Index
2010
2 July 2010, Ingram Olkin
26 May 2010, Shirley Pledger and Richard Arnold
9 February 2010, John Maindonald
2009
8 September 2009, Vijay Nair
2 April 2009, Ross Ihaka
2 February 2009, Brian Easton and Ryan You
2008
20 October 2008, Jim Ridgway
3 June 2008, Pra Murthy
23 April 2008, Nicholas Horton
3 April 2008, Ray Chambers
18 March 2008, Jiancang Zhuang
4 March 2008, Martin Bland
7 February 2008, J.P. Lewis
2007
4 December 2007, I-Ming Liu
20 November 2007, Mike Camden and Paul Cowie
24 October 2007, Jim Renwick
18 September 2007, Peter Thomson
6 August 2007, Nick Longford
9 May 2007, Shirley Pledger
18 April 2007 (6pm), Stephen E. Fienberg
18 April 2007 (Noon), Stephen E. Fienberg
16 April 2007, Stephen E. Fienberg
22 March 2007, Richard Arnold
2006
28 November 2006, Nanny Wermuth
22 November 2006, Estate Khmaladze
1 November 2006, Len Cook
12 October 2006, Dimitar Christozov and Stefanka Chukova
25 July 2006, Edith Hodgen, Rachel Dingle and Hilary Ferral
20 June 2006, Ian Westbrooke
16 March 2006, Geoff Chambers
2005
13 December 2005, Rod Lea
8 November 2005, Tony Vignaux
14 September 2005, Paul Jose
2 August 2005, Phil Lester
19 May 2005, Statistical Education
13 April 2005, Mark Weatherall
24 February 2005, Tim Ball
2004
7 December 2004, Estate Khmaladze
18 November 2004, Robin Willink
30 September 2004, Robert Davies
27 May 2004, Jeff Robinson
29 April 2004, James Liu
25 March 2004, Shirley Pledger
11 February 2004, Brian Easton
2003
26 November 2003, Stefanka Chukova and Yu Hayakawa
28 October 2003, Nick Longford
14 August 2003, Caroline Roughneen
9 July 2003, Mark Weatherall
11 June 2003, Leigh Bull
1 May 2003, Chris Francis
27 March 2003, Srinivas Chakravarthy
2002
17 October 2002, Frances Krsinich and Mike Camden
27 August 2002, Paul Dyer
24 July 2002, Jean Thompson
21 February 2002, Richard Arnold
2001
6 December 2001, Brian Bull
16 October 2001, Dean Hyslop and Dave Mare
8 August 2001, Brian Pink
2 July 2010, Ingram Olkin
Speaker: Ingram Olkin, Stanford University
Title: Meta-Analysis: History and Statistical Issues for Combining the Results of Independent Studies
Time: 6:00pm, Friday 2 July (light refreshments from 5:30pm)
Place: Room RH LT3, Rutherford House, Lambton Quay
Abstract:
Meta-analysis enables researchers to synthesize the results of independent
studies so that the combined weight of evidence can be considered and
applied. Increasingly meta-analysis is being used in medicine and other
health sciences, in the behavioral and educational fields to augment
traditional methods of narrative research by systematically aggregating
and quantifying research literature.
Meta-analysis requires several steps prior to statistical analysis:
formulation of the problem, literature search, coding and evaluation of the
literature, after which one can address the statistical issues.
We here review some of the history of meta-analysis and discuss some of
the problematic issues such as various forms of bias that may exist. The
statistical techniques that have been used are nonparametric methods,
combining proportions, the use of different metrics, and combining
effect sizes from continuous data.
About the Speaker:
Professor Ingram Olkin (Stanford University) is the NZSA Visiting Lecturer for 2010. His visit to New Zealand is sponsored by Statistics New Zealand and is associated with the International Conference on Statistical Methodologies and Related Topics, in conjunction with the NZSA 2010 Conference, 29 June - 1 July 2010, Massey University, Palmerston North.
Dr Olkin is an icon in the world statistical community, having been active for over 60 years. He is a member of many professional societies, has received many honours and awards, has held and holds many editorial positions, and has delivered numerous invited addresses all over the world. Ingram has coauthored 7 books, edited 10 books, and contributed 220 journal papers. His joint paper with Albert Marshall "A multivariate exponential distribution" was cited in over 600 articles.
Dr Olkin's work is aimed at ensuring that educators select the proper statistical tools for measuring the outcomes of their programs and methods, and that their interpretation of the results is similarly rigorous. His research includes the development of powerful new statistical methods for combining results from independent studies that have analysed the same topic. Meta-analysis is assisting researchers to reconsider long-standing educational problems with a fresh critical eye.
Dr Olkin is a Guggenheim, Fulbright, and Lady Davis Fellow, with an honorary doctorate from De Montfort University. He received his BS in mathematics at the City College of New York, his MA from Columbia University, and his PhD from the University of North Carolina. Dr Olkin's research interests include analysis of social and behavioural models; multivariate statistical analysis; correlational and regression models in educational processes; and meta-analysis.
26 May 2010, Shirley Pledger and Richard Arnold
Speakers: Shirley Pledger and Richard Arnold, Victoria University
Title: Clustering and Pattern Detection in Ecological Models Using Mixtures
Time: 6:00pm, Wednesday 26 May (light refreshments from 5:30pm)
Place: Room RH LT2, Rutherford House, Lambton Quay
Abstract:
Incidence matrices record the presence or absence of species in particular
habitats or locations, while abundance matrices record the count of each
species. Such matrices can be used to identify species which associate, and
the habitats in which they do so. This can be achieved by clustering the
rows or the columns of an incidence or abundance matrix, or clustering both
rows and columns simultaneously.
Finite mixture models – where species and habitats are separately and
probabilistically classified into groups – provide a tractable means of
carrying out this clustering, and yield results which are readily
interpretable in their ecological context. We will present a likelihood-based
approach to the two-way clustering using finite mixtures.
Further developments provide likelihood-based analogues of a range of
multivariate techniques, including correspondence analysis and
multidimensional scaling, which traditionally use matrix decompositions.
Generally the mixture modelling gives clustering and ordination results
similar to those from the standard methods, but with extra information
available from the likelihoods for pattern description and statistical
analysis.
9 February 2010, John Maindonald
Speaker: John Maindonald, Australian National University
Title: Mining a Cricketer Data Archive
Time: 6:00pm, Tuesday 9 February (light refreshments from 5:30pm)
Place: Room LT4, VUW Law School, Old Government Buildings, Lambton Quay (First floor, north wing at back)
Abstract:
The data, compiled by psychologist John Aggleton from information in
the "Who's Who of Cricketers", has details on the birth year and
bowling hand of British first class cricketers who were born in one of
the years 1840–1960. For those who died in 1990 or earlier, age at
death is recorded. Also noted are instances of death by accident,
or in war.
Martin Bland and Douglas Altman used the data to investigate whether
left-handed cricketers die young. There is more that can be gleaned
from these data. They provide a commentary on the dislocation that
World War 1 caused to British society.
This has been an exercise in data mining, with a relatively automated
use of what are now standard statistical learning approaches, here
used to criticise and supplement more classical approaches. As
happens here, data frequently have a value that extends beyond the use
for which they were initially collected.
8 September 2009, Vijay Nair
Speaker: Vijay Nair, University of Michigan
Title: Statistical Inverse Problems in Network Tomography and Monitoring Quality of Service Characteristics in Networks
Time: 6:00pm, Tuesday 8 September (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
The term network tomography refers to two classes of large-scale inverse problems that arise in the modeling and analysis of computer and communications networks. The first deals with passive tomography where aggregate data are collected at the individual router/node level and the goal is to recover path-level information. The main problem of interest here is the estimation of the origin-destination traffic matrix. The second, sometimes referred to as active tomography, deals with reconstructing link-level information from end-to-end path-level measurements obtained by actively probing the network. The primary application in this case is estimation of quality-of-service parameters such as loss rates and delay distributions. Internet service providers use this to characterize network performance and to monitor service quality. We will review both classes of inverse problems with a focus on active tomography. We will provide a review of recent developments, including the design of probing experiments, inference for loss rates and delay distributions, and applications to network monitoring. This is joint work with George Michailidis, Earl Lawrence, Bowei Xi, and Xiaodong Yang.
About the Speaker:
Vijay Nair is the Donald A. Darling Professor and Chair of Statistics, and the
Professor of Industrial and Operations Engineering at the University of
Michigan. He is also a current Vice President of the International Statistical
Institute. Prof Nair is the 2009 Shayle Searle Visiting Fellow in Statistics at
Victoria University.
