Dataharvest 2026 - the European Investigative Journalism Conference

Reporting on transgender people can be a minefield: official data is scarce, public understanding is patchy, and with their rights increasingly contested by politicians and commentators, the community itself is often wary of journalists.

Drawing on Investigate Europe’s project The Cruel Condition -published with Arte, New Lines, Taz, and others- this session explores how laws across Europe pushed trans people toward sterilising surgeries over decades. It will outline how we compiled first-of-its-kind data to produce an unprecedented cross-border estimate, and share practical insights on building trust with trans sources.

We will also look at other investigations using data to cover LGBT issues (e.g., on money flows to conservative groups), a topic quite often overlooked by data journalists, to consider possible pathways for future investigations amid an intensifying backlash against LGBT rights in many parts of the world. We will look at other examples and invite you to discuss how to be creative with unusual, and sometimes even non-existent, datasets.

Large language models can do more than generate text – they can help clean and structure messy data files as well as enrich datasets. As LLMs increasingly become a useful tool for data journalists, the Ellmer package is a useful resource for R users to easily work with LLMs. The Guardian data team has used the Ellmer R package to clean and organise thousands of emails from the Epstein files, to investigate private equity firms in the United Kingdom, and to classify recipients of climate finance.

Using some of these examples, attendees will learn when this package can be the perfect tool for your investigation, which are the good practices when using LLMs, how to connect to an API of an LLM, how to write an efficient prompt, how to submit the prompts in bulk using the batch function for structured data and how to evaluate your results and iterate for improvements.

This is an advanced R session and we will assume that attendees have some prior knowledge of R.

Learn how the investigative team at the Danish Altinget used scraped data from inspection reports to map structural inequality in Copenhagen’s nurseries and kindergartens, and how the method can be applied to other local areas and welfare institutions.

We scraped data from +350 inspection reports and created our own color coded ranking system to investigate the quality of daycare centers across Copenhagen.

Our investigation found that 14 daycare centers have serious problems, 3 had illegal practice and most of the problematic institutions located in a very small area. It lead to one daycare center being closed down (with the funds transferred somewhere better), an extensive political debate in the local elections and in the local government of Copenhagen and the most selling story in Altinget Privat’s history.

In this session we will go through the hands-on methods behind our investigation, and discuss how it can be applied to other countries, local areas or institutions.

We will focus on scraping data from websites and PDF's with a minimum of coding, of structuring data found in a large amount of documents and of maximizing impact of your investigation. Also, our talk will feature a warrior princess, a shield with a unicorn and a flaming sword!

Looking forward to seeing you.

Have you ever come across a striking data visual and wondered: "How the hell did they make that?" Your follow-up question may be: "Can I achieve that too with the resources I have?" In this session, we start with a deep dive into the (technical) production of two data visuals. Then, for the third visual, it's your turn to figure out how it might have been made. Finally, we launch our online inventory and subjective guide of data visualization tools - a work in progress you can contribute to!

Heads up: we won't teach any tool in depth, but we'll highlight a range of options so you can find an approach or tool that best fits your needs and constraints (e.g., limited budget, small team, learning curve…).

Scrollytelling based on spatial data is a powerful way to immerse readers both in the context of global events and in a local, personalised story.In this session, we will cover the technical aspects and the visual and narrative techniques that help create smooth, coherent, and engaging storytelling.We will look at some case studies based mainly on historical topics, but with approaches you can apply to a wide range of narratives.  By the end of the session participants will have learned:    

• how to develop the script for a visual story,
• what scrollytelling engines do, and how they work,
• what types of data you will need to work with, and
• how to prepare and process the data.

Court decisions are treasure troves of verified facts—but they're often locked in dense legal documents that the public never sees. This session will present a specific methodology for extracting, contextualizing, and translating judicial findings into compelling multimedia narratives that serve human rights advocacy, accountability, and education.

We want to present the methodology we developed in the Balkan Investigative Reporting Network in Bosnia and Herzegovina - BIRN BiH - on how to connect court findings into coherent narratives, translating court-verified facts into accessible, informative, and educational content.

We will use three BIRN BiH developed databases to demonstrate our methodology. Additionally, we want to present how we used court facts to develop educational tools. Many journalists and human rights documentalists treat court records as supporting evidence for stories. This session will show how court decisions themselves can be the primary story—and how systematic extraction creates accountability infrastructure that outlasts individual articles or projects.

Investigating data-driven stories on environmental health, long-term pollution, or biodiversity loss often means facing a structural problem: the data don’t exist, or have never been systematically collected, not even by public institutions. In these cases, the only option is to produce them from scratch.

This session presents a journalism-led methodology that integrates citizen science and social research as core tools of investigative data journalism. Developed through collaboration with scientists while preserving editorial independence, the approach is grounded in concrete case studies and practical experience.

Drawing on projects conducted with communities exposed to chronic industrial pollution, environmental degradation, and conflict, the session shares practical tools, workflows, and lessons learned for designing, collecting, validating, and using original data when datasets are missing.

Participants will learn how to:
-design and run a citizen science project for investigative reporting, from framing research questions and defining indicators, to engaging and training citizen scientists, data collection, quality control, and journalistic use of the data- establish effective collaboration with scientists during research design, data collection, and verification through shared protocols and a memorandum of understanding
- design a rigorous social research process to investigate community needs and priorities using surveys, interviews, and qualitative methods, including collecting and analysing texts to identify trends, patterns, recurring themes, and undercovered issues
- integrate community-generated data, social research findings, and scientific measurements into original investigations and engaging stories.

The session also demonstrates how this approach can generate multiple, high-impact outputs: original investigative reporting based on co-produced data, policy briefs grounded in empirical evidence, and social research outputs in scientific journals.

This session will explore a new method: scraping on-screen text, using immense volumes of videos—from police dashcam videos to 24-hour news TV channels to fuel your investigation.

Antoine Schirer, a designer turned journalist, will share the methods and scripts he used for a 2025 Reporters Without Borders investigation to dissect months of programs of the controversial French news channel CNews.

Using Python, OCR, and fuzzy string matching, more than a million news banners were analysed to expose how the channel gets around broadcasting legislation.

We will look at other examples and invite you to discuss how to be creative with unusual - even non-existent datasets.