Values from real sales
MethodDocument M

How we value a car

Every number on this site is a percentile of completed auction sales, computed after nine cleaning rules that removed 4,342 of 32,418 rows across 28 models. This page states the rules, how many rows each removed, and what the resulting figure does not cover.

Sections
6

§ M.01Where the numbers come from

The source is BidBetter's own auction sales database: completed results from 19 auction houses, 302,360 sold lots from 1 January 2016 to 1 November 2026. Only rows cleared for publication are read; anything else is excluded before any statistic is computed.

Each model page is restricted to one make, one model string (or the several strings a generation is stored under) and a year window. The window is chosen to match the generation, not to flatter the median: where a generation has no string of its own in the data, the year range does the work.

§ M.02The nine cleaning rules

Rules run in this order. The first four remove rows that are not sales or not prices; the fifth and sixth remove rows that are not the car; the last three remove rows that cannot be a price for this car. Across the 28 models on this site, 4,342 of 32,418 pulled rows were removed, leaving 28,076.

Rows removed by rule, across every model on this site. A row is counted once, under the first rule that catches it.
RuleRowsShare of removals
Not sold in US dollars
GBP, EUR, AUD, CHF, CAD, NZD and HKD sales are excluded, so every figure on this site describes the dollar market.
3,89789.8%
No usable date
A missing sale date, or one dated in the future.
431.0%
Not a vehicle
Parts, memorabilia, literature, trailers and scale models. A parts word only drops a row when the title carries no chassis number and no trim or body-style word, because auction houses write real cars as "Chassis no. 30837S104422 Engine no. 3104422".
611.4%
Replica or kit car
A fiberglass Speedster replica on a VW pan is not the car it copies. This rule matters most on the Porsche 356, where 164 of 1,339 in-window USD sales were replicas selling at a median of $43,000 against $118,000 for the real car.
1693.9%
Below the plausibility floor
Under 5% of the model's median. A $22 "911" is a data error, not a cheap car.
421.0%
Above the ceiling
Over 10× the model's median. These are real sales set aside rather than deleted, and the highest is disclosed on the page.
1303.0%

Two of these rules are worth calling out because they changed published figures. The replica rule was added after measuring the Porsche 356: leaving replicas in published a median of $99,999 for a car that really sells at $118,000. The ceiling rule is asymmetric on purpose — a $4,295,000 1976 Porsche 935 is a real sale of a real car, so it is held out of the median and disclosed rather than deleted.

§ M.03Percentiles, medians and bands

The median is the 50th percentile of the cleaned sale prices. The typical range is the 25th to 75th percentile: half of all sales fall inside it. Percentiles are linearly interpolated on the sorted prices, which is the same calculation PostgreSQL's percentile_cont performs, so a figure can be reproduced straight from SQL.

A model publishes a median only at 20 or more cleaned sales, and nothing publishes below five. Below 20 the page would be describing a handful of cars rather than a market.

Each page also shows the full observed range, the cheapest and dearest sale that survived cleaning. That range is set by outliers, so it is a boundary rather than a guide to what most cars bring.

§ M.04What these figures do not cover

Dollars only. Sales in other currencies are excluded. On some models that is a large share of the market — the Porsche 964, the Mercedes 190SL and the BMW E30 M3 trade heavily in Britain and Europe — so a dollar median is a median of the dollar sales, which is a narrower claim than "what the car is worth".

Model and year, not car. A page describes a model. It does not know about this car's engine, paint, paperwork or accident history, and no model-level figure can.

Mileage is not used to adjust. Only 6.6% of these sales carry an odometer reading with a unit attached. Where it is recorded it tracks price clearly (see mileage and value), but it cannot be applied to the other 93% without inventing a number.

Not an appraisal. This is a description of a market, not an inspection of a car, and it does not predict an auction result. See the disclaimer.

§ M.05Reproduce a figure

A model's median can be recomputed from the published sales without trusting this page: take the model's sales table, drop anything not sold in dollars, sort the prices and take the middle value. The page lists its most recent 15 sales with a link to each auction record, the model-year split, the body-style split and the transmission split, so the arithmetic is checkable against the record.

The dataset behind all of it is described on the dataset page, and the question pages are indexed from model values.

§ M.06Questions

Is a BidBetter value an appraisal?
How many sales back a figure?
Why is the median different from what I was offered?