2 April 2009, Ross Ihaka
Speaker: Ross Ihaka, The University of Auckland and the R Foundation
Title: R : Past and Future History
Time: 6:00pm, Thursday 2 April (light refreshments from 5:30pm)
Place: Room LT4, VUW Law School, Old Government Buildings, Lambton Quay (1st floor, north wing at back)
Abstract:
R is a free software environment for statistical data analysis. It
began life as a purely academic experiment, but has developed into one
of the more widely-used statistical software environments.
This talk will describe how R came into being and developed into its
current state. It will describe the current state of R development and
also cover more recent work which seeks to deliver even better
computing environments.
2 February 2009, Brian Easton and Ryan You
Speakers: Brian Easton and Ryan You
Title: Measuring Gambling Experiences of New Zealanders
Time: 6:00pm, Monday 2 February (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
In 2008, the Centre for Social and Health Outcomes Research and
Evaluation (SHORE) at Massey University published a study based on the
gambling experiences of 7000 New Zealanders. Brian Easton and Ryan You
will report on the statistical work they did to assess the extent to
which gambling affects individual's welfare.
20 October 2008, Jim Ridgway
Speaker: Jim Ridgway, School of Education, University of Durham UK
Title: The OECD Global Project – Measuring the Progress of Societies,
Thoughts and Actions
Time: 6:00pm, Monday 20 October (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
The OECD has launched the Global Project on Measuring the Progress of Societies (GP) to encourage the use of new indicator systems in order to inform and promote evidence-based decision-making within and across the public and private sectors. Put baldly, the GP aims to invigorate democracies by getting citizens involved with the creation of measures of societal progress, and by holding governments to account on a raft of measures, "beyond GDP". There is an underlying belief that an emphasis on economic indicators alone is a recipe for global disaster.
Statistics and statisticians have a central role to play, and key concepts in statistics need to be absorbed into the "common sense" of engaged citizens.
There are a number of key activities, including:
Eliciting what communities think "progress" means in the 21st century;
Advocating appropriate investment in building statistical capacity;
Developing and sharing good practices on the measurement of societal progress (with an emphasis on using sound and reliable methodologies);
Producing a broader, shared, public understanding of changing global conditions, while highlighting areas of the most dramatic changes and of inadequate knowledge; and
Stimulating international debate, based on solid statistical data and indicators, on both global issues of societal progress and comparisons of such progress.
This paper will provide experiences of some of the activities in the first OECD training session for statisticians, held in Siena in September 2008, and will show some visualization tools designed to facilitate the transition from data to knowledge.
The whole enterprise is conceptually and technically challenging. I hope we can explore these challenges in the discussion.
3 June 2008, Pra Murthy
Speaker: Pra Murthy, University of Queensland
Title: Warranty: An Introduction
Time: 6:00pm, Tuesday 3 June (light refreshments from 5:30pm)
Place: Room LT4, VUW Law School, Old Government Buildings, Lambton Quay (1st floor, north wing at back)
Abstract:
Most products (consumer, industrial and commercial) are sold with warranty. A warranty is a contractual agreement which requires the manufacturer to repair, replace or provide some compensation for failures occurring within the warranty period. There are various aspects to warranty and they have been studied from different perspectives. The seminar will give an overview of this.
23 April 2008, Nicholas Horton
Speaker: Nicholas Horton, Smith College, Northampton MA
Title: What your physician should know about statistics but perhaps doesn't: The implications of the increasing sophistication of statistical methods in medical research
Time: 6:00pm, Wednesday 23 April (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
A recent survey of original articles published in The New England
Journal of Medicine revealed increasing use of statistical methods over
time, compared with surveys conducted in 1979 and 1989 by Emerson and
Colditz. Of 311 articles published in 2004–2005, a substantial fraction of
articles utilized relatively sophisticated statistical methodologies
such as survival analysis (61%), multiple regression (51%) or power
calculations (39%). Only 13% of articles used just simple descriptive
statistics (e.g. percentages, means, confidence intervals). Knowledge
of material typically included in an introductory statistics course
increased this percentage to only 21%. The statistical training
required prior to entry into as well as that provided during medical
school is quite minimal, and this increasing sophistication complicates
the interpretation and dissemination of new results, particularly
for clinicians who have not received additional training in the
conduct of research. We discuss the implications of this increased
use of sophisticated statistical methods for medical and statistical
education.
3 April 2008, Ray Chambers
Speaker: Ray Chambers, University of Wollongong, and NZSA Visiting Lecturer for 2008
Title: Measurement Error in Auxiliary Information
Time: 6:00pm, Thursday 3 April (light refreshments from 5:30pm)
Place: Room LT4, VUW Law School, Old Government Buildings, Lambton Quay (1st floor, north wing at back)
Abstract:
Auxiliary information is information about the target population of a sample survey over and
above that contained in the actual data obtained from the sampled population units. The
availability of this type of information represents a key distinction between sample survey
inference and more mainstream inference scenarios. In particular, modern methods of
sampling inference (both model-assisted as well as model-based) depend on the availability of
auxiliary information to improve efficiency in survey estimation. However, such information
is not always of high quality, and typically contains errors. In this talk I focus on some
survey-based situations where auxiliary information is crucial, but where this information is
not precise. Estimation methods that allow for this imprecision will be described. In doing so
I will not only address the types of inference of concern to sampling statisticians (e.g.
prediction of population quantities), but also inference for parameters of statistical models for
surveyed populations.
About the Speaker:
Ray Chambers is Professor of Statistical Methodology at the University of Wollongong. He has extensive research interests in the design and analysis of sample surveys, official statistics methodology, robust methods for statistical inference and analysis of data with group structure. He is also the NZSA Visiting Lecturer for 2008.
18 March 2008, Jiancang Zhuang
Speaker: Jiancang Zhuang, Institute of Statistical Mathematics, Tokyo
Title: Visualizing patterns in earthquake clusters by using a point
process model
Time: 6:00pm, Tuesday 18 March (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
Point processes have been widely used in modelling seismicity, and other natural and
social phenomena. Among them, the epidemic-type aftershock sequence
(ETAS) model is a space-time
continuous-type point process for describing the clustering behaviour
of earthquake occurrences. This talk gives a real example using this
model to analyze characteristics and test hypotheses associated
with earthquake clustering. Such questions include the relationship between
the size of ancestor events to the number of triggered events
(offspring), the size of the region, etc. A form of model residual analysis
will also be described and shown how it can be used as a powerful tool in
model improvement.
4 March 2008, Martin Bland
Speaker: Martin Bland, Professor of Health Statistics, University of York, England
Title: Risedronate, the BBC, and me
Time: 6:00pm, Tuesday 4 March (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
Risedronate is a drug used to reduce fractures in women with osteoporosis. The story of the Sheffield Risedronate trial was national news. One of the clinical researchers, Aubrey Blumsohn, dissociated himself from the trial because he was refused access to the data he had helped to collect. He did not believe the analysis provided by the pharmaceutical company statistician. Then some of the data were obtained. I shall describe how I came to analyse the data on behalf of the BBC, what I found, and what happened next.
About the Speaker:
Prof Bland coauthored a paper with Doug Altman in 1986 that was published in the Lancet, and titled "Statistical Methods for Assessing Agreement". By 1992 it became a citation classic, and by April 2010 it had been cited over 16,800 times. A copy of the paper (and much other material) can be found on his web page: http://www-users.york.ac.uk/~mb55/.
7 February 2008, J.P. Lewis
Speaker: J.P. Lewis, Weta Digital (contractor)
Title: Statistical Applications in Computer Graphics
Time: 6:00pm, Thursday 7 February (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
Computer graphics modeling and animation have traditionally been approached as a craft, using models that are manually constructed and animated. With plentiful computer storage and techniques such as non-parametric regression it is increasingly possible to build models directly from data, as well as to characterize the ways that these models change. This talk will describe film and gaming applications such as character skinning, eye movement synthesis, and automatic caricature, and present research on identifying and characterizing structure in multidimensional data. The discussion will reference PCA, kernel density, robust norms, MDL, and graph cuts.
4 December 2007, I-Ming Liu
Speaker: Ivy (I-Ming) Liu, Victoria University of Wellington
Title: The analysis of ordered response data
Time: 6:00pm, Tuesday 4 December (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
This talk has two parts. First, I will review methodologies used for
analysing ordered categorical (ordinal) response variables by
introducing models for data with a single ordinal response variable.
The second part is based on the talk given at the International
Statistical Institute Conference in August 2007. It provides graphical
methods for checking the adequacy of the proportional odds model,
which is currently the most popular model for ordinal responses. The
methods using the cumulative sums of residuals focus on evaluating
model misspecification for specific covariates.
