Are car recalls voluntary, or does the government force them?
Key facts
- The share of recall campaigns NHTSA had to force fell from 26.63% in 2013 to 1.93% in 2022 - a 93% decline - and then tripled back to 13.97% in 2026.
- Counting vehicles instead of campaigns shows no such collapse. The share of recalled units under an agency-forced campaign was 35% to 55% in most years, and its 2022 low of 9.51% is a single-year trough, not a decade-long decline.
- The reconciliation is size. The median agency-forced campaign covered 259 vehicles in 2010-2014, 51,465 in 2015-2020 and 48,667 in 2021-2023. The agency did not stop forcing recalls; it stopped forcing small ones.
- Manufacturer-initiated recalls are the control, and they never moved. Their median size ran 424, 308, 352 and 321 vehicles across the same four eras. A file-wide recording change would have moved both series; only one moved.
- In 2010-2014, 40.2% of agency-forced campaigns covered fewer than 100 vehicles. By 2021-2023 that was 4.5%. The manufacturer-initiated comparison held flat at 28.4%, 34.6%, 33.6% and 34.1%.
- The post-2024 rebound is broad, not one carmaker: 324 forced campaigns across 120 distinct filers, the largest holding 14.8%.
- No manufacturer ranking is published here. It reproduces at only 0.562 between the 2015-2020 and 2021-2026 halves and is reported as refused.
Which measure is right - the campaign count or the vehicle count?
Both are correct arithmetic on the same file, and they answer different questions. The campaign count asks how often the agency had to intervene. The vehicle count asks how much of the recalled fleet its interventions covered. They diverge whenever the typical size of a forced recall changes - which is exactly what happened.
| Era | Campaigns | Agency-forced | Forced share | Median units, forced | Median units, manufacturer | Forced campaigns under 100 units |
|---|---|---|---|---|---|---|
| 2010-2014 | 3,224 | 704 | 21.84% | 259 | 424 | 40.2% |
| 2015-2020 | 5,166 | 341 | 6.60% | 51,465 | 308 | 8.5% |
| 2021-2023 | 2,815 | 89 | 3.16% | 48,667 | 352 | 4.5% |
| 2024-2026 | 2,471 | 324 | 13.11% | 5,204 | 321 | 21.9% |
Read the last two columns together. The forced-campaign median rises roughly 200-fold between the first two eras while the manufacturer-initiated median barely moves. That is the whole finding in two numbers.
Could this just be NHTSA changing how it fills in the form?
This is the check most likely to destroy the result, so it is worth being exact. If the agency had simply changed its recording practice in 2015 - say, by no longer tagging small campaigns as agency-influenced - the forced series would shift and the manufacturer series would not. That is precisely the pattern observed, so the recording-change hypothesis cannot be dismissed on the size data alone.
What argues against it is the internal control. The manufacturer-initiated campaigns are drawn from the same file, the same years and the same filers, and their size distribution is flat across all four eras (median 424, 308, 352, 321; under-100 share 28.4%, 34.6%, 33.6%, 34.1%). A clerical change to how the file is written would be expected to disturb the composition of the file as a whole. It did not. The change is confined to the one category whose definition is a decision by the agency.
We state the limit rather than resolving it: this analysis establishes that the forced-recall count and the forced-recall volume diverged, and that the divergence is specific to agency-initiated campaigns. It does not establish the agency's internal reason, which is not in the file.
Is the collapse just the Takata airbag recalls?
No, and this was the second check most likely to reverse it. Agency-forced campaigns in 2015-2020 are 56.3% air bags, overwhelmingly the Takata inflator wave, so the obvious worry is that one campaign family produces the entire size shift.
Removing every air-bag campaign from both sides leaves the finding intact:
| Era | Median units, forced (ex-airbag) | Forced under 100 units (ex-airbag) | Median units, manufacturer (ex-airbag) |
|---|---|---|---|
| 2010-2014 | 227 | 41.3% | 400 |
| 2015-2020 | 7,810 | 16.8% | 294 |
| 2021-2023 | 48,400 | 5.6% | 332 |
| 2024-2026 | 4,754 | 23.1% | 313 |
The 2021-2023 era is the decisive one: air bags are only 20.2% of forced campaigns there, and the median forced campaign is still 48,400 vehicles against a manufacturer-initiated median of 332.
What changed after 2024?
The composition did, completely. Agency-forced campaigns in 2015-2020 were led by air bags (56.3%). In 2024-2026 the leaders are electrical system (61 campaigns), equipment (43), structure (32) and back-over prevention (25), with air bags down to 5.2%.
The rebound is also unusually broad: 324 forced campaigns across 120 distinct Part 573 filers, with the largest single filer at 14.8% and the top three at 25.9%. And the forced campaigns are small again - median 5,204 units, with 21.9% under 100 vehicles, closer to the pre-2015 pattern than to the Takata decade.
Can you rank carmakers by how often NHTSA forces their hand?
No, and we computed the ranking before refusing it. Split between campaigns filed in 2015-2020 and those filed in 2021-2026, the manufacturer ordering reproduces at a tie-corrected Spearman of only 0.562 across 27 filers with at least 20 campaigns each.
