BACK TO SCHEDULE SPEEDTALK

Speed talk session 10

Statistics as a Social Good

4 June 2026
13:15 – 14:00
ŠIBENIK VI SHOW ON MAP

Session chair
Andrea Galić Nagyszombaty
Director, Statistical Methodologies, Quality and Customer Relations, Croatian Bureau of Statistics (CBS); Co-Chair of the Programme Committee

Read more Read less

Andrea Galić Nagyszombaty has been the Director of the Statistical Methodologies, Quality and Customer Relations Directorate at the Croatian Bureau of Statistics (CBS) since October 2019. She graduated from the Faculty of Economics and Business, University of Zagreb, after which she completed a master’s degree in macroeconomics. In 2016, she earned her PhD in economics, with a dissertation titled: 'Impacts of the Modern Systems of Tax Payment on Trends and Estimation of the Unofficial Economy'. She is the author and co-author of several papers in the field of economics, particularly focusing on the unofficial economy. She has over 20 years of experience in official statistics. During that time, she has worked in various areas within the CBS, including ten years in national accounts and several years as Head of the Department for Strategic Development of the Statistical System of the Republic of Croatia. She was appointed project manager for the 'Improving the Coverage of National Accounts of the Republic of Croatia – Non-Observed Economy’ project and has participated in a variety of working groups. In 2015, she completed an internship with the Scottish Government, in the Office of the Chief Statistician and Strategic Analysis, focusing on strategic development. Since 2019, she has been especially dedicated to enhancing the dissemination and communication of official statistics and has represented the CBS at numerous meetings and conferences. She invests particular effort and enthusiasm in the promotion of official statistics and the improvement of statistical literacy.

Presentation title
Whatever Affects one Directly, Affects all Indirectly
PRESENTATION

When microdata cannot be publicly disseminated, detailed documentation becomes essential.

Read more Read less

For machine learning applications in official statistics, transparent dataset description directly supports several quality dimensions: relevance, by clarifying the dataset’s analytical purpose; accuracy and reliability, by enabling replication of preprocessing and modelling procedures; and coherence and comparability, by situating the dataset within established statistical frameworks. This paper therefore provides a structured description of the AIML4OS–WP11-PT dataset, an administrative-based representation of the full population of inter-firm commercial transactions in Portugal for 2022. Although the confidential microdata cannot be shared, documenting its construction is crucial for its responsible use in ML-driven statistical research.

The dataset portrays the Portuguese production and supply-chain system as a directed, weighted network in which firms are nodes and business-to-business transactions form edges. Its comprehensive, threshold-free design enables analysis of how economic linkages emerge, how shocks propagate through production chains, and where systemic vulnerabilities may arise. Such a representation supports both descriptive analyses of economic connectivity and predictive modelling of networked behaviour, offering a foundation that can be compared and adapted across statistical systems.

Because of the high sensitivity of the underlying data, the dataset itself cannot be released. However, its processing algorithms and analytical methods are fully reproducible. This constraint highlights the importance of producing transparent summary network statistics, which convey structural information without revealing confidential firm-level details. Such summaries enhance accessibility and clarity, allowing users to understand the dataset’s key characteristics while respecting confidentiality protections.

The empirical analysis includes standard network measures—degree distributions, reciprocity, clustering coefficients, path lengths, and weighted connectivity—that capture fundamental features of complex systems. These indicators provide interpretable representations of network structure and dynamics. For machine learning, they are particularly valuable in assessing representativeness, sparsity, and potential sources of bias, thereby reinforcing the dataset’s accuracy and reliability even when direct access to microdata is restricted.

To explore structural heterogeneity, seven subnetworks were derived, organised by firm size (DIM), geographical region (NUTS3), and economic activity (NACE). These subnetworks reveal how connectivity patterns vary across sectors and locations, supporting more relevant and targeted analytical applications.

Finally, the characteristics of the AIML4OS–WP11-PT dataset align closely with those observed in other national production networks, such as those of Ecuador and Hungary. This convergence validates the dataset’s structure and enhances the comparability and generalisability of ML models built upon it, enabling the Portuguese case to contribute meaningfully to broader analyses of economic interdependencies.

Main author / Presenter
Alexandre Cunha
Statistics Portugal (INE)

Read more Read less

Alexandre Cunha holds a degree in Applied Statistics from Universidade do Minho. Has been working at Statistics Portugal (INE) since 2021, within the Data Collection Department, where he focuses on the processing, integration and analytical use of administrative data in support of official statistics.

CO-AUTHORS:

Presentation title
Strengthening User Trust through Quality Management in Official Statistics: Lessons from Somalia
PRESENTATION PAPER

Building and maintaining user trust in official statistics depends fundamentally on the perceived quality, accessibility, and relevance of data.