20 November 2007, Mike Camden and Paul Cowie
Speakers: Mike Camden and Paul Cowie, Statistics New Zealand
Title: Confidentiality and Census Counts: the best of both worlds
Time: 6:00pm, Tuesday 20 November (light refreshments from 5:30pm)
Place: Room LT2, VUW Law School, Old Government Buildings, Lambton Quay (note that GB LT2 is in the lecture block behind the old wooden building)
Abstract:
Tables of counts from our Census of Population and Dwellings are valued
highly by users, especially users from local government. The Census also
appear to be valued and strongly supported by the 4,249,737 respondents,
who form about 98% of the population. The two worlds are those of users
and respondents, and they are interested in data utility and safety
respectively. Users request detailed tables that are often very sparse.
Statistics New Zealand's past protections against sparseness include
random rounding to base 3 and a mean cell size rule. We decided, for our
2006 Census, to further enhance data utility and protect safety by setting
a threshold for sparse tables: counts above this are released and counts
at this value or below are suppressed.
The confidentiality rules are applied to each geographical area
separately, and our current geographical areas vary widely in population.
This variation is a source of sparseness, but we turn this liability into
an asset. For each of our widely varying Area Units, we calculate measures
for utility and safety, and get some expected and unexpected messages from
the data, via some data visualisation. We use this data to evaluate the
threshold approach, and also to help in setting the size for a possible
new set of geographical areas for NZ, designed for output.
We hope to use these results to heighten utility and safety for NZ's
Census 2011.
Presenters: Paul and Mike both work in the Statistical Methods Division of Statistics
NZ, and both spend much of their time aiming to ensure that NZ has the
best confidentiality systems available. Paul is a graduate from the
University of Auckland, in statistics and commerce, and Mike has done time
at Auckland, Waikato and VUW. The talk is an expanded version of
Statistics NZ's contribution to the Joint UNECE/Eurostat Work Session on
Statistical Data Confidentiality in December.
24 October 2007, Jim Renwick
Speaker: Jim Renwick, NIWA, Wellington
Title: Recent climate change assessment: physical science and
statistical interpretation
Time: 6:00pm, Wednesday 24 October (light refreshments from 5:30pm)
Place: Room HMLT001, Hugh MacKenzie Building, Kelburn Parade (note the change of location)
Abstract:
The Intergovernmental Panel on Climate Change (IPCC) recently released
its fourth Assessment Report (AR4), covering past observations of
climate changes and projections of likely future global changes. This
presentation will summarise the role of the IPCC, and outline some of
the key messages from the AR4 in terms of both the past and the future.
Beyond the physics of climate and climate change, the talk will cover
some aspects of climate change monitoring and prediction, including the
estimation of probability densities for climate parameters in the future.
Speaker: Dr James Renwick leads a research programme on climate
variability and change at NIWA in Wellington. His interests include
Southern Hemisphere climate variability and impacts of climate on the
New Zealand environment. He has worked in the area of climate and
atmospheric dynamics and statistics since the late 1970s, starting as a
weather forecaster at the Meteorological Service and completing a MSc
through ISOR at Victoria University in the late 1980s. He was a lead
author on Working Group I of the IPCC 4th Assessment Report, dealing
with observed changes to the large-scale circulation of the atmosphere.
18 September 2007, Peter Thomson
Speaker: Peter Thomson, Statistics Research Associates Ltd
Title: Hidden Markov models: some examples of their
application and reflections on their use
Time: 6:00pm, Tuesday 18 September (light refreshments from 5:30pm)
Place: Room LT3, VUW Law School, Old Government Buildings, Lambton Quay (Ground floor, north wing at back)
Abstract:
Hidden Markov models were first introduced in the statistical literature by Baum and
his colleagues in the late 1960s and have since become widely used in many disciplines
including meteorology, economics, finance and speech recognition, to name but a few. In
effect, a hidden Markov model (HMM) blocks time series data into consecutive periods of
time (regimes) within which the observations follow a simple regime-dependent time series
model. Switching to and from regimes is governed by an unobserved (hidden or latent)
Markov chain or variant thereof. In this way the time evolution of regimes is directly
modelled as is the evolution of the observations within regimes. This conceptually simple
and open structure allows direct modelling of time and spatial scales that may be present
in the data, as well as opportunities for enhanced interpretation and more physically based
models.
The advantages and limitations of using hidden Markov models in practice will be discussed with reference to the particular HMM applications in which we have been involved.
These will mainly focus on modelling rainfall (daily and high frequency) and GDP growth
rates (quarterly), with other applications in finance and river-flow modelling (actual and
potential) mentioned in passing. The focus will be on practical rather than technical
issues and, in particular, the need to explore and exploit the structure of HMMs.
6 August 2007, Nick Longford
Speaker: Nick Longford, SNTL, Reading, England
Title: Allocation of limited resources and related problems in small-area statistics
Time: 6:00pm, Monday 6 August (light refreshments from 5:30pm)
Place: Room LT4, VUW Law School, Old Government Buildings, Lambton Quay (1st floor, north wing at back)
Abstract:
Small-area statistics is the generic term used for estimation of one or a
set of quantities associated with each district of a country. Development
of methods for small-area estimation has been motivated mainly by the
increased demand for more and more detailed information about the social,
economic, health-related and other processes in a country that are in the
remit of its government. The key principle in small-area estimation is
borrowing strength, or exploiting similarity of the districts. The
overall research strategy may be summarised as follows: first we estimate
as efficiently as possible the quantities that our clients would ideally
like to have, and then we let them handle the estimates as if they were
the underlying (population) quantities, after instructing them about
standard errors or confidence intervals.
The talk will discuss an attempt to distribute a fixed budget to the
districts of a country for combating a specific problem, such as high
unemployment. The distribution is to be based on district-level estimates
of the quantity of interest (unemployment rate). It is easy to show, by
simulations, that efficient estimation is not conducive to effective
distribution of the budget. Some solutions better than efficient
estimation are proposed, but they do not have a straightforward
derivation. The problem is connected to the use of estimates in general,
and the conclusion implies that the separation of the activities of
estimation and decision making, which promotes integrity of one
kind, results in suboptimal decisions.
The talk will be of interest to lecturers of elementary statistics, as the
ANOVA, hypothesis testing, and a few similar rituals will be exposed as
way past their due dates, not only for the practice, but even for the
curriculum.
9 May 2007, Shirley Pledger
Speaker: Shirley Pledger, School of Mathematics, Statistics and Computer Science, Victoria University of Wellington
Title: Something for Nothing: Estimating age-related survival rates from
capture-recapture data when age is unknown
Time: 6:00pm, Wednesday 9 May 2007 (light refreshments from 5:30pm)
Place: VUW Law School (Old Government Buildings), Room GBLT4 (1st floor,
north wing at back)
Abstract:
Age-structured population projections are used extensively in population
dynamic studies and in risk assessment, and capture-recapture methods may be
used to provide survival rate estimates. However, the age of an animal first
captured as an adult will usually not be known.
We construct likelihood-based models which allow for uncertainty of birth
time. Age-dependent hazard rates and senescence may be modelled, and
appraised by AIC or likelihood ratio tests.
These models are well suited to isolated populations (with little or no
immigration), or to species with high site fidelity after a juvenile
dispersal phase. An application that uses the possum data set from the
Orongorongo Valley is given.
The same models are also applicable if "birth" is immigration and "death" is
emigration, and negligible actual births and deaths occur. This is the
situation with migratory birds, which stop over at a site to build up body
reserves before continuing the journey. If the best predictor of the current
probability of imminent departure is time since arrival, this should be
included in the model, even though the exact arrival time is unknown.
18 April 2007 (6pm), Stephen E. Fienberg
Speaker: Stephen E. Fienberg,
Department of Statistics, Machine Learning Department, and Cylab, Carnegie
Mellon University, Pittsburgh, PA 15213, U.S.A.
Title: When Did Bayesian Inference Become "Bayesian"?
Time: Wednesday 18 April 2007, refreshments at 5:30 pm, talk at 6pm
Place: McLaurin Theatre MCLT102, Kelburn Campus
Abstract:
While Bayes' theorem has a 250-year history, and the method of inverse
probability that flowed from it dominated statistical thinking into the
twentieth century, the adjective "Bayesian" was not part of the statistical
lexicon until relatively recently. This talk provides an overview of key
Bayesian developments, beginning with Bayes' posthumously published 1763
paper and continuing up through approximately 1970, including the period of
time when "Bayesian" emerged as the label of choice for those who advocated
Bayesian methods.
