (log of) household expenditure per capita rely on independent variables available both in the 1999 Census (first model) and in the 2001 Agricultural Census (second model), and in the VHLSS of 2002. We follow ideas given in work by Moura (1994, 1999) and mention how to measure the accuracy of our model. Time permit

Semester 2
Date	Presenter	Title	Abstract
1 Jul	Dr Chuang Taipei Medical University	Systems Theory and Health Care: Crossing the Quality Chasm	[Click to show or hide]
		The presentation will provide some of Dr Chuang’s research and results of applying a system theory-based approach in the development of appropriate integration and analytical models within the context of the health care system. Two models will be discussed: 1) the holistic healthcare system relationship model: a framework for evaluating the effectiveness of the interrelated subsystems of the healthcare system, namely the accreditation, QMR and HCO health care systems, on improvements in health care; 2) Systems Oriented Event Analysis model (SOEA): a mechanism of guiding the events investigation team to be able to integrate systems changes in the way of systems thinking. Applications of the SOEA model have been found to succeed where traditional root cause analyses have failed.
4 Jul	Dr Patrick McElduff Hunter Medical Research Institute	Cluster randomised trials - what do you do when there is a small number of clusters?	[Click to show or hide]
		Randomised controlled trials (RCT) are considered the gold standard for testing new treatments or interventions, which are designed to improve the health and care of patients. Typically the unit of randomisation in an RCT is the individual patient. However in some trials the intervention is applied at a service level, such as a clinic, hospital or school and in these trials it is necessary to randomise all patients at a particular clinic, hospital or school to the same treatment arm. These trials are generally called cluster RCT and the analysis of data from these trials must account for the potential correlation of outcomes between patients within a cluster. In this seminar I will present some of the issues around the analysis of these trials with reference to a recent cluster RCT of an intervention to manage fever, hyperglycemia and swallowing dysfunction in acute stroke patients in 19 hospitals within NSW.
5 Aug	Mr Barrie Stokes The University of Newcastle	Maximum Entropy Alternatives to Pearson Family Distributions	[Click to show or hide]
		In the spirit of ET Jaynes' Maximum Entropy Principle, a (Bayesian) prior distribution conforming or constrained, say, to known moments should have the maximum entropy of all such distributions. At a previous MaxEnt conference [MaxEnt 2009, Oxford Mississippi] a method of obtaining MaxEnt univariate distributions under a variety of constraints was presented. The Mathematica function Interpolation[], normally used with numerical data, can also process "semi-symbolic" data, and Lagrange Multiplier equations were solved for a set of symbolic ordinates describing the required MaxEnt probability density function. We apply a developed version of this approach to finding MaxEnt distributions having prescribed beta1 (skewness squared) and beta2 (kurtosis) values, and compare the entropies of the MaxEnt distributions to those of the Pearson family distributions having the same beta1 and beta2. (This work was presented in poster form at the recent MaxEnt 2011 conference in Waterloo, Canada.)
9 Sep	Dr John Best The University of Newcastle	A Two-Part Talk: 1. Tests for Latin Squares and 2. Tests for Randomized Block Categorical Data	[Click to show or hide]
		Two projects, joint with Prof John Rayner, which are still "work in progress" will be briefly described. The first relates to finding a nonparametric test for replicated latin squares and the second concerns a new algorithm for a Chi-square test of homogeneity when the counts are not based on independent observations (being derived from a randomized block layout). The talk will be at an introductory level and sensory evaluation data will illustrate the methods.
14 Oct	Dr Darfiana Nur The University of Newcastle	Nonlinear Time Series Modelling in Bionformatics and Finance: Bayesian Perspectives	[Click to show or hide]
		I would like to share some research progress/problems from my recent study leave. There are two main components in this talk: DNA sequence modelling in Bioinformatics and Bayesian analysis of a modified Smooth Threshold Autoregressive (STAR) models in Finance. For the DNA sequence modelling, I continued my research in Bayesian Hidden Markov modelling for some new DNA sequences and extended the research to model short DNA motif segmentation. Computing problems were encountered due to the complexity of the model and the size of DNA data. For the STAR models in Finance, I partly derived a Bayesian analysis of STAR and extended it to STAR-GARCH models of order p by obtaining the posterior distributions of their parameters based on both noninformative and conjugate priors.
4 Nov	Chris Oldmeadow Hunter Medical Research Institute	Bayesian Networks: overview and two applications in health and genetics	[Click to show or hide]
		Bayesian Networks (BNs) are a class of probabilistic graphical models for reasoning and decision making within systems that contain uncertainty. They are useful for both understanding underlying mechanisms of a system and for updating beliefs when new data are available. In a BN, the variables within the system are depicted as nodes in a Directed Acyclic Graph (DAG) and conditional probability distributions can be assigned by domain experts or estimated from the data. Under certain assumptions, causal dependencies between variables are captured by the DAG and thus these models are very popular for causal inference when only observation data are available. In this talk I will give a general overview of inference in Bayesian Networks, focusing on the case where the causal structure of the system is assumed to be known. Two applications in the fields of Health and Genetics will be presented.

View seminars for: 2011, 2012, 2013, 2014, 2015, 2016, 2017