Euroworkshop on Statistical Modelling
Workshop Description
(Extract from the official Proposal)
Project Topic:
Statistical Modelling is required whenever information has to be drawn from Data. It is the core discipline of Data Mining and has become impressively powerful over the last couple of years. The numerical developments in the last decades allow now to analyse, investigate and explore large and complex Data situations by statistical models. The field of Statistical Modelling and Data Mining is of extreme dynamism and innovation. The Euroworkshop on Statistical Modelling with a series of three specially focused small workshops is supposed to do justice to the dynamic research in Statistical Modelling. It intends to provide an European forum where high level research is taking place.
Theoretical and technical developments in statistical modelling are thereby most frequently stimulated and provoked by Data coming from nearly all areas of science and industry. As a countermove to this, scientists in all areas profit from newly developed and adapted approaches in Statistical Modelling. Due to this symbiosis of demands and offers from other areas in science, Statistical Modelling can be seen to the highest degree as multidisciplinary science.
In Statistical Modelling two major streaming have been developed with convincing success in the last years. These are Mixed Models and Nonparametric Models. Beside these two directions, Model Diagnostics have become impressively popular. The EuroWorkshop on Statistical Modelling contributes to these three themes by following 4 modules:
Three small workshops (modules 1-3), where each of them will focus on one of the three topics mentioned, i.e. Mixed Models, Nonparametric Models and Model Diagnostic.
The presentations and results developed in and due to the workshops will be published in a book or a series of books which serves as module 4. The book series will give a state of the art report and promotes and stimulates further research.
Mixed Models (Event 1, Module 1):
In Mixed Models unobserved heterogeneity is modelled by random effects. This means that individual, latent effects are incorporated in a statistical model to mirror the reality. For instance in a drug study, each human being reacts differently to a drug; some are very sensitive, some others react only weakly to the active substance. Similarly, in the analysis of unemployment, individual effects have to be considered, since each unemployed individual react differently; some develop or have skills to return to professional life, some others suffer from the failure. In both of these examples Mixed Models can cope for the individual heterogeneity. The general area of Mixed Models is however even wider, e.g. models measurement error or latent classes also yield models with mixed structure.
Highly sophisticated methods have been developed and applied for Statistical Modelling with Mixed Models. Extensive numerical and theoretical developments in the recent years allow us to fit rather complex models (see e.g. Breslow & Clayton, 1993, Journal of the American Stat. Association; or Aitkin, 1999, Biometrics). However, not all barriers resulting from the complex modelling are crossed yet and future research is demanded. The general intention is to make Mixed Models as realistic as possible. This requires future research in numerical and statistical problems. Moreover comparison and assessment of available methods is required (see also module 3 below).
Nonparametric Models (Event 2, Module 2):
As in all areas of science, models serve as a portrayal of the reality. Their quality and usefulness is thereby heavily dependent on the complexity of the model itself. A simple model can only mirror simple things. Moreover, as in all quantitative sciences, classical quantitative models suffer from being parametric, i.e. the reality is mirrored by a number of parameters. Nonparametric Models go beyond this scope by modelling relations and effects nonparametrically, i.e. parametric functions are substituted by nonparametric flexible curves. As a consequence, Nonparametric Models do not require a prior defined relation structure, but allow us to mirror and explore complex and totally unknown structures.
Various approaches have been developed in the last years, some tracing back to mathematics (Spline fitting, see e.g. Green & Silverman, Nonparametric Regression and Generalized Linear Models, Chapman & Hall), some others come from extending statistical concepts (Bayesian Smoothing, Local Fitting, see e.g. Fan & Gijbels, 1996, Local Polynomial Fitting, Chapman & Hall) . Numerical and theoretical results allow a wide range of applications of Nonparametric Models. Further research is required to make more profit of this powerful modelling technique. Investigation of theoretical aspects and applicability of the available routines in more depth is demanded to assess and describe their impact. Finally, a fair comparison of the available routines is still rudimentary but desired.
Model Diagnostics (Event 3, Module 3):
Generally speaking, every model is plausible as long as it is not falsified. The falsification of models however is the most challenging point in science. This holds in the same way for statistical models. To provide valid Statistical Models that fulfil the task of modelling reality, it is of fundamental importance to check and asses the model by Model Diagnostics. This third module and topic is therefore seen as the logical and compelling conclusion of the above two topics, i.e. Mixed Models and Nonparametric Models. A number of recent results in the framework of Nonparametric Models demonstrate the blooming progress in this active research field (see e.g. Hart, 1997, Nonparametric Smoothing and Lack-of-Fit Tests, Springer-Verlag or Kauermann & Tutz, 1999, Biometrika). In Mixed Models however, tools and concepts for model diagnostics are rudimentary until now and research is heavily demanded. Resampling approaches appear as promising in this context. Successful developments of Model Diagnostic tools will allow us to draw valid inferential conclusions from Data. This allows the scientist in all fields to profit from Statistical Modelling and to refine or correct their substance matter models.
