← Sources & Dashboards

Sources that publish usage counts

A short list of public data sources that show how often each dataset is viewed or downloaded — and, just as important, what those numbers actually count.

Why this page exists

A catalogue with ten thousand entries answers “what exists?” The question people actually have is “which of these is worth opening?” Usage counts are the only crowd-sourced answer most portals offer — not authoritative, but better than alphabetical. Very few publishers expose them, which is why this list is short rather than selective.

A count is not a quality score
Counts are cumulative, so an ordinary dataset from 2015 will out-rank an excellent one from 2024 on age alone. Most portals do not separate people from scripts, so a dataset polled nightly by an automated job can look wildly popular. And a download measures access, not usefulness — sometimes it just means the file was the only way to see the data. Read the counts as a rough signal of attention, never of merit.

Sources

Each entry states what is being counted, because views, downloads and API calls are three different things and portals seldom say which they mean.

Socrata portals — data.cdc.gov, healthdata.gov, and most state open-data sites CDC Per dataset
CDC PLACES county file: 181,877 views · 61,569 downloads  |  PLACES place file: 23,072 views · 15,700 downloads
Every dataset page shows a view and download count. The numbers are also machine-readable without an account — append the dataset id to /api/views/ and add .json: data.cdc.gov/api/views/swc5-untb.json returns viewCount and downloadCount among the metadata. The same pattern works on any Socrata-hosted portal, which is most state open-data sites.

What it counts: cumulative totals since the dataset was created, with no published separation of human from automated traffic. A file behind a nightly API job accrues views that no person ever saw. Treat the ratio of downloads to views as more informative than either number alone.

Zenodo Repository Per record
Every record page carries views and downloads. Zenodo is unusual in documenting its method: it logs an anonymised visitor id, the country of origin, the referrer, and a classification of each request as human, machine or robot — so the distinction most portals ignore is at least being recorded here.

What it counts: record-page visits and file downloads, tracked separately. Counts changed at the October 2023 platform migration, so figures spanning that date are not comparable with what was shown before it.

Draft — more entries to come. Candidates still to be checked before they earn a place here: ICPSR, Harvard Dataverse, Figshare, Kaggle Datasets and data.gov. Each goes on the list only once its counts have been confirmed as publicly visible and its definition of a “view” established.