640 likes | 654 Views
Explore the varied applications and innovations in statistical science, from genetic studies to business analytics. Discover the importance of statistical modelling in shaping public policy and improving evidence-based decision-making processes. Unveil the key role of technology in transforming data analysis and inference. Join us in understanding the complex systems through modular structures and global models for impactful results.
E N D
Diversities of gifts, but the same spirit Peter Green RSS Presidential address 18 June 2003
A discipline of diversity Public life Social science Science Technology Medicine Statistics Business and industry
philosophical foundations mathematical theory inferential principles design data collection techniques computation modelling A discipline of diversity
A discipline of diversity Interaction with the rest of the world is part of the subject itself
shelter and nourishment for statistics • a microcosm of
Connection or fragmentation? What holds us together?
Connection or fragmentation? Statistics in Society • getting the correct denominator in workforce statistics • computing DNA match probabilities • assessing clinical effectiveness • evaluating GM crop experiments
Connection or fragmentation? Heterogeneity of discipline • intellectual strength • structural weakness
medical industrial core social official
medical industrial core social official
medical industrial core social official
How the discipline develops • Promoting our strengths should be a key priority for the discipline and for the Society
How the discipline develops – demands of applications • Public policy • evidence-based decision making • performance measurement • Legal system • scientific evaluation of evidence • Social science • respect for quantification • public archives, National Statistics
How the discipline develops – demands of applications • Business • data-mining • Technology • uncertainty in telecomms, images • Science • all scales: Astronomy to Genomics • quantum level?
How the discipline develops – opportunity of technology • Sensors and instrumentation • Data-logging capacity • Communications • Number-crunching • transforming quantity and quality of data • enabling highly computer-intensive analysis
How the discipline develops – theoretical innovation • Relaxation of old philosophical quarrels • Rehabilitation of Bayesian methods • Key role of conditional inference • graphical modelling • Stochastic calculus • martingales • Point processes
The role of statistical modelling • underpinning all parts of the discipline • the most basic tabulation or summary involves conceptualisation • what can vary? • on what scale? • depending on what?
The role of statistical modelling • Discipline in creation of methodology • Framework • for study of foundations • for expressing principles • for provision of computational tools • Use more to communicate ideas • & break down barriers between theory and practice?
Structured systems • A framework for building models, especially probabilistic models, for empirical data
Markov chains Spatial statistics Genetics Regression AI Statistical physics Sufficiency Covariance selection Contingency tables Graphical models
Structured systems Key idea - understand complex system through global model built from small pieces • comprehensible • modular • each with only a few variables
Modular structure Basis for • understanding the real system • capturing important characteristics statistically • defining appropriate methods • computation • inference and interpretation
Conditional independence • X and Z are conditionally independent given Y if, knowing Y, discovering Z tells you nothing more about X • X Z Y X Y Z
Conditional independence as seen in data…. Does survival depend on ante-natal care? .... what if you know the clinic?
Conditional independence survival ante clinic survivaland clinicaredependent andanteandclinicaredependent but survival and ante are CI given clinic
AB AO AO OO OO A natural directed graph from genetics A AB A O O Mendel
Model for lip cancer data regression coefficient covariate random spatial effects relative risks observed counts
or non- Bayesian
Bayesian structured modelling • ‘borrowing strength’ • automatically integrates out all sources of uncertainty • properly accounting for variability at all levels • including, in principle, uncertainty in model itself
Bayesian structured modelling • ‘borrowing strength’ • automatically integrates out all sources of uncertainty • … for example in forensic statistics with DNA probe data…..
Bayesian structured modelling • ‘borrowing strength’ • automatically integrates out all sources of uncertainty • … for example in modelling complex biomedical systems like ion channels…..
Ion channelmodel model indicator transition rates hidden state Hodgson and Green, Proc Roy Soc Lond A, 1999 binary signal levels & variances data
model indicator C1 C2 C3 O1 O2 transition rates hidden state binary signal levels & variances data * * * * * * * * * * *
Structured systems’ success stories include... • Genomics & bioinformatics • DNA & protein sequencing, gene mapping, evolutionary genetics • Spatial statistics • image analysis, geographical epidemiology • Temporal problems • longitudinal data, financial time series, signal processing
The methodology gap • Subgroups develop their own ideas and jargon • Weaker communication between than within • Little evidence in RSS journals • But wide use of outdated and inappropriate statistical techniques in some areas
The methodology gap - the pressures: • pace of working life specialisation quick approximations • training more focussed • both theoretical and applied
The methodology gap • RSS provides something for almost every specialism • but how many of us exploit that?
Making more of methodology • Relevance to applications is the main stimulus and justification • But, for the sake of the vigour of the subject and cross-fertilisation between applications, there is a vital role for ‘generic methodology’ • not mathematical statistics • not application-specific
Generic methodology • The generalised likelihood ratio test • Fisher scoring • The practice of fitting dose-response relationships by MLE all existed before….
Generic methodology …. but the generalised linear model framework did not just unify, • it generated new application-specific technique • it promoted good practice generally
A good methodology paper might cover all of ... • underlying philosophical principles • mathematical development • statistical modelling of a real process • computational implementation • data analysis • model criticism • interpretation of inference and performance