ICE Detention Stays at Georgia Facilities

Limitations

Calendar years, not federal fiscal years: `year` and `month` come from `stay_book_in_date_time` (the date the STAY began - for a stay transferred into Georgia, the date it began at a facility elsewhere), taken in Eastern time: ICE records each facility's local wall-clock time (DDP labels it UTC), so a stay that began at a facility in another time zone is converted to its Eastern date (963 stays moved to the next day, 23 of them into another month), so figures here will NOT match ICE's own or DDP's published FY totals (federal FY runs October 1 - September 30). Coverage is the release window 2022-10-01 through 2026-08-06: calendar 2022 is October-December only and 2026 stops on 6 August, so both end years are partial and a year-over-year chart dips at both ends. Stays that began BEFORE the window (1,961 with a Georgia stint, reaching back to 2004) are the left-censored residue of detentions still open when the FOIA window opened, not a series, and are dropped. WHICH STAYS COUNT: a stay is one uninterrupted period in ICE detention, possibly across several facilities (transfers do not end a stay), and this topic counts every stay with AT LEAST ONE stint (facility book-in) at a Georgia facility, however short, taken from the stint-level release of the same extract (DDP's duplicate-flagged stints dropped). Of the stays that began in the window, 78,248 touched Georgia; the stays file's own single-facility readings give 31,385 by first facility, 68,369 by the facility where the person spent the longest and 44,679 by last facility, so a 'detentions in Georgia' figure depends on the reading. Version 1.x of this dataset used the longest-facility reading; the 9,879 stays added in 2.0.0 passed through Georgia but spent longer at a facility in another state (for 88% of them one in AZ, MS, TX or LA). Among those that had ended, the median stay spent 7.6 days in Georgia (a median 26% of the stay), and 72% ended at a Georgia facility. Georgia is both an entry point and a way station: 48% of the window's stays whose first facility was in Georgia ended in another state, chiefly Louisiana and Texas, while people first booked in Fulton are largely moved on to Stewart and Folkston. 53 Georgia stints in the stint file carry no stay_ID and cannot be linked to a stay; a stay whose only Georgia time was one of them is missed. WHOLE-STAY MEASURES: `avg_stay_length_days`, `avg_facilities_per_stay`, `release_reason`, `arresting_agency_group`, `citizenship_country`, `demographic` and the `year`/`month` keys describe the WHOLE stay, including time held outside Georgia before it arrived or after it was moved on; a stay moved to Louisiana and removed from there is `removed` here. Only `avg_days_in_georgia` is Georgia-specific: the mean days the same ended stays spent at Georgia facilities (47.4 of 60.0 days on average across all ended stays in the window). COUNTY: each stay is placed in ONE Georgia county - the one where it spent the most days, summing its Georgia stints by the facility's county (an open stint runs to the end of 2026-08-06); ties go to the county of the most recent Georgia book-in, then the lower FIPS. 31,159 stays touched more than one Georgia location (most often the Atlanta hold room and Stewart, or Folkston and Stewart); they count only in their most-time county, so a county row is NOT every stay that passed through that county's facilities (use ice_detention_stints for book-ins by facility county), and county rows can never sum above the state total. Overlapping Georgia stints of one stay (3 stays in this release, not flagged as duplicates by DDP) are counted once. Geography is the DETENTION FACILITY'S county, never the person's residence: Stewart, Charlton, Irwin, Fulton and Clayton hold ICE facilities (Floyd and Cobb county jails also hold ICE detainees for short periods), and nothing here says where the detained person lived. 99.98% of Georgia stays resolved to a county (16 state-only, whose most-time Georgia location has no county in the source), so county rows sum to slightly less than the state total. Disclosure control (the platform's own rule; the source has none): county rows are annual totals with every categorical pinned to `all` (`month` and `citizenship_country` are published at state grain only), on a complete 159-county x 5-year grid; any county count below 5 - ZERO INCLUDED - is withheld as NULL, with complementary suppression so a withheld cell cannot be recovered by subtraction, and every average is withheld wherever its N is below 5 (at every grain) or withheld. A county NULL therefore means 'fewer than five, possibly none', never a real zero; state rows are complete and a state 0 is a real zero. At county grain num_stays - num_stays_ended (the still-open count) is published by implication even when small; it is a count of a category at a large facility, not an identity, and the state series publishes it directly. CENSORING: `avg_stay_length_days` and `avg_days_in_georgia` average ENDED stays only (7.6% of Georgia stays in the window were still open at extraction, 24% of those that began in 2026), so the most recent months are biased short - the long stays still in progress are missing from their book-in month - and an apparent fall (or rise, as they close) in average length near the end of the window can be a censoring artifact rather than a change in practice. `num_stays - num_stays_ended` is the open count in each cell. `demographic` is ICE's own gender determination (`Male` / `Female`), not self-reported; ICE's `Unknown` counts toward `all` only, so male + female is at most `all`. `citizenship_country` names the 22 countries most frequent in Georgia records; everything else is `other`. `arresting_agency_group` is the Deportation Data Project's program-group flag for the arrest that led to the stay (ICE vs CBP), not an ICE field; a NULL flag is published as `unknown`. `release_reason` collapses ICE's release labels to six outcome groups plus `still_detained` (no book-out by 2026-08-06).