BACK TO SCHEDULE POSTER

Poster session 2

4 June 2026
15:45 – 16:30

Presentation title
From Control to Capability: How a Data Quality Framework Empowers Modern Data Stewardship Roles
Data governance becomes meaningful for organizations through the systematic management of data quality.

Read more Read less

This is especially important for administrative data, where high-quality register data is a key input for official statistics, policy analysis, and data reuse. Managing data quality at the source reduces duplication of data collection, improves statistical outputs, and increases trust in public-sector data. In Estonia, this approach is reflected in national data quality guidance developed to support consistent and sustainable implementation of data governance across public institutions.

The Estonian data quality guidance is built on a structured metadata model that defines how data quality is described, assessed, and improved in administrative registers. The model includes quality dimensions, measures, and indicators, as well as quality rules and business rules that formalize expectations for data capture and maintenance. Supporting artefacts such as a business glossary and data dictionary ensure consistent interpretation of data elements and classifications. Together, these components connect register management with statistical and analytical use.

Statistics Estonia is a national competence centre for data governance, supporting institutions and registers in implementing the data quality guidance. Through shared models, methodological support, and practical recommendations, this role promotes harmonization across registers and strengthens the reuse of administrative data for statistical purposes.

The framework is implemented through clearly defined data stewardship roles. Data owners set quality expectations and ensure compliance with legal and reuse requirements, while data stewards operationalize these expectations by maintaining metadata, monitoring quality indicators, and coordinating quality improvements. By embedding structured data governance actions into register management, organizations can improve data reusability, support high-quality statistics, and enable data stewardship roles as key drivers of effective data governance.

Main author / Presenter
Annika Uibopuu
Statistics Estonia

Read more Read less

Head of the Data Governance Team at Statistics Estonia, supporting national institutions in implementing data quality frameworks, metadata standards, and data stewardship practices to improve administrative data reuse and official statistics.

Presentation title
STATRevision: A Shiny Application for Revision Analysis
At Statistics Austria, revisions of published statistics are distinguished between occasional and routine revisions.

Read more Read less

The latter occur on an ongoing basis and are carried out in a planned manner according to a predefined publication procedure. The need to publish early estimates of relevant statistics, often based on incomplete data, makes it necessary to replace preliminary results with newer or final ones. Since routine revisions may alter preliminary results significantly, it is important to understand the direction and magnitude of such changes, as well as whether they are systematic or not. Following a recommendation from the 2022 ESS peer review report for Austria, the Center for Methodology at Statistics Austria developed the Shiny application STATRevision. Its main goal is to perform regular analyses of routine revisions in a standardised and systematic manner. The main benefit of using this tool is that it provides an overview and detailed information about the direction, extent and composition of revisions, based on summary statistics and visualisations. These results can help to improve the estimation process of preliminary estimates and highlight the need for earlier data availability. This presentation provides background information on revision analysis in general and outlines the current state of revision analysis at Statistics Austria. The main part will focus on the STATRevision application by demonstrating its key features and its use at Statistics Austria.

Main author / Presenter
Thomas Glaser, Alexander Kowarik
Statistics Austria

Read more Read less

Thomas Glaser graduated with degrees in Statistics and Sociology from the University of Vienna and has been working at Statistics Austria since 2007. He is part of the unit “Center for Methodology” at Statistics Austria. Besides the application of statistical methods to sample surveys, his responsibilities concerning quality management cover quality guidelines, quality reporting and revision analysis. Dr Alexander Kowarik is head of the methods unit at Statistics Austria with more than 15 years of experience working at an NSI. He is an active contributor to the R open-source community with a focus on official statistics application and participating in several international projects related to the usage of new data sources for the production of official statistics.

Presentation title
Assessing Data Quality in a Web-First Mixed-Mode Establishment Survey
In the context of official statistics, establishment surveys increasingly rely on web and mixed-mode data collection to counter declining response rates and rising fieldwork costs.

Read more Read less

While previous research has shown that mixed-mode designs can maintain response rates at lower costs, their implications for data quality remain insufficiently studied in establishment surveys. This study examines whether introducing a web-first mixed-mode design affects data quality compared to a traditional interviewer-based design.