Book: Euroworkshop on Statistical Modelling (Module 4):
The book respectively the book series "Euroworkshop on Statistical Modelling" will report the results and developments presented at or due to the 3 workshops above. For each event, 5-8 contributions are planed to be published in the book series. The Scientific Committee and the Keynote-Speakers will ask participants to contribute to this project. Cooperation and joint work, in particular between Young Researches and senior scientists will be emphasised and promoted. The Scientific Committee with the help of the Keynote-Speakers will also include a résumé in the book series which is derived by all participants in the last session of each workshop (see below). This résumé summarises first results which were derived at the workshop and names the relevant scientific problems which were discussed.
Programme Content:
General Description:
As described above, the project Euroworkshop on Statistical Modelling consists of 4 modules, three small high level scientific workshops (events 1-3) and a book project as fourth project, which summarises the results derived in the three workshops.
All workshops are organised in a similar fashion which provides continuity. Changes are made where this is necessary. The workshops will take place in Schloß Höhenried, a Bavarian castle near Munich, over the first weekend in November 2000, 2001 and 2002. The dates and the length are chosen with care. First, November is an off peak season for conferences, i.e. experts are available, and secondly, the length allows for a totally focused treatment of the topic. The framework chosen for the workshop mirrors the opinion obtained in an a prior survey pursued by the Organisers.
Workshops 1-3 (events 1-3):
The 30 participants meet under examination conditions to work undisturbed in the considered topic. Each workshop consists of 6 sessions, each of them focussing on one particular subtopic. For each workshop 5 Keynote Speakers are invited and further renowned scientists are asked to participate and contribute. Young Researchers are particularly encouraged to present their material. The selected Keynote Speakers should scientifically stimulate the audience and provide the forum for extensive discussions. The Scientific Committee takes pride in choosing appropriate Keynote Speakers for the sessions, who are able to handle the challenge lying in their talks. All Keynote Speakers, but in particular the first one, are asked to provoke and discuss questions from the audience in order to generate an open and productive atmosphere. This in particular helpful and supportive for Young Researchers. All Keynote Speakers are asked to present their material on an understandable but scientifically exact level. The entire audience, in particular the Young Researchers with a more focussed knowledge, should be able to follow, comprehend and discuss the presented material.
In the first session an outstanding expert gives an overview about the state of the art in the topic treated by the workshop. In the following four sessions special aspects of the topic covered by the workshop are illuminated in more depth. The focus of these sessions is on a presentation and discussion of the cutting edge of research in the particular topic. The Keynote Speakers are asked by the Organisers to particularly emphasise the applicability of the presented methods. They should demonstrate the interdisciplinary flavour of statistical modelling.
Book series (module 4):
In the final session of the workshop the derived and presented results are summarised. Open questions and problems will be named and written down in a résumé. This résumé provides a "road map" for future research. In the last session the Organisers starts with the help of the Keynote-Speakers their editorial job for the book or book series "Euroworkshop on Statistical Modelling". The participants and the Keynote Speakers are asked to contribute, cooperation between the participants is demanded and strongly encouraged. This holds in particular for senior scientists and Young Researchers. The book will be published with care, the articles will circulate among the authors prior to printing in order to guarantee a unique notation and to allow for cross-references.
Organisation and Management:
The organisation of the Euroworkshop on Statistical Modelling is done by the Organisation Committee. This consists of:
Göran Kauermann, (Coordinator and Young Researcher)
Ludwig-Maximilians Universität München, D, (University of Glasgow, UK, after 09/00)
Herwig Friedl
Technische Universität Graz, A
John Hinde
University of Exeter, UK
Emmanuel Lesaffre
Katholieke Universiteit Leuven, B
The three workshops (event 1-3) take place in Schloß Höhenried near Munich, Germany. The conference location with its well equipped lecture rooms and the on-site accommodation will guarantee a productive working atmosphere.
All participants will be informed prior to the workshop about the particular shape of the workshop. The final programme, built by the Organisers and the Keynote-Speakers, will be made public in advance. All participant will have complete information about the programme, the list of participants and the contributed presentations (abstracts) prior to the workshop. This gives the possibility of personal preparation for the workshop. All progress made in the organisation will be made public on the webpage. This also holds after the workshop, so that the webpage and the mailing lists will serve as virtual discussion and information forum.
