PulsCar analysis of NHTSA public data, September 2026 Topic: steering that is hard to turn / loss of power steering assist License: CC BY 4.0 (attribution: "PulsCar analysis of NHTSA public data, September 2026") ------------------------------------------------------------------ 1. SOURCE FILES ------------------------------------------------------------------ All inputs are NHTSA public flat files, downloaded 17 September 2026 and loaded into a local SQLite database: - COMPLAINTS (FLAT_CMPL) - owner complaints (Vehicle Owner Questionnaires) - RECALLS (FLAT_RCL) - Part 573 safety recall records - TSBS (FLAT_TSB) - manufacturer communications index - INVESTIGATIONS (FLAT_INV) - ODI investigation records Field definitions: https://static.nhtsa.gov/odi/ffdd/cmpl/CMPL.txt Complaint rows in the database: 777,040 unique reports, received 1 January 2015 through 8 September 2026. That receipt window is the denominator for every percentage below. The failure itself can pre-date the report, so model years older than 2015 appear normally. ------------------------------------------------------------------ 2. THE COMPLAINT QUERY ------------------------------------------------------------------ FTS5 match on the free-text description: ("power steering" OR "steering assist" OR "steering wheel" OR steering) AND ("hard to turn" OR "difficult to turn" OR "stiff" OR "locked up" OR "loss of power steering" OR "lost power steering" OR "no power steering" OR "hard to steer" OR "difficult to steer") Raw matches: 15,895. Then kept only rows whose NHTSA component string contains "STEERING": final working set = 13,618 complaints. The component filter is the only exclusion applied. It removed 2,277 rows (14.3%) in which the phrase appeared but NHTSA had not coded the report to the steering system at all. FALSE-POSITIVE READ. Before publishing any count we read 30 randomly selected matches in full (seed 11). Twenty-seven were on topic: the owner describes heavy, stiff or unassisted steering. Three were marginal and were left in with this note: one salvage-title vehicle whose primary failure was the transmission, one multi-system failure (engine, brakes and steering together), and one engine stall in which assist was lost because the engine stopped. Observed false-positive rate: 3 of 30 (10%), and all three of them do in fact report heavy steering - they simply are not steering-system defects. We did not tighten the query further, because every narrower variant we tried dropped genuine short reports such as "Steering Sticks and Very Stiff". ------------------------------------------------------------------ 3. WHAT WE COMPUTED ------------------------------------------------------------------ hard-steering-complaints-by-make.csv Complaint count, share of the 13,618, median failure mileage, the size of the mileage sample, share towed, share alleging a crash, and the share of reports whose text names EPS / electric power steering. Top 40 makes. hard-steering-top-models.csv Same set by make+model: count, median / first-quartile / third-quartile failure mileage, the three most common model years, share towed. Top 40. hard-steering-by-model-year.csv Complaints by vehicle model year, 1996-2026. hard-steering-owner-reported-costs.csv Distribution of dollar figures owners typed into their own complaints. hard-steering-eps-recalls.csv Every recall campaign NHTSA coded to component "STEERING:ELECTRIC POWER ASSIST SYSTEM": 93 distinct campaigns, campaign number, report date, manufacturer, makes, model-year span, potentially affected population, our remedy classification and the NHTSA record URL. MILEAGE. Rows with miles <= 0 or >= 400,000 were dropped as unusable. 9,211 of 13,618 complaints (67.6%) carry a usable mileage. Median 91,000; quartiles 57,000 and 130,000; 26.6% below 60,000 miles. TOWED / CRASH. Computed from the flat file's TOWED and CRASH flags on the same 13,618 rows, against the all-complaints baseline of 777,040 rows. hard-steering set: 5.5% towed, 3.3% crash alleged, 0.3% fire all complaints: 5.1% towed, 4.4% crash alleged KEYWORD SHARES (case-insensitive regex on the description, same 13,618 rows): warning light or dash message named ......... 4,973 (36.5%) mentions a recall ........................... 4,255 (31.2%) "while driving" ............................. 5,161 (37.9%) towing mentioned in the text ................ 1,738 (12.8%) no code found / could not be duplicated ..... 1,594 (11.7%) names EPS or electric power steering ........ 1,301 ( 9.6%) names the rack / steering gear .............. 1,876 (13.8%) names the column or intermediate shaft ...... 626 ( 4.6%) names the pump .............................. 462 ( 3.4%) names fluid ................................. 455 ( 3.3%) names a belt ................................ 158 ( 1.2%) WHAT THE REPAIR WAS. 5,014 of the 13,618 reports (36.8%) mention a repair. Only 1,647 of those 5,014 (32.8%) name the part. Among the ones that do: steering rack / steering gear ............... 940 (57.1% of named) steering column / intermediate shaft ........ 241 (14.6%) EPS motor, module or torque sensor .......... 