Li-Chun Zhang is Professor of Social Statistics at University of Southampton and Senior Researcher at Statistics Norway. His research interests include graph sampling, finite population sampling, inference of machine learning, sample survey estimation, non-response treatment, measurement errors, small area estimation, index number calculations, editing and imputation, register-based statistics, population size estimation, statistical matching, record linkage. Li-Chun Zhang has authored the book Graph Sampling and co-edited the book Analysis of Integrated Data. He is currently Co-Editor-in-Chief of Journal of Official Statistics, and he has served as Associated Editor for journals such as the Journal of Royal Statistical Society (Series A) and Harvard Data Science Review. He has participated in the EU framework projects EURAREA, DACSEIS, RISQ and BLUE-ETS, the ESSnet projects Small Area Estimation, Data Integration, Quality of Multisource Statistics and Trusted Smart Statistics, the H2020-projectInGRID-2, the ESRC-projects ADRCE and NCRM-SAE.

Many technological, socio-economic, biological phenomena exhibit a graph structure that is the central interest of study, or the edges may provide effectively access to those nodes that are the study units. Either way, graph sampling is a statistical approach to study real graphs, which is universally applicable based on exploring the variation over all possible subgraphs that can be taken from the given population graph, according to a specified method of sampling, irrespectively of the unknown properties of the target population graph. Graph sampling encompasses the established theory of finite population sampling, because any latter situation can be represented as a special case of the former. All the so-called “unconventional” finite population sampling techniques, such as indirect, network, adaptive cluster or line-intercept sampling, can be more effectively studied as special cases of graph sampling. Whereas snowball sampling and various random walk sampling yield respectively breadth-first or depth-first observations in genuine graph problems. Graph sampling is also necessary for representation learning of large graph-structured data, including texts, images or networks as particular instances. Many relevant topics are together referred to as graph representation learning, including embedding of nodes or edges, node clustering or community detection, as well as neural networks that apply to data structured as graphs instead of vectors. The course provides an introduction to the central concepts of graph sampling, the most common sampling methods, and the construction of graph sampling strategy. A wide range of application areas and topics are illustrated, such as epidemiological studies, environmental and spatial statistics, network analysis, graph representation learning.

Zhang, L.-C. (2022). Graph Sampling. Chapman and Hall/CRC. https://www.routledge.com/Graph-Sampling/Zhang/p/book/9781032067087

Session 1
General introduction
Bipartite incidence graph sampling and estimation

Session 2
Breadth-first snowball sampling
Depth-first walk sampling

Session 3
Graph representation learning
Node embedding
Network classification and node clustering

Brunero Liseo is Professor of Statistics at Sapienza Università di Roma. His research interests include Bayesian methodology, Bayesian- Frequentist interface, approximate methods of inference, finite population sampling, sample survey estimation, measurement errors, small area estimation, population size estimation, record linkage. Brunero Liseo has co-edited the book Nonparametric Statistics. He is currently Editor-in-Chief of the journal “Statistical Methods & Applications. He supervised 12 Ph.D. students and published 75 papers in peer-reviewed journals.

The course is an introduction to a relatively new method for making prediction intervals with a prespecified coverage guarantee for any finite sample size. While this method is very popular in machine learning, it is not yet broadly used in survey sampling. In these lessons, we first briefly recall quantile regression, then introduce the conformal method and finally explore and discuss its modifications to address design- and model-based challenges arising in survey sampling and small area estimation. For the sake of brevity, we will only consider the case of continuous response variables. All lessons will be supplemented by practicals with R.

Lecture notes References: Shafer G. and Vovk, V. (2008) A Tutorial on Conformal Prediction, Journal of Machine Learning Research, 9, 371-421. Lei, J., G'Sell, M., Rinaldo, A., Tibshirani, R., Wasserman, L. (2018) Distribution-Free Predictive Inference for Regression. Journal of the American Statistical Association, 113(523), 1094-1111. Barber, R. F., Candes, E. J., Ramdas, A., & Tibshirani, R. J. (2021). Predictive inference with the jackknife+. Annals of Statistics, 9, No. 1, 486–507 Tibshirani, R. J., Foygel Barber, R., Candes, E., & Ramdas, A. (2019). Conformal prediction under covariate shift. Advances in neural information processing systems, 32.

Session 1
Intro to conformal prediction (CP): exchangeability and main algorithms (full and split CP, jackknife+).

Session 2
Challenges in CP and beyond: Conditional CP and weighted exchangeability; adaptivity.

Session 3
Conformal Prediction in Survey Sampling, Small Area
Estimation and Population Size Estimation