18 April 2007 (Noon), Stephen E. Fienberg
Speaker: Professor S.Feinberg, Maurice Falk Professor of Statistics and Social Science, Carnegie Mellon University
Title: Bayesian Mixed Membership Models for Soft Clustering
Time: 12:00pm, 18 Apr 2007 (Wednesday)
Place: Seminar Room, Cotton 431, Victoria University of Wellington
Abstract :
In many problem settings involving clustering and classification, units can
conceivably belong to multiple group. Bayesian mixed membership models
provide a natural way to address such "soft" clustering and classification
problems. These models typically rely on four levels of assumptions:
population, subject, latent variable, and sampling scheme. Population level
assumptions describe a general structure of the population that is common to
all subjects. Subject level assumptions specify the distribution of
observable responses given the population structure and individual
membership scores. Membership scores are usually unknown and hence can also
be viewed as latent variables which can be treated as fixed or random in the
model. Finally, the last level of assumptions specifies the number of
distinct observed characteristics (attributes) and the number of
replications for each characteristic. We describe four applications of mixed
membership modeling: (i) to disability indicators from the National Long
Term Care Survey, (ii) abstracts and bibliographies of research reports in
The Proceedings of the National Academy of Sciences, (iii) genetic SAGE
libraries, and (iv) protein-protein interactions in yeast (this involves
extensions that incorporate stochastic block-modeling). Our methods include
the computation of full posterior distributions as well as various forms of
variational approximations. In the examples, we also discuss issues of model
assessment and specification.
16 April 2007, Stephen E. Fienberg
Speaker: Professor S.Feinberg, Maurice Falk Professor of Statistics and Social Science, Carnegie Mellon University
Title: Making Forensic Science More Scientific: Statistics and the Evaluation of
Forensic Evidence
Time: 12:00pm, 16 Apr 2007 (Monday)
Place: Railway West Wing, Victoria University of Wellington
Abstract:
In the US, forensic science is under increasing attack. This is the consequence of the confluence of a number of elements including (a) continued revelations of wrongful convictions linked to faulty forensic evidence, (b) the resounding success of DNA and other genetic evidence in a forensic context, and (c) the "CSI Effect"---the expectation of infallible high tech forensic tools that are part of the popular weekly crime show, Crime Scene Investigation. In this talk I will describe a potpourri of forensic tools (e.g., the polygraph, eyewitness testimony, traditional fingerprinting, and new computer forensic tools), legal cases in which they arise, and the role statistics plays in their evaluation and legal credibility.
22 March 2007, Richard Arnold
Speakers: Richard Arnold and John Townend,
Victoria University of Wellington
Title: Using earthquakes to measure stress in the earth's crust
Time: 6:00pm, Thursday 22 March 2007 (light refreshments from 5:30pm)
Place: VUW Law School (Old Government Buildings), Room GBLT4
Abstract:
When an earthquake occurs a planar weakness in the rocks of the earth's
crust (a fault) is overwhelmed by the ambient stresses (pressures), and the
rocks on one side of the fault slide, releasing seismic waves. The direction
of the slip and the orientation of the fault plane on which the slip happens
contain information about the nature of the stresses which caused the
earthquake. The information about the stress provided by a single earthquake
is very weak, but many separate earthquakes can be used to constrain the
stress parameters more strongly. This talk will present a new Bayesian
solution to extracting information about the stress which drives
earthquakes.
28 November 2006, Nanny Wermuth
Speaker: Nanny Wermuth, Chalmers/Gothenburg University, Sweden
Title: Distortions of effects
Time: 6:00pm, Tuesday 28 Nov 2006
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
Unnoticed confounding may severely distort the direction and strength
of the effect of an explanatory variable on its response variable, as
given by a stepwise data generating process. For direct confounding
this effect is known. If it arises from a common unobserved explanatory
variable, it is relevant mainly for observational studies, since it is
avoided by successful randomization. By contrast, indirect confounding,
which is discussed now, is an issue also for intervention studies. For
general stepwise generating processes, we provide matrix and graphical
criteria to decide which types of direct or indirect confounding may be
present and when they are absent.
22 November 2006, Estate Khmaladze
Speaker: Estate Khmaladze, Victoria University of Wellington
Title: On distributions that do not follow asymptotic theory and other anomalies
Time: 6:20pm, Wednesday 22 Nov 2006
Place: Lecture Theatre 3, Ground Floor, Rutherford House, 23 Lambton
Quay
Abstract:
It is so interesting to calculate or simulate things in probability
and statistics that our science has almost been converted into a sort
of experimental science. However, the most interesting such
calculations are when, instead of showing how great the theory is,
they disagree with mathematical statements. In cases like this we
typically do not understand the phenomena enough or our intuition is
somewhat conservative and, hence, there is a chance of discovery.
In the talk we present two phenomena of this kind. One is exact values
for non-crossing probabilities for finite samples and their limit. We
will argue that we do not understand enough the tails of empirical
distributions. In other words, we do not understand enough the Law of
the Iterated Logarithm: one of the basic laws of probability and
statistics. The second is another look on samples and likelihood ratios.
We illustrate the unusual yet typical behaviour of ordinary likelihood
ratio statistics, for samples of size n=100, and even n=20.
The material of the first part can be found in the speaker's joint paper
with a former PhD student, Eka Shinjikashvili: J. Appl. Prob., 2002. For
the illustrations in the second part we are indebted to the curiosity of
Nick Webb, a current VUW Graduate student.
1 November 2006, Len Cook
Speaker: Len Cook, Former NZ Government Statistician (1992–2000) and
Head of the UK Office for National Statistics, the Registrar General of
England and Wales, and the first UK National Statistician (2000–2005)
Title: What might official statistics in the Antipodes learn from the British
statistical system, and vice versa?
Time: 6:00pm, Wednesday 1 Nov 2006
Place: Lecture Theatre 3, Ground Floor, Rutherford House, 23 Lambton
Quay
Abstract:
Over the past sixty years, the UK, Australia and New Zealand have had
quite different approaches to the organisation of official statistics,
but these differences are now diminishing. The forces we face in
common are now the more critical determinants of performance.
Antipodean institutions can learn from the responsiveness of the UK
statistical system, while the UK has long needed the stronger focus on
investment of both New Zealand and in particular Australia. The UK
environment is a tough one, and the quest for broader robust and timely
solutions exists in more fields than in the Antipodes, partly because
of the size of the UK, and the extent of its analytical history.
We are well equipped to share more, as a global media, international
collaboration in research and global convergence of policy settings
demand coherence nationally, regionally and now globally in statistical
practice.
12 October 2006, Dimitar Christozov and
Stefanka Chukova
Speakers: Dimitar Christozov, American University in Bulgaria; and
Stefanka Chukova, Victoria University of Wellington
Title: Estimation of the
mean cumulative function from automotive warranty data: a stratification
approach
Time: 6:00pm, Thursday 12 October 2006
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings, 15 Lambton Quay
Abstract:
This study deals with a type of truncation that occurs with typical automotive warranties. Warranty coverage and the resulting claims data are limited by mileage as well as age. Age is known for all sold vehicles all the time, but mileage is only observed for a vehicle with a claim and only at the time of the claim. Here we deal with the univariate case, taking either age or mileage as the usage measure. We evaluate the mean cumulative number of claims or
cost of claims and its standard error as functions of the usage
measure. Within a nonparametric framework, we extend the usual methods in order to account for the fact that the odometer readings are available only for a vehicle with a claim and only at the time of the claim. We illustrate the ideas with real data on four cases based on whether the usage measure is age or miles and whether the results are adjusted for withdrawals from warranty coverage. This study is an extension of Chukova and Robinson's (2005) findings.
Firstly, we weaken their assumption of having a linear relationship
between the mileage accumulation and age of a car and, secondly, our model accounts for multiple claims whereas the previous model was based on the last claim only.
Reference: S. Chukova, J. Robinson (2005). Estimating mean cumulative functions from truncated automotive warranty data. Modern Statistical and Mathematical Methods in Reliability, Eds. A. Wilson, N. Limnios, S. Keller-McNulty, Y. Armijo, World Scientific, Singapore, 2005, pp.121–135.
25 July 2006, Edith Hodgen, Rachel Dingle and Hilary Ferral
Speakers: Edith Hodgen, Rachel Dingle and Hilary Ferral, New Zealand Council for Educational Research
Title: Statistics: a growth area for NZCER
Time: 6:00pm, Tuesday 25 July 2006
Abstract:
The New Zealand Council for Educational Research (NZCER) has been in existence for 72 years, but it is only for the last four that the council has employed at least one statistician. Having started with one, we have appointed an additional statistician almost every year since then.