The analysis is based on an experiment conducted within the IAB Establishment Panel, an annual survey of establishments carried out by the Institute for Employment Research (IAB). In 2018, a sequential web-first followed by face-to-face mixed-mode design has been tested against the traditional single-mode face-to-face design. To assess measurement quality, survey responses from both experimental groups are linked to administrative data derived from employer-level social security notifications. Measurement error is evaluated for about 20 variables which capture numbers of employees with specific characteristics.

Beyond measurement error, the study considers further indicators of data quality: item nonresponse, social desirability responding, use of filter questions, and answer consistency within the questionnaire. To address selection and nonresponse bias, calibration weights are applied throughout the analysis.

The results indicate that while measurement error in web interviews is for some variable higher than in face-to-face interviews, differences between the mixed-mode and single-mode designs are generally small. Overall, the findings provide empirical evidence that a web-first mixed-mode design can be implemented in establishment surveys without substantial losses in data quality. The study thus offers important guidance for statistical offices and survey practitioners when modernizing data collection strategies.

Main author / Presenter
Corinna König, Joseph Sakshaug
Institute for Employment Research (IAB), Ludwig Maximilian University of Munich (LMU), Germany

Read more Read less

Corinna König works as a researcher at the Statistical Methods at the Institute for Employment Research (IAB). Her research focuses on survey methodology for establishment surveys, with a particular emphasis on mixed-mode designs and data quality. Joseph Sakshaug is distinguished researcher, deputy head of research, and head of the Data Collection and Data Integration Unit in the Statistical Methods Research Department at the IAB. He is also university professor of statistics at the School of Mathematics, Computer Science, and Statistics at the LMU Munich (Ludwig-Maximilians-Universität München) and honorary full professor at the School of Social Sciences at the University of Mannheim. Previously, he was associate professor (senior lecturer) in Social Statistics at the University of Manchester (UK), and assistant professor (junior professor) of Statistics and Social Science Methodology at the University of Mannheim (Germany). Mr Sakshaug received his MSc and PhD in Survey Methodology from the University of Michigan-Ann Arbor, and his BA in Mathematics from the University of Washington-Seattle. From 2011 to 2013, he was an Alexander von Humboldt postdoctoral research fellow at the IAB and the LMU Munich. He is also a faculty member in the International Program in Survey and Data Science, and an adjunct research assistant professor at the Survey Research Center of the Institute for Social Research at the University of Michigan. His research focuses on data quality issues in complex surveys, the integration of multiple data sources, and empirical research methods.

CO-AUTHOR:

Presentation title
Evaluating LLMs for Open-Ended Survey Coding and Classification
Open‑ended survey questions are widely used to capture respondents’ attitudes, reasoning, and experiences, but the process of coding free‑text responses into analytical categories is resource‑intensive and interpretive.

Read more Read less

Recent developments in large language models (LLMs) have increased interest in their potential use for automating aspects of survey coding, particularly through the application of existing codebooks using natural‑language instructions. Despite growing experimentation in this area, there remains limited clarity about how such approaches should be examined and discussed within the methodological frameworks traditionally used in survey research.

This paper examines the use of large language models for coding open‑ended survey responses from a survey‑methodological perspective. The analysis draws on open‑ended responses from UK social and political surveys that were originally coded by trained human coders using established survey procedures. Model‑generated codes are considered in relation to human coding practices, without treating human judgement as an absolute reference point.

By situating the discussion of LLM‑based coding within survey methodology, the paper seeks to clarify how automated approaches can be examined using concepts familiar to survey researchers. The focus is on methodological alignment and analytical framing rather than on definitive performance claims, with the aim of supporting informed discussion about the role of automated tools in the processing of open‑ended survey data.

Main author / Presenter
Laone Maphane
University of Southampton, United Kingdom

Read more Read less

Dr Laone Maphane is Lecturer in Research Methods and AI Skills at National Centre for Research Methods (NCRM), University of Southampton. Her expertise lies in AI applications for survey methodology and quantitative research, including use of large language models (LLMs), computational macroeconomics, online data-collection methods, and AI for environmental sustainability and policy-relevant applied research.

CO-AUTHOR:

Presentation title
Combining Markov Systems and Fluid Dynamics for the Analysis of Migration Flows
PAPER

Migration flows between regions are frequently modelled using Markov systems, which provide a natural stochastic description of population transitions.