Read more Read less

This paper examines how improvements in statistical quality have contributed to strengthening user trust in Somalia’s official statistics between 2017 and 2025, drawing on global performance metrics and national user feedback. The analysis focuses on evidence from the Open Data Inventory (ODIN), the World Bank’s Statistical Performance Indicators (SPI), and findings from the first and second National User Satisfaction Surveys conducted by the Somalia National Bureau of Statistics (SNBS).

According to the Open Data Inventory 2024, Somalia ranked 157th globally with an overall score of 40 out of 100, reflecting low data coverage (32) and relatively stronger data openness (46) . These results indicate persistent challenges in data completeness and frequency, despite progress in publishing machine-readable and accessible datasets. In contrast, the World Bank’s Statistical Performance Indicators show substantial system-wide progress, with Somalia’s SPI increasing by approximately 147 percent from 19.6 in 2019 to 48.4 in 2022, and further improving to 49.7 in 2023. This improvement reflects gains across the five SPI pillars—data use, services, products, sources, and infrastructure—signaling a maturing national statistical system.

User perceptions provide critical insight into how these quality improvements translate into trust. The first National User Satisfaction Survey found that 55 percent of users were satisfied with the overall quality of official statistics, although satisfaction with timeliness, accuracy, frequency, disaggregation, and coverage remained below 45 percent . The survey also revealed gaps in responsiveness to data requests and limited awareness of official statistics among some users. Findings from the second User Satisfaction Survey indicate increased stakeholder engagement and improved satisfaction with accessibility, usability, and service delivery, particularly for first-time users, while highlighting continued demand for stronger analytical quality and interpretation .

By triangulating international quality indicators with national user feedback, this paper demonstrates that targeted improvements in statistical quality—especially openness, accessibility, and user engagement—are central to building and sustaining trust in official statistics in fragile contexts. The Somalia experience underscores the importance of continuous quality management, transparent communication, and responsiveness to user needs as foundations for credibility and trust in official statistics.

Main author / Presenter
Abdulkadir Gedi
Somali National Bureau of Statistics (SNBS)

Read more Read less

I am a data science and data management specialist at the Somali National Bureau of Statistics (SNBS), with over eight years of experience in the production, management, and dissemination of official statistics at national and international levels. My work focuses on data quality management, SDG monitoring and reporting, statistical data portals, and the implementation of Statistical Data and Metadata Exchange (SDMX) standards. I hold a Master of Science (M.Sc.) in Computer Science and Engineering (Data Science) from United International University, Dhaka, Bangladesh. I have completed advanced professional training in machine learning for official statistics, big data analytics for policy planning, data anonymization, spatial data infrastructures, and data leadership through programs delivered by UNSD, WB, ESCWA, IMF, SIAP, ADB, and ILO. My professional interests include strengthening data credibility, transparency, and user trust in fragile and post-conflict statistical systems.

Presentation title
Strengthening Data Quality in Official Statistics: The Case of the Türkiye Violence Against Women (VAW) Survey
PRESENTATION

In the compilation of official statistics, data collection of high quality is of great importance, notably in the case of socially sensitive issues such as the “Violence Against Women Survey” (VAW), which used various methodological and quality-enhancing measures throughout the implementation process to maintain accuracy, consistency, and efficiency.

Read more Read less

This paper examines these measures taken in the official statistics production process and evaluates the overall contribution of this study to improving data quality within the national statistical system. In the sampling stage, the research was designed to decrease the non-response rate by selecting households with at least one woman aged 15-59, which was the target group of the research. By sampling only from such households, the research minimized unsuccessful interview attempts, increased overall response rates, and helped field teams use their time and resources more effectively. The length of each interviewer's interview was tracked during the interview process and regularly reported to the Regional Directorates involved in the implementation process to allow irregular interview patterns to be detected early and interventions to be supported when necessary. Interview duration monitoring served as a function of quality control and ensured that interviews were conducted and in accordance with established standards. In the 2024 survey, consisting of many modules and questions, consistency between questions was monitored throughout the fieldwork using software developed specific to the study, and feedback was sent to Regional Directorates in real time so that corrections could be made. Some basic demographic information obtained from administrative records was previously entered into the CAPI system, reducing data entry mistakes and shortening survey time. The consistency among survey responses and administrative records was then checked to make sure the final set of data was high-quality and reliable.

Main author / Presenter
Eda Evin Şahin
Turkish Statistical Institute (TurkStat)

Read more Read less

I have been working at the Turkish Statistical Institute since 2004 and currently hold the position of Specialist. I am a PhD candidate in Demography, with extensive experience in population registers and household surveys. I have contributed to the implementation of major official statistics projects, most recently the Surveys on Violence Against Women and Crime Victimization. My professional responsibilities include questionnaire design, development of data entry systems, data processing and analysis, preparation of statistical publications, and reporting.

CO-AUTHOR: