Public-Records Data Studies Built to Be Cited

Sweden's offentlighetsprincipen, in chapter two of Tryckfrihetsförordningen, gives anyone the right to ask a public authority for its records, and most authorities answer within a week at no charge. That is where our studies come from. A dataset assembled from records nobody else bothered to request keeps earning citations years after publication. A panel survey stops earning the week it lands. Where the data comes from: Domstolsverket for court throughput and case duration. Kronofogden for enforcement. Bolagsverket for filings. SCB and Medlingsinstitutet for wage and labour statistics. Arbetsförmedlingen for vacancy data. UHR and Skolverket for admissions and school results. For Nordic comparison sets, Danmarks Statistik, SSB and Tilastokeskus. Outside the Nordics, Eurostat, the EU Open Data Portal, and Freedom of Information Act 2000 requests to UK bodies filed through WhatDoTheyKnow so that both the request and the response stay public. Method, and the part that is not negotiable: The method note is written before the analysis, and it names the extract date, the exclusion rules and what the denominator is. A second analyst re-runs every published figure from the raw file. Where the data is a census we say so and give no margin of error. Where it is a sample we print the interval and the n on the same line as the number. If a finding disappears once an obvious control is added, the finding does not publish. That has ended two projects at our own cost. How it is published: The page carries the charts. The dataset ships as CSV alongside the query or script that produced it, licensed CC BY 4.0 and deposited on Zenodo for a DOI. This is the step most shops skip and the one that matters. A licensed, citable dataset gets taken up by researchers, government reports and Wikipedia editors, and those references do not decay the way news links do. Charts come as SVG for the web and flat PNG at print resolution, because picture desks still ask for both. What we refuse to build: A thousand-person panel survey asking people about something they have no way of knowing. Any study where your company is the sole source of the data and also its subject — a reporter reads that as marketing, correctly. Charts with the n removed. Percentages on a base under fifty. A headline that survives only because of where the buckets were cut. The launch is a separate decision: We will pitch the study, and it is priced separately because it is different work with a different failure mode. Some studies do better placed with two named academics and a committee clerk than with forty reporters, and we will tell you which one you have rather than run the same launch either way. Not for you if: You need the finding to say something in particular. You cannot wait six to twelve weeks between the records request and publication, because authorities answer when they answer and there is no escalation that helps. Or your legal team will not clear publishing a number that reflects badly on a sector you sell into — which happens, and it is better discovered now than after the analysis. What arrives: The dataset and its licence. The method note. The built page as static HTML or as a React component written against your existing design tokens. The chart set in both formats. An embed snippet for third parties. And a note on when the source data refreshes, so you know when the study needs rerunning and when it stops being true.