We plan to outline the range of work carried out at NZCER, some of the challenges faced by those doing quantitative research in New Zealand schools, some of our emerging areas of excellence, and a few case studies.
20 June 2006, Ian Westbrooke
Speaker: Ian Westbrooke, Research Development and Improvement Division, Department of Conservation, Christchurch
Title: Meeting statistical needs in a conservation management organisation
Time: 6:00pm, Tuesday 20 June 2006
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton Quay
Abstract:
DOC is responsible for all New Zealand's conservation estate and
for conserving native ecosystems and biodiversity, including
threatened species. To carry out such diverse management roles
effectively, scientific input is essential. We have around 60
national science staff, plus a number working as technical staff
in the 13 conservancies and on other projects. I joined in 2000 in
the newly-created position of statistician.
I plan to describe the range of scientific work at DOC and show
some examples from land, marine and social areas. I will outline what
I see as key statistical needs at DOC, how some of those needs are
currently being met, and how those with statistical skills can assist
conserving NZ's native biodiversity. One task is encouraging and
empowering staff throughout DOC to analyse the vast piles of
unanalysed monitoring and other observational data. Encouraging the
use of Excel and statistical packages to explore data and pose
management questions can lead to significant gains. For more formal
studies, design and analysis issues raise a huge range of statistical
questions.
16 March 2006, Geoff Chambers
Speaker: Geoff Chambers, Cell and Molecular Biosciences, Victoria University of Wellington
Title: Out of Taiwan? Genetics sheds new light on Maori origins
Time: 6:00pm, Thursday 16 March 2006
Place: Lecture Theatre 1, Old Government Buildings, 15 Lambton Quay
Abstract:
The migration into and settlement of remote Oceania by the modern
Polynesian and Maori peoples is a significant achievement, which has
been the subject of extensive research and scholarship. The routes
taken by the Austronesian peoples and the origins of the migrations
are subject to ongoing debate. We present a range of findings from
molecular genetic studies of Polynesian and New Zealand Maori
populations and a synthetic total evidence theory that we suggest can
account for key elements of the migrations.
Note: This talk is based on a paper presented at the International
Symposium on The Dispersal of Austronesians and the Ethnogenesis of
the People in the Indonesian Archipelago (Chambers, 2005); published
in NZ Science Review 63: 75–80.
13 December 2005, Rod Lea
Speaker: Rod A Lea, Senior Scientist, ESR
Title: Genome Informatics and Disease Susceptibility
Time: 6.00pm, Tuesday 13 December 2005
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton Quay
Abstract:
The advances in biotechnology over the past decade have resulted in the
production of an enormous quantity of molecular data. Currently there are
major research efforts worldwide to develop statistical computing methods
capable of extracting meaning from large, complex molecular genetic datasets
(genome informatics). This seminar will overview the field and provide some
examples of how statistics are used to understand the genetics of complex
disease in humans.
8 November 2005, Tony Vignaux
Speaker: Tony Vignaux, Emeritus Professor of Operations Research, VUW
Title: Incorporating factors such as Chance and Risk
Time: 6.00pm, Tuesday 8 November 2005 (light refreshments from 5.30pm)
Place: Lecture Theatre 2, Old Government Buildings, 15 Lambton
Quay
Abstract:
A description of personal encounters with decision situations
involving chance and risk. This will include practical examples
of some of: sensitivity analysis, queue theory, simulation,
game theory, stock control, investment analysis, decision analysis,
dimensional analysis, and genetic algorithms.
Deterministic OR and straightforward statistics, including Bayesian
methods, will be ignored in this talk.
14 September 2005, Paul Jose
Speaker: Paul Jose, School of Psychology, VUW
Title: Making Moderation and Mediation Quick, Easy, and Clear
Time: 6.00pm, Wednesday 14 September 2005 (light refreshments from 5.30pm)
Place: Lecture Theatre 2, Old Government Buildings, 15 Lambton
Quay
Abstract:
At present there is considerable confusion about how and when to do
moderation and mediation. Statistics textbooks discuss moderation and
mediation as special cases of ANOVA and multiple regression
respectively, but these two techniques are rarely discussed in the same
context in these books. In my research in developmental, social, and
clinical psychology I often use both methods, sometimes on the same
datasets. In my teaching (research methods in psychology) I promote the
use of these two techniques. My overwhelming response from students and
academicians is one of interested ignorance. One staff member asked me
"what button do I push in SPSS to make moderation happen?". My second
insight about this area is that current statistical packages (like SPSS)
do not perform either moderation or mediation easily and cleanly.
This situation has led me to create two Excel macros that permit a user
to take output from a stat package like SPSS and within minutes generate
publication-ready figures. Recently we have rewritten these programmes
in HTML and have made them available free of charge on my home page.
Plans are afoot for a stand-alone programme that will conduct both of
these techniques from raw data.
My presentation will be a basic discussion of the statistical techniques
of moderation and mediation as well as a demonstration of my two
programmes, ModGraph and MedGraph. Interested parties may wish to view
my home page before the lecture to gain a deeper understanding of these
programmes and techniques: http://www.vuw.ac.nz/psyc/staff/paul-jose/index.aspx.
2 August 2005, Phil Lester
Speaker: Phil Lester, School of Biological Sciences, VUW
Title: Using discriminant analysis to predict the establishment
success of exotic ant species in New Zealand
Time: 6.00pm, Tuesday 2 August 2005 (light refreshments from 5.30pm)
Place: Lecture Theatre 2, Old Government Buildings, 15 Lambton
Quay
Abstract:
Biological invasions can dramatically alter ecosystems. An ability to predict
the establishment success for exotic species is important for biosecurity and
conservation purposes. Using discriminant analysis I examine the exotic New
Zealand ant fauna for characteristics that predict or determine an exotic
species' ability to establish. Quarantine records show interceptions of 66 ant
species: 17 of which have established, 43 have failed to establish, whereas
nests of another six are periodically observed but have failed to establish
permanently (called 'ephemeral' establishment). Mean temperature at the highest
latitude and interception variables were the only factors significantly
different between established, failed or ephemeral groups. Aspects of life
history, such as competitive behaviour and morphology, were not different
between groups. However, in a stepwise discriminant analysis, small size was a
key factor influencing establishment success. Interception rate and climate were
also secondarily important. The resulting classification table predicted
establishment success with 71% accuracy. Because not all exotic species are
represented in quarantine records, a further discriminant model is described
without interception data. Though with less accuracy (65%) than the full model,
it still correctly predicted the success or failure of four species not used in
the previous analysis. Despite the low estimated accuracy, Malahanobis distances
from this model were used to successfully predict the status of five ant species
not used in the above analysis. Predicting which species will establish in a new
area appears an achievable goal, which will be a valuable tool for conservation
biology.
19 May 2005, Statistical Education
Speakers: Four 10-minute talks, as outlined below, followed by up to
45 minutes of discussion among presenters and the audience
Title: Into uncharted waters: school statistical education from 2005 onwards
Time: 6.00pm, Thursday 19 May 2005 (light refreshments from 5.30pm)
Place: Lecture Theatre 2, Old Government Buildings, 15 Lambton
Quay
Abstract:
Alex Neill (NZCER) will introduce and chair the meeting, as current
convenor of the Education Committee of the NZ Statistical Association.
The session has been devised collaboratively by the Education Committee,
NZSA and the Wellington Maths Association (WMA) for two sets of people:
Wellington mathematics teachers and Wellington statisticians.
The purpose is to update both groups on recent and impending changes
to statistics in schools. There are major possibilities and risks
for statistics within school mathematics. We hope to create
discussion and plans about how the two sets of people can work
together, to help ensure that the future for statistical education is
useful and exhilarating for teachers, students and the statisticians
who assist them.
There will be four 10 minute presentations:
Mike Camden, Statistics New Zealand: The status of the mathematics and
statistics part of the current revitalisation of the school curriculum,
and the potential for a paradigm shift in mathematics education.
Alasdair Noble, Massey University: The current review of the
one-year-old NCEA Level 3 and Scholarship Standards.
Terry McAuliffe, Tawa College: Statistics in NCEA Mathematics at Levels
1, 2 and 3: victories and challenges.
David Vere-Jones, SRA Ltd and VUW: Changes afoot in the UK and
Australia.
The discussion will be led off by Megan Clark (VUW) and Robin Averill
(WCE), and will focus on the question:
How can the statistical community work with teachers, to make their
task lighter and more enjoyable, now and in the future?
