The geography of reported immigration enforcement activity
This map reads all 4,299 rows from DefrostMN-AllActivityData.xlsx, including 171 records outside the previous seven-county study area. The initial view is centered on the Twin Cities; “Fit all records” reveals the full extent. Each plotted mark represents a source report, not necessarily an independent incident or a verified enforcement action. Confidence and activity categories follow the workbook.
Time selection uses activity_date, not record-entry date. Dates are treated as local calendar dates without timezone conversion. Playback advances by one day; it does not imply continuous observation or movement between points. Point colors show date, category, or confidence as selected. The bars show report counts, not enforcement rates. Reporting intensity, observer coverage, repeated sightings, and recording delays may vary over time.
No Good/Pretti dates, breakpoints, surge phases, or event windows are imposed or estimated here. A future phase analysis should infer candidate changes from the time series and assess reporting changes before relating those candidates to external events. The earlier report contains exploratory fixed-date comparisons; those comparisons do not establish phases.
The dataset, map library, roads, water, county boundaries and city labels are embedded locally. The map makes no tile requests and works offline. Roads and water are generalized U.S. Census TIGERweb reference geography downloaded October 16, 2026, covering 95°W–91°W and 41.8°N–45.8°N. This includes the full report extent. Only primary/secondary roads and water polygons at least 0.1 km² are shown; local streets are omitted. These reference layers reflect current geography, not necessarily the observation period. Report coordinates are unchanged. County outlines: U.S. Census TIGERweb ACS 2024. Map interaction: Leaflet 1.9.4, BSD-2-Clause. Roads: Census TIGERweb Transportation. Water: Census TIGERweb Hydro.
ACS community overlay: Census January 1, 2024 boundaries linked by exact GEOID to official 2020–2024 ACS five-year estimates. County subdivisions represent cities, townships and other minor civil divisions; a city spanning counties may appear in separate parts. Incorporated places are a separate alternative geography, including cities and villages, but not townships. Whole areas intersecting the reference-map extent are included. There are 2,146 county subdivisions and 941 incorporated places. Three areas have no matching ACS table row and display “Unavailable”: the Dakota County portion of Cannon Falls city and the two Baldwin city geography entries. The two boundary systems are never added together. Hover or click only when the overlay is enabled. The overlay is static population context, not a population estimate for an individual report or date.
Population and race/ethnicity use table B03002, nativity uses B05012, and median household income uses B19013 in 2024 dollars. Hispanic/Latino includes any race; displayed White, Black and Asian categories are non-Hispanic and single race. Foreign born does not identify immigration status. Percentages divide category counts by their table total. Margins of error are 90%; derived percentage MOEs use the ACS subset-proportion formula, with the ratio formula when needed (Census calculation guidance). Unavailable or special values are shown as unavailable, not zero. Shading uses fixed percentage intervals (0, 10, 25, 50, 100%) or income intervals ($0, $50k, $100k, $150k, $250k+), not significance classes. Downloaded tables, raw boundaries, joined CSV/GeoJSON and metadata are stored locally under analysis/data/acs_communities_2024/. Source: Census 2024 ACS Summary File.
Prepared October 16, 2026. All changes are local; the source workbook and Google Drive files are unchanged.
More reports, a larger footprint, and weaker clustering are different changes. Explore each separately, with the same study area and distance scales every day.
Calendar coverage and sample size come first. Click a bar to select its date.
Follow the line colors from early to late. The two reference patterns stay fixed across dates.
Hover over a curve for its date and sample size; click to inspect that date below.
Higher L−r = stronger local concentration. A line closer to the black curve resembles the overall reporting geography; a line closer to zero resembles uniform randomness. These are different comparisons. The pooled reference includes pairs across days and is not an average of daily curves. Its 199 uniform simulations use n=4,128; their pointwise band is not a significance band for daily or weekly curves. This summary is independent of the spatial-reference dropdown; the Function control below switches both profiles between K and L−r.
RMS curve departure over the same 1–10 km distance grid. Lower values mean closer agreement; this is a difference between profiles, not the geographic range of activity.
The shaded band is a 95% pointwise simulation envelope for the selected date’s sample size. It is not a confidence band, a simultaneous test across distances, or a test comparing the selected dates. Curves are suppressed when fewer than 15 reports or too few interior focal points are available.
Hull area and maximum distance are sensitive to sample size and outliers; nearest-neighbor distance also depends on sample size. The 90% radius describes a circle around that window’s centroid, not an administrative territory.
How far apart are pairs of reports within each selected window?
Line: 10th–90th percentiles; box: 25th–75th; marker: median. These describe observed pair distances, not uncertainty. Pairs are dependent and this plot is not edge corrected.
All daily and weekly K/L calculations use the same seven-county polygon: Anoka, Carver, Dakota, Hennepin, Ramsey, Scott and Washington. Coordinates are projected to NAD83 / UTM zone 15N and measured in kilometers. The map still contains every source report. Map filters and the map viewport do not affect this analysis.
For each distance r, use only focal reports at least r from the study-area boundary, while counting all their neighbors within r. The finite-sample estimator is K(r) = A × total neighbor count / [(n − 1) × eligible focal reports]. The centered transform is L(r) − r = √[K(r)/π] − r. Distances run from 0.25 to 10 km in 0.25 km steps. Require at least 15 reports and at least max(10, 25% of n) eligible focal reports at each radius. A reduced-sample K estimate need not be monotonic.
Uniform randomness: 199 simulations with the same n placed uniformly inside the fixed polygon, including water. Positive L−r indicates more close pairs than this geometric benchmark, but urban density alone can produce that result. This is not a realistic model of enforcement opportunity.
Overall reporting geography: sample n report locations without replacement from all 4,128 study-area reports, repeated 199 times. This retains the dataset’s uneven geography and repeated locations. It asks whether the selected window differs from the overall reporting footprint. It is not an independent baseline, an exposure model, or an inhomogeneous K estimate.
K/L evaluates pair concentration at increasing distances. RMS spread, the 90% centroid radius, hull area and diameter describe the footprint. Activity can spread outward while remaining clustered within local areas. A peak of L−r is not automatically a cluster’s diameter or an operational scale.
A trailing week pools seven complete calendar days before estimating K/L. Adjacent weeks overlap and are not independent replications. Repeated sightings, gathering/staging and geocoding precision can affect very short distances. The optional sensitivity curve collapses only exact same-day coordinates; it is not incident deduplication and has no separate simulation envelope.
The reference bands are pointwise 2.5–97.5% simulation ranges, requiring at least 80% valid simulations at a radius. They do not correct for searching many dates or distances. No p-values or automatic phase boundaries are reported. Sparse days are retained in volume charts but suppressed in K/L. Changes in observer coverage and reporting effort can mimic changes in activity.
The saved daily/weekly metrics and study-window membership provide a reproducible basis for later demographic modeling. That stage should distinguish population exposure, land use and observation effort from interaction among reports; consider inhomogeneous K and spatial count models with appropriate denominators. No demographic effect is estimated here.
Historical context: Operation Metro Surge was underway in December; January 6 marked a publicly announced expansion. Good’s death on January 7 is contextual information, not a model breakpoint. The January 6–8 defaults are an exploratory comparison requested by the user; any dates can be selected. The observed rise in reports alone does not establish its cause.
Methods: spatstat K-function documentation. Timeline: Minnesota Attorney General; AP, January 6. Seed: 20261016. Source reports are not necessarily independent incidents or verified enforcement actions.