Read more Read less

Additionally, continuum and flow-based approaches are often employed to analyze migration dynamics. This study proposes a unified mathematical structure that connects these two perspectives by interpreting Markov dynamics as the motion of a compressible continuum.

Starting from the Kolmogorov forward equation, the stochastic constraint that probabilities sum to one ensures that the dynamics remain within the probability simplex. By introducing a suitable coordinate transformation centered at the stationary distribution, assumed to exist in the general case, the system is reduced to a lower-dimensional representation. This formulation allows explicit expressions for position, velocity, acceleration, and density evolution to be derived, offering a clear and interpretable framework for analyzing the system’s evolution. The approach is analytically tractable for systems with two or three states, which are particularly relevant for simplified representations of migration between countries, regions, or urban centers.

The proposed approach is primarily methodological and aims to support the analysis and interpretation of migration-related official statistics. By providing a physically interpretable, continuum-based representation of Markov migration models, it offers a complementary tool for studying stability, possible convergence, and structural properties of migration systems, using either observed or simulated data.

Main author / Presenter
Boukouvala Anna
Hellenic Statistical Authority (ELSTAT), Aristotle University of Thessaloniki (AUTH), Greece

Read more Read less

Currently a PhD candidate in Mathematics at Aristotle University of Thessaloniki, where research is focused on Markov systems as continuum media - a topic that bridges theoretical mathematics and applied modeling. The MSc in Statistics and Modelling was previously completed within the framework of the EMOS program, following a BSc in Mathematics at the same university. Research interests encompass stochastic processes, mathematical modeling, and their applications across a variety of scientific and statistical domains.

CO-AUTHORS:

Presentation title
Developing a Synthetic Census Public Use File within a Quality Management Framework
Public Use Files (PUFs) provide access to microdata while protecting the confidentiality of individual records.

Read more Read less

Traditional PUFs rely on rule-based anonymisation techniques, such as aggregation, recoding, and suppression. While effective in limiting disclosure risk, these approaches often substantially reduce analytical utility, particularly for multivariate and household-level analyses. The growing demand for more flexible and informative microdata access calls for new methods and technologies that support a transition towards modern statistical processes.

This paper presents the development of a synthetic Public Use File for the 2021 Census data as a model-based alternative to traditional census PUFs. Synthetic household- and person-level microdata are generated from a subset of the census data to preserve key statistical properties and hierarchical relationships while ensuring that no released record corresponds to a real individual or household. The synthetic dataset enables secure data exploration, methodological development, model testing, and data sharing without compromising privacy, while remaining comparable to traditional PUF dissemination practices.

The approach is embedded in a quality management framework that explicitly balances data utility and confidentiality protection. Data utility is evaluated through comparisons of marginal distributions, multivariate relationships, and selected analytical outcomes across the original census data, the traditional PUF, and the synthetic PUF. Confidentiality protection is ensured through model-based disclosure control and the absence of direct linkage between synthetic and original units.

The results demonstrate that synthetic census PUFs can enhance analytical utility relative to traditional anonymised PUFs while maintaining high confidentiality standards. The paper illustrates how synthetic data generation exemplifies the integration of new methods and technologies into data dissemination processes within a transparent and reproducible quality management framework.

Main author / Presenter
Ivona Kereta
Croatian Bureau of Statistics (CBS)

Read more Read less

Ivona Kereta is a Senior Advisor in the Sampling, Statistical Methods and Analyses Department at the Croatian Bureau of Statistics (CBS). She holds a master’s degree in Project Management. Since joining CBS in 2021, she has been working in the areas of survey sampling, seasonal and calendar adjustment of time series, and statistical disclosure control. She has also gained experience through her work on various statistical projects in the fields of agriculture, labour force statistics, income and living conditions, and other domains. In these projects, she contributed to improving sample selection processes, data weighting, and imputation of missing values. In addition to sampling-related tasks, she is actively involved in seasonal and calendar adjustment and in the continuous improvement of methodological processes. She also participates in defining rules and methods for the protection of confidential statistical data, as well as in monitoring trends and implementing new methods in these areas.

CO-AUTHOR: