Methodology & sources
Every number, and exactly where it comes from.
Apartment History is an aggregator, not a grader. Here's each source we pull, what we do with it, and the honest limits — so you can weigh it yourself.
Where the data comes from
| Signal | Source | What it is | The honest limit |
|---|---|---|---|
| Reviews | Public Google reviews | The full review history for a building — rating, date, text, and any management response. We rebuild the timeline from these. | Older reviews carry relative dates (“a year ago”), so dates before ~12 months back are approximate to the year. |
| Crime | FBI Crime Data Explorer (NIBRS/UCR) | Official offenses reported by the city’s police agency, per 1,000 residents, against the state and national average. | City/agency-level — not the building. Non-reporting years (e.g. the 2020–21 NIBRS transition) are shown as gaps, never zero. |
| Walkability | EPA National Walkability Index | A 0–20 score for the building’s census block group, from street connectivity, transit access, and nearby destinations. | Block-group level; a vintage national dataset, refreshed when the EPA updates it. |
| Air quality | Open-Meteo air quality | Recent (~90-day) average and peak US AQI plus PM2.5 near the address. | Regional grid (~11 km), so buildings in the same area share a reading. |
| Internet | FCC National Broadband Map | Providers, technology (fiber/cable), and top advertised speeds available at the address. | Advertised maximums; actual availability can vary by unit. |
How we handle it
Sourced, not scored
We surface what reviewers said and what public records show. We never assign our own grade, star, or “verdict” — the reader judges.
Trajectory over average
A single average hides everything. We compute the rating over time and detect the change-point — the month the rating actually moved — using statistical change-point detection (the Pelt algorithm).
Flags, not accusations
We flag unusual patterns — a sudden burst of 5-stars, reviews that don’t match a long complaint history — as “worth a closer look.” We never call a review fake or name a person.
Hygiene, lightly
We collapse exact-duplicate text and note small samples. We don’t silently delete reviews or “clean” the data toward any conclusion.
We store only the derived result — the trends and counts — not a copy of the review corpus, and we link out to the original source. See a building in action on the demo or explore everything we track.