What I did
The data
Three free files from NHTSA, the US road safety agency: owner complaints, safety recalls, and its defect investigations.
What gets ranked
Every vehicle and part that owners complained about in the past year, for example "2023 Honda CR-V, steering". That is about 22,000 vehicles and parts each month.
The method: logistic regression
For each vehicle and part I collect 73 facts known at the end of the month: complaint counts and growth, crash and fire reports, vehicle age, recent recalls and open investigations. Logistic regression learned a weight for each fact from 2015 to mid-2021, where the outcomes are known. It adds the weighted facts into one score: the estimated chance of a recall in the next six months.
The test
For each of 36 months, January 2023 to December 2025, I sort every vehicle and part by its score. The top 100 is the watch list. Then I count how many on it were really recalled in the following six months.
Result: 17.3% of the logistic regression watch list was recalled within six months
I ran the same test on three ways of choosing 100 vehicles and parts.
The same test for six ways of building the list
Share of the monthly top 100 recalled within six months. The dot is the average over 36 months. The line is the margin of error.
What I found
- The logistic regression watch list beat the complaint-count list in all 36 months when the weights were updated every month.
- Recall and investigation facts help more than complaint facts. A vehicle's and maker's past recalls predict the next recall better than complaints do.
- Complaint facts matter most for vehicles with no recent recall, where there is no recall history to go on.
- Adding the text of the complaints did not help. It made the watch list slightly worse.
- Most of the watch list is not recalled. About of every 100 vehicles and parts on it are not recalled within six months.
The watch list, month by month
Pick a month. The watch list for that month is the vehicles and parts with the highest score at the end of the month, using only information available then. The last column shows what happened afterwards.
Score is the estimated chance of a recall in the next six months, calculated by logistic regression from the 73 facts. The list is sorted by this score.
| Rank | Score | Vehicle | Part | Complaints in the past 12 months | Rank by complaint count | What happened next |
|---|
No rows match these filters.
Select a row to open that vehicle in the look-up tab.
Look up a vehicle
Covers the vehicles and parts that were on a top-250 watch list, were recalled in 2023 to 2025, or had 10 or more complaints in a year.
Try one of the cases from the report: , , .
Complaints received per month
Dashed lines mark the month a recall was filed.
Rank each month, by score and by complaint count
Rank 1 is the top. Above the dotted line means it was on the top-250 watch list. Gaps are months when this vehicle and part was not ranked, for example the 12 months after its own recall.
Recalls on this vehicle and part
I tested the logistic regression watch list on every month from January 2023 to December 2025
For each of the 36 months I built the watch list using only information available at that time. Then I checked the real recall records for the following six months to see which listed vehicles were recalled.
How many on each month's watch list were recalled within six months
Each point is one month. The blue line is the logistic regression watch list. The grey line is the complaint-count list. The numbers swing because one recall can cover many model years at once.
Which information helps the most
Share of the monthly top 100 recalled within six months when the list is built from different information.
How many months of warning the watch list gave
For each recalled vehicle and part that was on the top-250 watch list: how many months in a row it had been on the watch list before the recall was filed. Recalls filed in 2024 and 2025.
Two extra checks
Open investigations and recent recalls are the warning signs that matter most
I rank the warning signs by how much worse the watch list gets when that kind of information is taken away.
Share of the watch list recalled within six months when one kind of information is removed
Each bar shows the result when one kind of information is taken away. Taking away the investigation facts or the recall facts hurts the most. Taking away the complaint counts, or the facts about what share of a vehicle's complaints are about one part, changes almost nothing.
What raises or lowers the score
The 14 facts with the largest weights. A bar to the right means the fact raises the score. A bar to the left means it lowers the score. Complaint counts are measured over several overlapping periods, so some point left and some right. Read those together.
How each part changes the score
A bar to the right means that part gets a higher score, all else equal. Hover over a bar to see how often the part was recalled in 2015 to mid-2021, when the weights were learned, and in 2023 to 2025. Some rates changed a lot, so these weights need updating over time.
Three cases
What the watch list cannot do, and how it was tested
Limits
- Most vehicles and parts on the watch list are not recalled. The watch list is a way to decide where to look first.
- About 70% of recalls have no earlier complaint on the same vehicle and part, so no method based on complaints can reach them. The other 30% cover 94% of recalled vehicles.
- The watch list leans on a maker's recall history. Makers that recall often get a high score even with few complaints.
- Parts are matched only by broad category. Complaints and recalls are linked using 27 part categories, such as "steering" or "electrical system".
- I had no sales numbers. So I cannot tell whether a vehicle has many complaints because of a real problem or because many of them were sold.
- I looked at the 2023 to 2025 results while still building this. To check I had not adjusted things to fit those years, I tested one unused month, January 2026, at the very end.
How it was tested
- Forward in time only. The weights are learned from the past and the watch list is checked against the future.
- No look into the future. Every fact for a month uses only records dated up to that month.
- A pause after a recall. Complaints jump once a recall is announced, so a vehicle and part is not ranked for 12 months after its own recall.
- Why test forward in time?
Complaints rise before a recall and jump after it
Average complaints per month for a recalled vehicle and part. Month 0 is the month the recall was filed.
Linking recalls to earlier complaints
Share of 2015 to 2025 recalls with at least one earlier complaint on the same vehicle and part.
Data sources
Three free files from the NHTSA datasets page, downloaded November 19, 2026: owner complaints, safety recalls since 2010, and defect investigations. The method is logistic regression with 73 facts. I also tested random forest, gradient boosting and LightGBM.