Explore

Pick a dataset — Use the sidebar, or type an alias in the Dataset box — 311, collisions, inspections. Aliases exist because a Socrata id like erm2-nwe9 is unguessable; the id is still shown, and still works.

Choose a measure and a grouping — countanswers “how many”. Group by a categorical column to break that down. Only columns that group into meaningful buckets are offered — grouping by a timestamp gives one bucket per row, which is a list, not an aggregate.

Filter from observed values — Filter values autocomplete from what the column actually contains. This matters more than it looks: a guessed spelling returns zero rows and reads as “nothing happens here” rather than as a mistake. Community boards, for instance, are spelled 04 MANHATTAN, not 104 MANHATTAN.

Read the connections carefully — Row join means both datasets carry the same real-world identifier — a tax lot, a building, a restaurant — so records correspond. Same area means they can only be compared once aggregated to a borough or district; it does not match individual records. No key means related subject matter and no way to join at all.

Nothing is downloaded — Every query is pushed to Socrata, which aggregates server-side and returns kilobytes. A first run over a very large dataset can take ten seconds or so; the result is cached, so a repeat is quick.