13 April 2005, Mark Weatherall
Speaker: Mark Weatherall, Wellington School of Medicine and Health Sciences;
Joint work with R.M. Pickering and S. Harris, Health Care Research
Unit, University of Southampton, UK
Title: Graphical sensitivity analysis with different methods of imputation
for a trial with probable non-ignorable missing data
Time: 6.15pm, Wednesday 13 April 2005 (light refreshments from 5.45pm)
Place: Lecture Theatre 4, First Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
Graphical sensitivity analyses have recently been recommended for
clinical trials with non-ignorable missing binary outcome. We
demonstrate an adaptation of this methodology for a continuous
outcome of a trial of three cognitive-behavioural therapies for
mild depression in primary care, in which one arm had unexpectedly
high levels of missing data. Fixed value and multiple imputation from
a normal distribution (assuming either varying mean and fixed SD, or
fixed mean and varying SD) were used to obtain contour plots of the
contrast estimates with their P values superimposed; their confidence
interval; and the root mean square error. Imputation was based on
both the outcome value alone, or on change from baseline. The plots
showed fixed value imputation to be more sensitive than imputing from
a normal distribution, but the normally distributed imputations were
subject to sampling noise. The contours of the sensitivity plots were
close to linear in appearance with the slope approximately equal to
the ratio of the proportions of subjects with missing data in each
trial arm.
24 February 2005, Tim Ball
Speaker: Tim Ball, Statistical Consulting for Continuous Improvement
Title: Golden Opportunities - A Case Study
Time: 6.15pm, Thursday 24 February 2005 (light refreshments from 5.45pm)
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
As a commercial statistical consultant, I view my most important
role as one of education - helping my client long-term improve
their operations, rather than just doing a "good" statistical study
for the immediate project on hand.
I will illustrate this talk with examples from my work with a company
that produces gold ore reference materials - covering statistical and
practical problems and aspects of the work done with the client
during our, so far, 7+ years of cooperation.
The talk will be high in pictorial content, and low in mathematics.
It will also contain some interesting facts about the production and
measurement of gold. I hope to get over the thrill I get, as a
consultant, in discovering new things about fields in which I have
only layman knowledge.
7 December 2004, Estate Khmaladze
Speaker: Estate Khmaladze, Victoria University of Wellington
Title: Probability and Statistics in Tbilisi, Georgia in retrospect:
1963-1983
Time: 6pm, Tuesday 7 December 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
Georgian Mathematics, from the very beginning of the 20th century,
was strong - one of the strongest schools within the Soviet Union,
at any rate. But it was mostly mechanics, elasticity theory,
mathematical physics and complex variables, functional analysis and
differential equations, and also strong topology and algebra.
Probability theory and statistics took root in Georgia much later.
One can start with the famous Tbilisi Conference of 1963, or
earlier - from the mid 1930s, with the course of lectures of
Razmadze and frequent visits of Kolmogorov and P.S. Alexandrov. How
these sciences developed since then in Georgia, and what was the
level and forms of statistical life - that is what the talk will try
to present. We will look also on scientific and personal connections,
and frequent meetings with Kolmogorov, or Smirnov, or - from a younger
generation - with Prohorov or Bolshev, Skorohod or Shiryaev, Ibragimov
or Gnedenko.
One aspect that is always so interesting is the human aspect: this
huge amount of great, sparkling talents of individuals, which often is
somehow hidden and passes almost unnoticed by an outside observer, but
which has to be there and is so inspiring and beautiful. I will speak
about a few individuals as well. We will dwell a little bit upon how
the interest in applications, this most important intrinsic feature of
any good research in our science, manifested itself on Georgian ground,
but had only little prospect in Soviet times.
All this will be personal recollections of the speaker, presented,
however, in the hope that there are some common features, or some very
specific features, which will be interesting to look at for others.
18 November 2004, Robin Willink
Speaker: Robin Willink, (Applied Mathematics) Industrial Research Ltd
Title: Statistical methods in metrology (measurement science)
Time: 6pm, Thursday 18 November 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 2, Old Government Buildings, 15 Lambton
Quay
Abstract:
Scientists at the Measurement Standards Laboratory of New Zealand
(MSL), which forms part of Industrial Research Ltd., are
responsible for maintaining New Zealand's standard units (such as
the kilogram, second and metre) to ensure `traceability' to the
standards of other countries. They also carry out research in
measurement science and conduct commercial calibration work.
This talk will summarize research that I have carried out over
the last five years through links with scientists at MSL. Issues I
will discuss will include:
the calculation of a reference value for an artefact measured
in a comparison involving the corresponding laboratories in many
countries,
the `degree-of-equivalence' summarizing the difference between
the results of any two laboratories,
the way in which the metrology community are encouraged to
evaluate and express the potential error in any measurement estimate.
Much can, and probably will, be said about the need for practical
methods, the room for clear communication to metrologists (who are
predominantly physicists), and the interpretation of probability.
The issue of the best way to combine systematic and random errors
is still alive - although these terms, and even the term 'error',
are out of favour. Academic statisticians might get a shock!
30 September 2004, Robert Davies
Speaker: Robert Davies, Statistics Research Associates Limited
Title: Crash Risk and Road Surface Characteristics
Time: 6pm, Thursday 30 September 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
New Zealand's state highways are surveyed annually and surface
characteristics such as roughness, skid resistance and rutting are
recorded at either 10 or 20 metre intervals. Curvature and gradient
are also recorded. The analysis described in this presentation
relates the road vehicle crash-rate to the road surface and geometry
data. The eventual aim is to find cost effective measures for
reducing the crash-rate through highway improvements. The main
analysis method is a variant of Poisson regression. Particular
problems that need to be handled by the analysis include the volume
of data involved, the errors in crash location, the questionable
nature of some of the data and the inevitable incompleteness of
the data. Nevertheless the analysis suggests that useful and
credible results can be obtained.
27 May 2004, Jeff Robinson
Speaker: Jeff Robinson, General Motors R&D Center, Michigan USA
Title: Estimating Mean Functions from Warranty Data with Dual
Useage Measures
Time: 6pm, Thursday 27 May 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 2, Old Government Buildings, 15 Lambton Quay
Abstract:
This talk summarizes some practical estimation problems that occur
in the analysis of typical automotive warranty data. Warranty
coverage and the resulting claims data are limited by mileage as
well as age. Age is known for all sold vehicles all the time, but
mileage is only observed for a vehicle with a claim and only at
the time of the claim. So an unknown number of vehicles leave
coverage due to the mileage constraint, and the number of vehicles
eligible to generate a claim at any age-mileage combination is
subject to uncertainty. The latter problem makes a full bivariate
treatment of this problem challenging. Here we concentrate on
univariate solutions (for both age and mileage) that account for the
mileage truncation. We take a nonparametric approach, so the methods
are extensions of the usual calculations for the mean cumulative
number of claims or cost of claims and its standard error
(Nelson, 2003). We extend slightly applications suggested by Hu &
Lawless (1996 JASA) for dealing with mileage withdrawals, and
simultaneously address the problem of claim reporting delay first
discussed by Kalbfleisch, Lawless and Robinson (1991). Real
automotive warranty data examples illustrate the methods.
29 April 2004, James Liu
Speaker: James Liu, Victoria University of Wellington
Title: Dynamics of interpersonal political environment and party
identification: longitudinal studies of voting in Japan and NZ
Time: 6pm, Thursday 29 April 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
Panel data over the course of a year examined the longitudinal
dynamics between social networks, social identifications, and
voting behavior among a general sample of registered voters in Wellington New Zealand and a national sample in Japan.
Stability in the interpersonal political environment (or the
political preferences of a person's social network) was
predicted independently by stability of party identification
(or social identity with political party) and lower education
in Japan; whereas stability in party identification was
predicted by stability of interpersonal political environment
and age in both countries. Stability of party identification
predicted voting consistency in both countries, whereas
stability of interpersonal political environment made an
independent contribution to voting consistency in Japan only.
There were cultural differences in levels of interpersonal
political environment stability. Results provided support for
the dynamical systems theory of groups, which claims that
interpersonal political environment and party identification are
dynamically interrelated to provide heuristics under uncertainty.
Key words: dynamics of voting, interpersonal environment, party
identification, Japan, New Zealand.
25 March 2004, Shirley Pledger
Speaker: Shirley Pledger, Victoria University of Wellington
Title: Using finite mixtures to model heterogeneity in
capture-recapture models
Time: 6pm, Thursday 25 March 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Ground Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
Capture-recapture models for both open and closed populations are
bedevilled by estimator bias caused by assuming individuals are
homogeneous in their survival and capture probabilities. Although
all models are wrong, some which use finite mixtures to allow for
individual heterogeneity are providing worthwhile bias reduction.