A random split-half of the same data returns 0.788 - noticeably better, and misleading, because it tests only whether the sample is large enough rather than whether the ordering is stable over time. The temporal split is the honest test and it is the one reported. The same trap flattered a make-level ranking in this desk's work on 13 October.
So the per-manufacturer shares exist in our working data and are deliberately not printed as a league table.
What a VIN check can and cannot tell you here
This analysis is about campaigns, not about individual cars, and the distinction matters.
It can tell you what a recall record means. A recall on a used car is not evidence that a regulator forced the issue; nearly nine in ten - 89.34% - are filed by the manufacturer. Equally, a manufacturer filing is not evidence that nobody pushed - the field records paperwork, not pressure.
A Zilocar VIN check returns what has been reported on one specific vehicle: accident and damage records including airbag deployment, odometer history and rollback indicators, salvage-auction and theft records, ownership and sales-listing history, and any recall present on that car.
It cannot tell you who initiated that recall. The initiator lives on the Part 573 filing, which is keyed to a campaign, not to a VIN. A VIN check also cannot show a recall's remedy status, an open NHTSA investigation, or per-unit firmware. For whether a specific recall on a specific car has actually been repaired, use NHTSA's own VIN lookup at nhtsa.gov/recalls or a franchised dealer.
What this analysis cannot claim
Four things, stated plainly. It is not a measure of enforcement effort: INFLUENCED_BY records who is named on the filing, and informal agency pressure that ends in a manufacturer filing is invisible here by construction. It is not a defect rate - POTAFF is the manufacturer's launch-time estimate of units potentially affected, not cars on the road, still unrepaired, or still in the US. It is not a manufacturer ranking, which failed cross-validation above. And 2026 is a partial year, closing at the file's newest report-received date of 8 October.
Method
Source. NHTSA ODI Part 573 recall flat file FLAT_RCL_POST_2010.txt, from https://static.nhtsa.gov/odi/ffdd/rcl/FLAT_RCL_POST_2010.zip, public domain, downloaded 2026-10-14 07:19 UTC. The server reported Last-Modified: Wed, 14 Oct 2026 07:04:51 GMT; the archive passed a CRC check; the extracted file carries 245,845 rows.
Freshness assert, run before anything was computed. The newest Part 573 report-received date (RCDATE) in the file is 2026-10-08, six days before retrieval.
Universe. 245,845 rows → restricted to RCLTYPECD = V (vehicles, as opposed to tyres, equipment and child restraints) → collapsed to one record per CAMPNO, because the file carries one row per make/model/year within a campaign while the initiator, the unit count and the filing date are properties of the campaign. That gives 13,676 distinct vehicle campaigns, all of which fall in the 2010-2026 band.
The field. · INFLUENCED_BY (field 14) takes three values. Across the 13,676 vehicle campaigns analysed here it reads MFR on 12,218 (89.34%), ODI on 1,350 (9.87%) and OVSC on 108 (0.79%), with no blanks; across all record types in the file, including tyres, equipment and child restraints, the counts are 13,581 / 1,453 / 270 with exactly one blank. ODI is NHTSA's Office of Defects Investigation and OVSC its Office of Vehicle Safety Compliance; both are counted here as agency-forced, and the split is reported separately per era because collapsing them would hide a compliance-versus-defect distinction.
Definitions. Agency-forced share is the proportion of a year's campaigns with INFLUENCED_BY in (ODI, OVSC). The units basis uses POTAFF, summed per campaign. The size comparison uses medians rather than means because the distribution is dominated by a handful of multi-million-unit campaigns.
The internal control, and why it is the load-bearing check. Manufacturer-initiated campaigns from the same file and the same years are reported alongside every size statistic. They are what distinguishes a change in agency behaviour from a change in how the file is written: a clerical rewrite would be expected to move both series, and only one moves.
Robustness, four checks, two of which bite. (1) Removing all air-bag campaigns - the Takata wave is 56.3% of forced campaigns in 2015-2020 - leaves the size shift intact, with the 2021-2023 ex-airbag median at 48,400 against a manufacturer median of 332. (2) The manufacturer ordering fails a temporal split at 0.562 over 27 filers and is refused in print; a random split-half would have returned 0.788 and flattered it into publication. (3) The 2024-2026 rebound is spread over 120 distinct filers with a 14.8% maximum, so it is not one manufacturer's defect wave. (4) Approximately 20.4% of the 2024-2026 forced campaigns were filed by school-bus, motorhome and heavy-truck manufacturers rather than passenger-car makers; that figure comes from a keyword match on filer names and is therefore approximate, and it is a reason to read the car-buyer relevance of the rebound conservatively.
Rank statistics are tie-corrected. Tied values take the average rank, because an uncorrected rank function returns a figure that depends on input order.
Limits. No denominator of vehicles sold, so no figure here is a per-vehicle rate. POTAFF is a launch-time estimate. INFLUENCED_BY records the filing, not the pressure behind it. 2026 is partial, closing 2026-10-08.
Frozen dataset. The full per-year series - campaign counts, forced counts and shares, unit totals, forced-unit shares, both medians and both under-100 shares - is published at nhtsa-initiated-recall-count-vs-size-2026-10-14.csv, 17 rows plus a header.