174 (10.6%) pump ........................................ 129 ( 7.8%) battery, alternator, fuse, ground, wiring .... 99 ( 6.0%) hose, line, fluid, reservoir ................. 41 ( 2.5%) tie rod, ball joint, control arm ............. 14 ( 0.8%) belt ......................................... 9 ( 0.5%) This is a distribution of what owners wrote down, not of what fails. The categorization is a regex over the description; a complaint is assigned to the first category it matches, in the order listed in the script. OWNER-REPORTED DOLLAR FIGURES. Every "$N" in the description was extracted, kept only when 100 <= N <= 10,000 and when the 280-character window around it contained both a steering word (steering / rack / pinion / column / EPS / gearbox / torque sensor) and a money word (repair, fix, replace, quote, estimate, cost, charge, paid, bill, price), and did not contain a disqualifying word (damage, totaled, air filter, headlight, brake pad, tire, A/C, transmission replacement, alignment). One figure per complaint, the largest. Result: 678 complaints. median $2,000; quartiles $1,500 and $2,600; 85.0% at or above $1,000 FALSE-POSITIVE READ: we read 30 of the 678 windows in full (seed 3). Two were wrong - a $139 diagnostic fee and a $10,000 total that covered fuel pumps as well as steering. Observed rate 2 of 30 (7%). LIMITATION, and it is the big one: people file NHTSA complaints when a repair is expensive or refused. A $15 fuse almost never generates a complaint. These figures describe the repairs owners complained about, NOT the price of fixing heavy steering. ------------------------------------------------------------------ 4. RECALLS ------------------------------------------------------------------ Component filter: component LIKE '%STEERING:ELECTRIC POWER ASSIST%'. 93 distinct campaigns; the sum of the per-campaign "potentially affected" figures is 8,743,313 vehicles. That sum double-counts any vehicle covered by more than one campaign, so it is an upper bound on vehicles, not a vehicle count. Remedy classification (our reading of the NHTSA remedy text, not NHTSA's): "software/reprogram" when the remedy text says software, reprogram or update and does not say replace in its first 200 characters - 20 campaigns; "inspect/replace" when it says replace - 55 campaigns; "other" - 18. All steering components together (component LIKE '%STEERING%'): 796 distinct campaigns in the file. ------------------------------------------------------------------ 5. BULLETINS AND INVESTIGATIONS ------------------------------------------------------------------ TSB FTS query: ("power steering" OR "steering effort" OR "electric power steering" OR EPS) AND ("hard to turn" OR stiff OR heavy OR effort OR notchy OR "sticky") 132 distinct NHTSA bulletin IDs. By make, the largest counts are Chevrolet 26, Ford 19, GMC 15, Ram 14, Jeep 13. Investigations: ODI actions whose component contains STEERING and whose subject or summary names power steering or steering assist. Open at the data cut: PE26-002 (2022-2023 Ram ProMaster, opened 2 March 2026, 19 owner reports), PE25-009 (2017-2018 Chrysler Pacifica, opened 5 September 2025, 94 owner reports), RQ23-003 (2013-2016 Ram 1500, opened 4 August 2023, 380 reports). Closed: EA17-004 (2010 Ford Fusion, opened as PE16-011 on 28 September 2016, upgraded 20 November 2017, closed 9 February 2024 with no safety-defect trend identified). Primary documents opened and read for this article: Honda service bulletin 19-036, 11 July 2019 (NHTSA ID 10162889) https://static.nhtsa.gov/odi/tsbs/2019/MC-10162889-0001.pdf Acura service bulletin 19-021 (NHTSA ID 10163140) https://static.nhtsa.gov/odi/tsbs/2019/MC-10163140-0001.pdf GM bulletin 16-NA-007, March 2017 (NHTSA ID 10138021) https://static.nhtsa.gov/odi/tsbs/2017/MC-10138021-9999.pdf Ford SSM 51255 (NHTSA ID 10228872) https://static.nhtsa.gov/odi/tsbs/2022/MC-10228872-0001.pdf BMW bulletin B320319, April 2019 (NHTSA ID 10159122) https://static.nhtsa.gov/odi/tsbs/2019/MC-10159122-9999.pdf Recall records read through the NHTSA recalls API, e.g. https://api.nhtsa.gov/recalls/campaignNumber?campaignNumber=15V340 ------------------------------------------------------------------ 6. LIMITATIONS ------------------------------------------------------------------ - Complaints are self-reported, unverified and unweighted. They are not failure rates. We do not know how many of each model are on the road, so a larger count for a high-volume model is expected and says nothing about reliability. Nothing here supports "make X is the most/least reliable". - Publicity drives complaints. A recall or a news story raises the count for that model without changing what breaks. - Free text is inconsistent. Every text-derived share above is a floor: an owner who does not write the word "fuse" may still have had a fuse. - Mileage is present on 67.6% of rows and is the owner's estimate. - Samples smaller than roughly 100 complaints are reported as small samples and should not be read as model-level rates. - www.nhtsa.gov blocks automated requests (HTTP 403), so recall records are linked through api.nhtsa.gov, which serves the same NHTSA record.