We pretend the animals come from finitely many classes with
unknown membership lists, and assume homogeneity within classes.
If heterogeneity is present, this provides better fitting models,
less biased estimates of parameters (e.g. abundance) and better
confidence interval coverage than the assumption of overall
homogeneity, but at the price of less precision.
Some models and results from both closed and open populations
will be discussed.
11 February 2004, Brian Easton
Speaker: Brian Easton
Title: The Estimation, Use and Misuse of Household Equivalence
Scales
Time: 6pm, Wednesday 11 February 2004 (light refreshments from 5.30pm)
Place: Lecture Theatre 4, First Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
Household Equivalence Scales are used to adjust the incomes of
households of different composition to equivalent (material) standards
of living. The simplest is to adjust them on a per capita basis, but this
does not allow for either economies of scale or that children have
different requirements to adults.
Brian Easton created his first scale thirty years ago. Last year, as a
part of their report on Health and Household Economy Project, he
and Suzie Ballantyne included a chapter which is the most
comprehensive review of the work done in New Zealand. It showed
that the scales in New Zealand have little scientific validity, and that
different scales have quite different outcomes for practical matters
such as to the level and the composition of poverty.
The presentation will be a tribute to Claudio Michelini, of Massey
University, who died unexpectedly in 2000, while working on the
most scientific household equivalence scales we have available.
See Also: Household Equivalence Scales
http://www.eastonbh.ac.nz/modules.php?name=News&file=article&sid=460,
Validation and the Health and Household Economy Project
http://www.eastonbh.ac.nz/modules.php?name=News&file=article&sid=229,
and Claudio Michelini
http://www.eastonbh.ac.nz/modules.php?name=News&file=article&sid=228.
26 November 2003, Stefanka Chukova and Yu Hayakawa
Speaker: Stefanka Chukova and Yu Hayakawa, Victoria University of
Wellington
Title: Warranty analysis: An overview and some new probabilistic
models
Time: 6pm, Wednesday 26 November 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 4, First Floor, Old Government Buildings,
15 Lambton Quay
Abstract:
In today's market, product warranty plays an increasingly important
role in both customer and commercial transactions. The use of
warranties is widespread and they serve many purposes, including
protection for manufacturers, sellers, and buyers. They are also
signals of quality and elements of marketing strategy.
In this talk, firstly we will focus on a detailed taxonomy for
mathematical models of one-dimensional warranty, introduce the
idea of degree of repair, and discuss the extension to two-
dimensional warranty and related problems. Secondly, we will
present a specific warranty model. The cost of each warranty claim
depends on the non-zero length of the repair time. Two types of
free replacement warranty policies are considered: non-renewing
and renewing. An alternating renewal process is employed to model
the operating and repair times. New results for alternating renewal
processes in a finite horizon are derived. These results are used to
evaluate the warranty costs over the warranty period and over the
life cycle of the product under both types of policies.
28 October 2003, Nick Longford
Speaker: Nick Longford, De Montfort University, Leicester, UK
Title: Stability of household income in European countries in the
1990s and a NZ connection
Time: 6pm, Tuesday 28 October 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton
Quay
Abstract:
The talk explores the patterns of change in the annual household
income in the countries of the European Community during the
years 1994–1999. The income is modelled by mixtures of log-normal
distributions, and the mixture components are interpreted as
representing one sub-population with steady increments and others
with various levels of volatility. The method is extended to models
for a combination of log-normal and categorical variables. An index
of income stability is defined for the countries. Graphical summaries
of the results are emphasized.
Current and planned research in another application of mixture models
is outlined and some results about the local-area house-price inflation
in New Zealand are presented.
Coauthors: M.G. Pittau (University of Rome 'la Sapienza', Rome, Italy)
14 August 2003, Caroline Roughneen
Speaker: Caroline Roughneen, Trinity College, Dublin
Title: Study of Engineering as a Career Choice
Time: 6pm, Thursday 14 August 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 4, Old Government Buildings, 15 Lambton
Quay
Abstract:
Recent research shows that engineering as a career choice is in
decline in Ireland. This talk will examine the determinants that
attract female students, in particular, into engineering studies at
tertiary level. It will also identify obstacles to female students
applying for and enrolling in engineering courses and suggest
practical measures that would lead to a more equitable balance
between men and women.
The methodology includes two national surveys in Ireland and
one national survey in New Zealand. These surveys are targeted
at Careers Advice teachers in Ireland and final year students in
both countries. The study will establish the availability and
usefulness of engineering career information provided by
engineering firms, tertiary level institutes and the engineering
professional body. The information provided by this study will
be used to develop best practices to promote engineering courses
at tertiary level and identify measures or marketing strategies that
would best interest female students in careers in engineering.
9 July 2003, Mark Weatherall
Speaker: Mark Weatherall, Wellington School of Medicine and Health Sciences
Title: Prevention of falls and fall related fractures in community dwelling
older adults: A meta-analysis of estimates of effectiveness based on
recent guidelines
Time: 6pm, Wednesday 9 July 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Rutherford House, 23 Lambton Quay
Abstract:
Background: Two recent falls prevention guidelines have been
published but did not include quantitative estimates of effectiveness
based on the literature that was reviewed to support their
recommendations.
Aims: To produce quantitative estimates of effectiveness of falls
prevention programmes from the randomised controlled trials cited
in the guidelines together with an updated literature search to
August 2002.
Methods: Meta-analysis of randomised controlled trials cited in
falls guidelines and studies identified by an updated literature search.
Randomised controlled trials were identified from the falls
guidelines and a literature search which met the following criteria:
Trials in community dwelling older people, one year follow up, and
outcome measures reported as number of subjects with at least one fall
or number of subjects with a fracture.
Results: The guidelines identified four studies of 'Exercise as a sole
intervention' which when combined with one further study identified
on the literature search gave a fixed effects odds ratio favouring this
strategy of 0.81 (95% CI 0.58 to 1.14), number needed to treat to
prevent one person having a fall 19.5. The guidelines identified seven
studies of a 'Multiple intervention' strategy that gave a random effects
odds ratio favouring this strategy of 0.64 (95% CI 0.47 to 0.88). Four
further studies were identified by the literature search. The updated
odds ratio favouring this intervention strategy was 0.65
(95% CI 0.52 to 0.81); number needed to treat was 9.8. Only two
studies had data for fracture and a fixed effect odds ratio favouring
falls interventions for fracture prevention was 0.50
(95% CI 0.18 to 1.4), number needed to treat to prevent one person
having a fracture 45.5.
Conclusion: Semi-quantitative statements of evidence can both
understate and overstate the effectiveness of falls prevention strategies.
There is moderate evidence of efficacy for fall prevention particularly
for multiple intervention strategies.
11 June 2003, Leigh Bull
Speaker: Leigh Bull, DOC
Title: Sizing up shearwaters: morphological variation in the genus Puffinus
Time: 6pm, Wednesday 11 June 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Rutherford House, 23 Lambton Quay
Abstract:
The shearwater genus Puffinus was used to investigate how the
interactions between an organism and its environment are manifested
in its morphology. This study represents one of the few comprehensive
analyses of size variation in seabirds, and the first for Puffinus. Three
levels of morphological variation were investigated: interspecific
differences between species, sexual size dimorphism between the sexes,
and intraspecific geographic variation over a species range.
Multivariate statistical techniques such as canonical discriminant
analysis and MANOVAs were used to quantify the patterns of
morphology in the shearwater genus Puffinus. Differences in phenotypic
variation among traits and between sexes, as measured by the coefficient
of variation, were used to investigate the possible mode and intensity of
selection operating on each trait and sex respectively. Correlations were
calculated to investigate the possible relationships between the
morphology and ecological variables such as food, insularity and climate.
Food, through its influence on competition, dispersal, growth, fecundity
and survival, appears to play an important role in the relationship between
the morphology and ecology of Puffinus individuals.
1 May 2003, Chris Francis
Speaker: Chris Francis, NIWA (coauthor: Bryan Manly, WEST Inc, Cheyenne, Wyoming)
Title: Simultaneous testing for mean and variance differences with nasty data
Time: 6pm, Thursday 1 May 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 101, Murphy building, Kelburn Parade
Abstract:
Possibly the most commonly performed type of statistical test is for
differences in means. It is often advisable to precede such a test by
another test for differences in variances because the mean-tests
commonly perform poorly in the presence of such differences.
