Department of Statistical Modelling

Brief characterization of its research field

Statistics in its various forms, such as e.g. mathematical statistics, biostatistics, applied statistics, econometrics, psychometrics, or even bioinformatics and ‘data science’, represents one of the branches of information sciences which use data to draw inference about the reality. This is specifically achieved using the concept of designed experiments which relates to different fields of research, such as agriculture (e.g. R. A. Fischer’s experiments), biomedical sciences (providing ’big data’ from genomic studies), medicine and pharmacology (e.g. involves statistical methodology for analysing data from clinical trials), engineering, economy (econometrics), social sciences (incl. psychology, pedagogy), and many other fields of scientific research. Observational studies, whether cross-sectional or longitudinal, offer another source of information, the latter rendering time series data. Observational studies may also involve ‘big data’ (e.g. data from satellite worldwide weather monitoring). Consequently, statistical science spans the realm of both basic and applied interdisciplinary research.

Key research trends in the field at hand

Various topics of statistical science are being researched by the members of our department. They include the following: linear and generalised linear models for independent data, mixed variants of the two classes of models for correlated data (e.g. cross-sectional and epidemiological studies), generalised additive models (GAM) allowing for flexible covariate modelling, methods for dimensionality reduction (big data), methods for analysing censored data (longitudinal, cohort studies), missing data imputation, hierarchical regression models, robust methods designed to obtain correct inference in the presence of outliers, change-point problems in time series, etc. In recent years the volume and complexity of data increased dramatically.

Department Vission

Mathematical statistics is one of the building blocks that is reflected in data management, processing and extraction of useful information from the data while controlling the level of confidence in our conclusions that are often based on large volumes of complex and noisy inputs. The ‘modelling emphasis’ indicates our interest in developing advanced methods and algorithms which take advantage from the design phase of an experiment and lead to optimised decision support in various scientific disciplines mentioned above. The department puts a key emphasis on promoting statistical approaches to solving various problems of basic and interdisciplinary research where the inspiration for developing advanced statistical methodology and algorithms comes primarily from related field of research. Propositions arising from related scientific disciplines are reformulated as statistical hypotheses and new testing procedures are developed to answer the scientific queries.

Substantiation of the vision

The department will contribute to a further development of statistical theory and methodology, primarily in relation to the above mentioned scientific disciplines and principal research topics. Interdisciplinary collaboration with colleagues from social sciences will result in developing new methods for validation of knowledge and psychologic tests, sensitivity analysis for items and tests in relation to different target groups, adaptive testing and mixed regression modelling. New approaches will be developed for testing structural changes in time series and panel data. Department members will continue in biomedical research collaboration, involving e.g. analysis of congenital defects, research of methods and algorithms for the analysis of categorical data (exact and interval estimation), development of advanced statistical methods for the analysis of trends in mortality and its short-term and long-term prediction. Methods will also be developed and refined to use large number of noisy data, often stemming from heterogeneous sources with various prior assumptions. Of special interest will be environmental data where statistical modelling can serve as a common framework for working with huge amounts of data from satellites, climate and meteorological models and various observational networks. Dynamic and/or spatial statistical modelling, of either non- or semi-parametric form, will also be at the centre of our interest. Frequentist as well as Bayesian inference from dynamic and spatial models will be drawn for such data, involving models’ identification and parameter estimation, with applications in e.g. ecology, meteorology and anthropology, medicine and technical disciplines. Dynamic and spatial inference will be applied to e.g. energy consumption modelling. Department’s activities will also include GAM and state-space modelling for vibrodiagnostics, as part of the Czech Academy of Sciences’ programme ‘Strategy 21’.