The Keynote Speakers will be informed in more depth about what the Organisers expect from them. They are also asked to actively participate in the scientific organisation, e.g. by suggesting or selecting the other speakers. A particular focus will always be on the encouragement of and discussion with Young Researchers.
The final session of the workshop gives the résumé of the meeting. Open problems and research demands will be formulated. The session also serves as an editorial meeting for the planed book series (module 4). The essence of the last session will give a contribution in the books. Further contributions are given by participants. Cooperation between Young researchers and senior scientists will be particularly promoted, i.e. senior scientists are asked to write a joint contribution with Young Researchers. The book series will be published after the third workshop.
Location and Infrastructure:
The workshops (event 1-3) take place in Schloß Höhenried, Bernried near Munich, Germany. The Schloß Höhenried is a Bavarian castle, located at the Starnberger See, the most famous outdoor recreational area of Munich. The location is more than suitable and rather perfect for the workshop type proposed.
The Conference Venue
The Schloß Höhenried belongs to the property of Klinik Höhenried, a famous and outstanding clinic for heart disease patients. The castle (Schloß) stands separately, surrounded by a beautiful park with direct access to the lake Starnberger See. The house is equipped with lecture and meeting rooms of different size, the largest fitting 50 people. The atmosphere in the castle is quiet and very convenient. It is stimulating and we expect that all participants will enjoy their stay and make use of the location. For instance the surrounding park invites walks which in turn yield a perfect and informal framework for intensive discussions between senior scientist and Young Researchers. Moreover, the interior of the castle offers enough space for smaller working groups. The lecture room is equipped with a large U-shape table. This easily allows everybody to participate and contribute to the discussions. We expect that a family like atmosphere will promptly develop.
The Infrastructure
The venue possesses 30 on site bedrooms. The rooms are of high standard, very clean and each one has a private modern bathroom. Room service contributes to the convenience, the staff are helpful and judicious. Breakfast is provided in the castle, lunch will be served in the nearby dining hall of the clinic. Dinner is available upon request, but the Bavarian restaurants in the nearby village Bernried provide an appealing alternative.
Transportation
Schloß Höhenried is most easily reached from Munich by commuter trains, which run every 20 minutes. The nearest train station to the castle is less than 5 minutes away by car (taxi costs about 10-15 Euro). Total travel time from Munich central station is less than 45 minutes.
Munich itself is perfectly connected by train and air links. Munich airport is the second largest in Germany and served by all major European and other international airlines. The airport is well connected to the commuter train net, so that Schloß Höhenried can be reach within 90 minutes from Munich airport.
Young Researchers:
The workshop will let high level science take place in Europe. Moreover, it is specially designed for Young Researchers and demonstrates, that Statistical Modelling is a challenging and fruitful research field. During the workshop Young Researchers meet with outstanding scientists which allows for enlightening discussions. The Young Researchers get into contact with the cutting edge of science. The general intention is to bring senior scientists and Young Researchers together in order to provoke cooperation.
The programme is designed so that participants can learn and internalise the theory and applications presented. The impact of the topic discussed at the workshop is understood and applied in statistical modelling. Young researchers can distribute their knowledge and make use of their gained lead, either in their own research or in industry as their future working field. The workshop serves as a contribution to make Europe more attractive for Young Researchers in Statistical Modelling and related fields.
The Euroworkshop on Statistical Modelling brings together Young Researchers and outstanding scientists. The three separate workshops offer a forum for active research. Selected Young Researchers with sound knowledge can profitably draw from the workshop in various ways.
First, the outstanding selection of high level scientists guarantees a first class programme where only the best and most useful results are presented. Secondly, the atmosphere resulting from the arrangements implies a forum for informal discussions of research results. The Keynote Speakers are asked to take their time and effort for discussions with Young Researchers. Since all workshop members stay at the castle, a friendly and family like atmosphere will develop soon. As a consequence, the discussion of research and cooperation will be able to take place in a rather informal way. In general, the Young Researchers get a smooth entry into the high level scientific community.
Finally, in the last session of each workshop, the Scientific Committee asks for several (5-8) book contributions from the participants. The Scientific Committee thereby strongly encourage senior scientists to cooperate with Young Researchers e.g. by writing a joint contribution. This contribution should mirror the state of the art of the discussed topic, but it is intended that the senior scientists and the Young researches continue their cooperation with active research.