We describe an approach, based on randomization, that tests for both
these differences, and evaluate its performance via a simulation
experiment.
The approach is shown to be useful as a robust, conservative method in
cases when the samples come from very non-normal distributions.
One possible outcome of the test is the conclusion that there are mean
and/or variance differences, but it is not possible to say which.
27 March 2003, Srinivas Chakravarthy
Speaker: Srinivas Chakravarthy, Professor of Operations Research and Statistics, Kettering University, Flint, Michigan, USA
Title: Impact of worker cross-training in service systems
Time: 6pm, Thursday 27 March 2003 (light refreshments from 5.30pm)
Place: Lecture Theatre 101, Murphy building, Kelburn Parade
Abstract:
We consider a service system with two types of customers. In such an
environment the servers could either be specialists who can serve a
specific customer type or generalists who can serve either type of
customers. Although cross-trained workers are more flexible and help
reduce system delay, they also contribute to the increased costs and
reduced service efficiency. The objective of this paper is to provide
insights into the choice of a right workforce mix of flexible and
dedicated
servers that minimizes the system related costs. Matrix-analytic methods
are used in this study.
17 October 2002, Frances Krsinich and Mike Camden
Speakers: Frances Krsinich and Mike Camden, Statistics New Zealand
Title: The use of noise to confidentialise tables
Time: 6pm, Thursday 17 October 2002 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Rutherford House, 23 Lambton Quay
Abstract:
SNZ confidentialises tables of magnitude data (e.g. total turnover for
industry by region) by first identifying sensitive cells for suppression
using a dominance rule and then performing secondary cell suppressions
to prevent derivation of the sensitive cells from marginal totals.
Although this is the approach adopted by most statistical agencies, it
has some significant limitations - secondary cell suppression is
time-consuming and results in the suppression of non-sensitive data.
The US Census Bureau recently proposed a new approach - unbiased
perturbation of the sampling weights such that 'noise' in the resultant
confidentialised table is targetted towards the sensitive cells. This
approach is much simpler to apply than cell suppression and, arguably,
results in less overall information loss. We have been researching the
application of this method to the Annual Enterprise Survey, and hope to
introduce it in 2003.
In the course of this work, we have identified a possible generalisation
of the method to tables of count data, such as those produced from the
Population Census. For some customised outputs, such as bulk outputting
of many related tables of census data, this may be a more appropriate
method to use than random rounding to base 3, the current method for
tables of counts from Census.
In this talk, we will outline the reasons for confidentialising data,
and list some existing methods. We will detail the version of the noise
method that SNZ hopes to implement for tables of magnitude data, then
discuss how a noise method for tables of count data could work, and what
we would need to do to assess it.
27 August 2002, Paul Dyer
Speaker: Paul Dyer, Head of Investment Strategy, AMP Henderson,
Wellington
Title: Are financial markets predictable? Some implications for
investors
Time: 6pm, Tuesday 27 August 2002 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton Quay
24 July 2002, Jean Thompson
Speaker: Jean Thompson, JAD Associates, Wellington
Title: 'Real world' statistical data and the role of the consultant
statistician
Time: 6pm, Wednesday 24 July 2002 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton Quay
Abstract:
At the beginning of 1962 when I joined the Applied Maths Laboratory of
the DSIR in Wellington, the statistical work I got involved in
substantially comprised helping various agricultural, biological and
ecological scientists analyse their data. I was to learn quickly that
unless the statistician had a part in designing the data collection
system, even exhaustive and highly-sophisticated analysis could often
yield little new information. Sadly, in essence, not much has changed
in the intervening 40 years.
There is still a perception in many quarters that the statistician is
the "necessary evil" to get the data analysed and frequently they are
not consulted until long after an investigation is well under-way. This
is a great shame, for while huge advances have been made in the methods
we can use to analyse data and the facility with which it can be done,
there is still the vexatious issue of data quality. In the ideal world
one would be part of the whole process but it is more
usual for a consultant to be brought in late in the day often after much
damage has been done. However, we must be careful not to necessarily
blame the investigator for hard-to-handle data. Some areas of
application simply do not produce nice clean data no matter how one
tries, even with good input from a statistician.
It was a good 20 years ago that the Applied Maths Division of the DSIR
started commercial consulting and thereby opened the floodgates for
highly variable data. After its demise in 1992 I continued as a private
consultant doing the same sort of things I had been doing for my public
sector employers, helping researchers and people in business understand
the processes they cared about.
In this talk I will summarize some of the situations I have encountered
over the last 5 years and focus on what I see as important issues for
applied statisticians and their teachers.
The philosophy and material I will present was given in a one-hour
seminar at the University of Canterbury in April 2001 and also at the
NZSA Conference in Christchurch in December 2001.
A panel discussion will follow Jean's talk, starting at approximately
6.30pm. Jean will be joined on the panel by Robert Davies and Tim Ball,
who are both consulting statisticians with considerable experience. The
panel will respond to questions from the audience, with the intention of
addressing further 'real world' statistical experiences.
21 February 2002, Richard Arnold
Speaker: Richard Arnold, Victoria University of Wellington
(joint research with Tony Vignaux and Denis Sullivan)
Title: Bayesian spectral analysis of white dwarf light curves
Time: 6pm, Thursday 21 February 2002 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton Quay
Abstract:
White Dwarfs are stars near the end of their lives which may pulsate
with periods of a few minutes, and which can be observed to brighten and
dim as they pulsate. In this paper a Bayesian approach due to Bretthorst
(1988) is applied to the analysis of the light curves of white dwarfs.
The method yields estimates of the dominant pulsation frequencies which
are highly accurate and superior to estimates obtained from classical
Fourier techniques. Moreover, the method can be applied to datasets
where there are missing values, where the sampling is not uniform in
time, and in particular allows the combination of observations from
different observing runs. In this paper we analyse observations of a
particular white dwarf on 8 separate occasions over a two year period.
6 December 2001, Brian Bull
Speaker: Brian Bull, National Institute of Water and Atmospheric Research (joint research with Dave Gilbert)
Title: Catch-at-age data from New Zealand fisheries
Time: 6pm, Thursday 6 December 2001 (light refreshments from 5.30pm)
Place: Lecture Theatre 3, Old Government Buildings, 15 Lambton Quay
Abstract:
Catch-at-age analysis estimates the age and size distributions of fish
taken by a commercial fishery, based on samples of fish from individual
catches. The resulting age frequencies are key inputs to the stock
assessment process, which assesses the sustainability of the fishery.
Catch-at-age analysis is undertaken for most major New Zealand finfish
stocks, including hoki, orange roughy, and snapper.
I will talk about the sampling programme and the quirks of the resulting
data. The analysis is statistically straightforward but involves several
interesting issues, including the choice of stratifying variables, the
bootstrapping technique used to calculate the variability of the
estimate, and the two competing methods used to convert length
distributions to age distributions.
16 October 2001, Dean Hyslop and Dave Mare
Speaker: Dean Hyslop and Dave Mare, The Treasury
Title: Understanding Changes in the Distribution of Household Incomes in New
Zealand Between 1983–86 and 1995–98
Time: 16 October 2001
Abstract:
This paper presents an analysis of changes in the distribution of gross
household income and income inequality over the period 1983–98. The
analysis applies a semiparametric approach to study the effects of
changes in the distribution of household types, and changes in National
Superannuation (old age pension), household socio-demographic attributes
and employment outcomes, and in the "economic returns" to such
attributes and employment outcomes on the distribution of income, and
uses kernel density methods to
estimate these effects. This approach provides a visual appreciation of
the shape of the income distribution, and is important in understanding
how each of these factors affected different parts of the distribution
over the period. We also estimate the effects of each of these factors
on changes in various summary measures of inequality over the period.
The results find that changes in household structure (particularly the
declining proportion of two-parent families), attributes, and employment
outcomes each contribute to the observed
increase in inequality, while the changes in returns are estimated to
reduce the level of inequality. Collectively these factors account for
about 50 percent of the observed increase, depending on the measure of
inequalit used. The results confirm other research findings that the
changes were concentrated during the late 1980s.
This is a Treasury working paper 2001/21 and is available at
www.treasury.govt.nz/workingpapers/2001.
8 August 2001, Brian Pink
Speaker: Brian Pink, Government Statistician, Statistics New Zealand
Title: The Impact of IT on Official Statistics
Time: 5:30pm, Wednesday 8 August 2001 (refreshments from 5pm)
Place: 1st Floor Meeting Room, Science House, Royal Society of NZ, Turnbull St
Return to
Local Groups Page
