Methodology

Data Collection

An automated pipeline runs every six hours, collecting articles from major Irish news publications via news RSS feeds and searching for reports of road fatalities. Sources include RTÉ, the Irish Times, the Irish Independent, the Irish Examiner, BreakingNews.ie and TheJournal.ie.

Data Extraction

Each article is read by an AI model (Claude) which extracts structured details: the date the collision occurred, the location (county, town, road), the number of people who died, and the vehicles involved. The model is given the article's publication date so that relative references such as "on Saturday" resolve correctly.

Locations are geocoded using OpenStreetMap's Nominatim service. Where an exact location cannot be determined, we fall back to the town or county centroid, so map pins are indicative rather than precise.

Deduplication

One death is usually reported by several outlets, sometimes days or months apart, and often with differing detail. To avoid counting a person twice, we compare records across five signals: how close the dates are, the number of deaths, the location, the similarity of the descriptions, and the vehicles involved. Confident matches are merged automatically; borderline cases are flagged for review.

When records are merged, every source article is kept, so each incident retains all of its citations.

Official Figures and Benchmarking

Official figures come from the Road Safety Authority, published as open data through the Central Statistics Office (table ROA29). We refresh them automatically and compare them against our own count every time the site updates.

The two will never match exactly, for two legitimate reasons. First, the definitions differ: official statistics count deaths from collisions on public roads, excluding car parks and private land, in line with EU convention. Ours are based on what is reported as a road death in the press. Second, recent official figures are provisional and are revised as Garda investigations conclude.

Any remaining gap is a measure of our own error, and we publish it rather than hide it. You can see the current comparison on the statistics page.

Why We Don't Backdate

Scraping news archives backwards through time sounds like it should work. In practice it does not. Measured against the official totals, our recovery of closed years was as low as 5%: for 2022 we found seven of the 152 people who died.

Publishing those numbers as if they were counts of the dead would understate reality by more than a hundred people a year, so we don't. Historical years on this site show official figures. Our own incident-level records for those years remain on the map, but they are a partial archive of what was reported, not a count.

Corrections

In July 2026 we found and fixed three faults that had inflated our figures: crash dates given the wrong year, which caused the same death to be stored more than once; news round-up articles ("four killed in 24 hours") recorded as a single incident carrying several crashes' combined toll; and year-on-year comparisons measured against our own incomplete record of the previous year, which made modest changes look like dramatic ones.

Reports published before that date carry a correction notice. We leave the original text in place rather than quietly rewriting it.

Known Limitations

• Not every road death is reported in the news, so we will miss some.
• Our current-year count typically runs slightly above the official figure for the same period.
• Some duplicates still slip through, particularly where one article names the person who died and another does not.
• Dates are inferred from article text and are occasionally wrong.
• Location accuracy depends on the detail in the report.
• This is not official data. For authoritative statistics, refer to the RSA.

Spotted a Mistake?

If you spot an incident that looks duplicated, misdated or wrong, please tell us — that feedback is the most reliable check we have. Data is collected every six hours and the site redeploys whenever it changes.