The Worst US Grade Crossings Are Low-Traffic Lines With Persistent Local Hazards, Not High-Traffic Urban Crossings

USE CASE. State DOTs administer the federal Section 130 program (23 USC 130) which provides grade-crossing safety improvement funding allocated to state DOTs by formula. The standard FRA Accident Prediction Formula (APF) used in Section 130 prioritization ranks crossings by predicted accidents per year as a function of train traffic, train speed, vehicle traffic, and warning device type. The APF is implicitly normalized by exposure but it is NOT a per-train-movement historical rate, and it does not directly surface crossings whose actual historical accident rate is wildly higher than the APF predicted rate. A direct per-million-train-movement historical ranking is the missing diagnostic that identifies crossings where a structural local hazard (sight obstruction, road geometry, vehicle volume not in the inventory, dispatch pattern) is producing accidents far above what the inventory parameters predict. RESULT. Of 8,833 crossings with at least one 2020-2025 accident and a non-zero daily-train count in the inventory: total system rate = 33.04 accidents per million train movements. The top 5 raw-accident-count crossings (the conventional ranking): 025422P Glendale AZ BNSF 24 accidents at 12 trains/day (rate 913/million); 025617C Phoenix AZ BNSF 20 accidents at 12 trains/day (761); 025430G Phoenix AZ BNSF 15 accidents at 12 trains/day (571); 923252B Jersey City NJ Norfolk Southern 14 accidents at 4 trains/day (1,598); 743643X Jersey Village TX 7 accidents at 2 trains/day (1,598). These are mostly high-volume urban crossings where exposure drives the raw count. The top 5 RATE crossings (the non-obvious ranking, restricted to ≥3 accidents): 348502B BARTLETT TN CSX 9 accidents at 1 train/day (rate 4,110/million, 125× system mean); 352618J BESSEMER AL 5 accidents at 1 train/day (2,283); 637912X MIDLAND CITY AL 4 accidents at 1 train/day (1,826); 923252B JERSEY CITY NJ Norfolk Southern 14 accidents at 4 trains/day (1,598); 743643X JERSEY VILLAGE TX 7 accidents at 2 trains/day (1,598). STRUCTURAL FINDING. Of the top 10 by per-million-train-movement rate, six are 1-train-per-day crossings — meaning that the worst per-exposure accident rates in the country are at branch lines and local industrial lines that see only one train per day. These are not the crossings that routinely appear in news coverage of dangerous grade crossings, which focuses on high-traffic urban locations. The Bartlett TN example (gxid 348502B) is striking: 9 highway-rail crossing accidents in six years on a one-train-per-day line — one accident every 8 months at a crossing where the train passes through once a day. That ratio cannot be explained by exposure; it is a local hazard signature (likely sight obstruction, road geometry, or vehicle traffic far above the rural mean), and it is exactly the kind of crossing that the conventional raw-count Section 130 prioritization tends to deprioritize because the absolute accident count is small. CAVEATS. (1) The 'totaldaylightthrutrains + totalnighttimethrutrains' inventory field captures THROUGH trains; it excludes switching moves, transit moves, and maintenance equipment, so the true train-movement denominator may be higher for some crossings (especially industrial yard-adjacent crossings), pulling the rate down. (2) The accident counts include both vehicle-train collisions and pedestrian-train incidents; for some urban crossings the high count is pedestrian-driven rather than vehicle-driven. (3) The 8,833-crossing universe is restricted to crossings with at least one 2020-2025 accident in the database; it does not represent the entire 438,538-crossing inventory. The structural finding holds within this accident-prone subset. (4) The FRA APF used in Section 130 implicitly captures some of this via its historical-accidents weighting term, but the Bartlett-style 1-train-per-day crossings do not jump out of an APF ranking as clearly as they jump out of a per-million-train-movement ranking.

When a car or truck or person gets hit by a train at a grade crossing, the Federal Railroad Administration logs it. They publish two databases: every accident going back decades, and an inventory of every grade crossing in the country with details like how many trains pass through per day. State DOTs use this data to decide which crossings to fix first with federal Section 130 funding. The standard ranking they look at is the raw accident count: which crossings had the most accidents recently? The top of that list is dominated by high-volume urban crossings where lots of trains and lots of cars meet. That makes intuitive sense — busy crossings have more accidents because they have more exposure. I joined the two databases and asked a different question: instead of raw accident count, what is the accident RATE per train passing through? The answer flips the picture. The worst per-train-passage accident rate in the country isn't at a busy urban crossing — it's at a small crossing in Bartlett, Tennessee that sees ONE train per day. That single crossing has had 9 accidents in the last six years. One accident every eight months on a line that the train passes once per day. That ratio is not exposure-driven. Whatever is causing those accidents is a local hazard at that specific crossing — bad sight lines, a weird road angle, more vehicle traffic than the inventory shows, something. Six of the top 10 worst per-train-passage rates in the country are also one-train-per-day crossings: small lines in Alabama, Florida, Georgia, Louisiana, Indiana, where the same crossing keeps having accidents even though almost no trains use it. None of these crossings appear in the conventional 'most dangerous grade crossings' news stories, which always cite the high-volume urban crossings. Why this matters: state DOTs and the FRA prioritize Section 130 grade-crossing improvement funding by an Accident Prediction Formula that mostly weights exposure. The Bartlett-style crossings — where the accident rate per train is sky-high but the absolute count is small — get deprioritized because in raw-count terms they look unremarkable. They shouldn't. They are exactly the crossings where a small physical change (a gate, a stop sign, a sight-line clearance) would prevent the most accidents per dollar spent, because the pattern shows a structural local hazard, not exposure-driven randomness.

348502B Bartlett TN CSX 9 accidents @ 1 train/day = 4,110/million · 352618J Bessemer AL 5 @ 1 = 2,283 · 637912X Midland City AL 4 @ 1 = 1,826 · 923252B Jersey City NJ Norfolk Southern 14 @ 4 = 1,598 · 743643X Jersey Village TX 7 @ 2 = 1,598 · 624729E Memphis FL 3 @ 1 = 1,370 · 279669W Conyers GA 3 @ 1 = 1,370 · 301813U Bogalusa LA 3 @ 1 = 1,370 · 352622Y Bessemer AL 3 @ 1 = 1,370 · 152495X Aurora IN 3 @ 1 = 1,370

025422P Glendale AZ BNSF 24 accidents (12 trains/day, rate 913/million) · 025617C Phoenix AZ BNSF 20 (12, 761) · 025430G Phoenix AZ BNSF 15 (12, 571) · 923252B Jersey City NJ Norfolk Southern 14 (4, 1,598) · 763630B Midland TX UP 12 (21, 261)

13,011 grade-crossing accidents in 2020-2025 across 10,146 distinct crossings · 8,833 crossings with non-zero daily-train inventory data · 11,409 joinable accidents · 345.4 million joinable train movements · system mean rate 33.04 accidents per million train movements · 486 crossings with ≥3 accidents in the period · 6 of top 10 per-rate crossings have only 1 train per day