Methodology · version 1.3 · snapshot 2026-07-19
How one number gets earned
This build grouped 2,155,302 ODI vehicle component rows into 398,146 deduplicated matched complaint incidents. Here is the pipeline, the math, and the limits.
Vehicle component rows and deduplicated complaint incidents are different populations. Neither is a fleet count. Rail lengths are illustrative because the stages use different units.
The pipeline · five stages, no black box
From raw record to index
Snapshot input
IngestThe 2026-07-19 NHTSA ODI snapshot contains 2,155,302 vehicle component rows.
Make/model normalization and incident deduplication
Normalize537,783 matched component rows collapse into 398,146 deduplicated complaint incidents associated with covered vehicles.
Same segment and era
CohortEach of 874 model years is compared only inside one of 127 same-segment, same-era cohorts.
Hazen midrank percentile
ScoreThe 0 to 10 signal index is a Hazen midrank percentile, a rank that places each model year between its peers, using the survival-weighted age-adjusted complaint rate within the cohort. It is not sales-normalized.
Whole interval must clear 8.0
FlagA deterministic parametric bootstrap, a repeated simulation of the expected complaint counts, uses 2,000 resamples. 143 of 874 records have an entire 95% interval at or above 8.0.
Separately, the estimated per-1,000 trend divides cohort-lifetime complaints by estimated vehicle-years. Vehicle-years use compiled US sales, a 0.25 prior-calendar-year plus 0.75 current-calendar-year allocation, and the shipped survival curve. Rates appear only when estimated VIO is at least 5,000. Complaint counts are NHTSA reports, not all failures.
The math · honestly stated
One formula, no thumb on the scale
The numerator
Deduplicated matched complaint incidents for the model-year. These are reports submitted to NHTSA, not all mechanical failures.
The trend denominator
Estimated vehicle-years use compiled US sales, a documented model-year allocation, and the shipped survival curve. Registration counts are validation anchors, not inputs to the curve. This denominator powers the separate per-1,000 trend and does not change the complaint-signal ranking.
The interval
A bootstrapped 95% range from 2,000 resamples. Wide interval, quiet claim. If it cannot be bounded, it does not ship.
There is no editorial weighting and no manual override. The score comes from the model-year complaint rate within its comparison cohort, with uncertainty carried into the published interval.
Built in the open
Check the work yourself
Every number on this site can be rebuilt from the public NHTSA ODI flat files using the five stages above. No private data, no scores for sale, no black box. When the method changes, the version number changes with it.
The fine print · set in large type
What the index is not
It measures complaint intensity against a same-segment, same-era cohort. A loud problem and a common problem can score alike. Every record links to the published method and the official NHTSA source dataset; MotorCensus does not republish individual complaint narratives.
Complaint counts are owner reports, not confirmed defects. NHTSA reviews each complaint with other evidence and uses no fixed complaint count to open an investigation. MotorCensus does not conduct defect investigations.
Every published index appears with its sample size and 95% interval. A displayed interval may round to the same tenth at both ends; that is still a valid bounded interval, not a missing value.
731 published model-years · free · no login
Read the method and open the record
Browse NHTSA complaint-signal records with visible sample sizes, intervals, and limitations. The current data snapshot is dated 2026-07-19.