How to read a set of auction results

Sale prices are skewed: a handful of exceptional cars sit far above everything else, and an average is dragged up by them. The median — the middle sale — is not. On this site the average of the Porsche 356 set is $166,417 against a median of $120,000, a gap of $46,417 created entirely by the top of the distribution.

That gap is the first thing to check in any set of results. If the average sits well above the median, the set contains cars that are not representative, and a figure quoted from the average is describing them.

The middle half of these 1,146 sales fell between $75,000 and $208,550. That band is the useful part of the answer: it says where a normal car lands, and it is wide on purpose, because the market is wide.

The band is also why a single median hides the most useful split. Within this set, coupe cars sell at a median of $84,000 on 612 sales, convertible cars sell at a median of $177,800 on 273 sales, roadster cars sell at a median of $280,000 on 104 sales. One median across all of them describes none of them well.

A median of three sales is a story about three cars. The threshold on this site is 20 cleaned sales before a median is published at all, and every page prints its count next to its median so the reader can judge it.

Counts also tell you what a set is missing. On Porsche 356, 408 of 1,554 pulled rows were removed — 26% — most of them sales not made in dollars. A set that has quietly discarded a third of its rows is not describing the whole market, and the page should say so.

The highest sale in this set is recorded on the page rather than folded into the median. 4 sales above $1,200,000 are held out of the median and the ranges, and the highest — $3,100,000 for a 1961 Porsche RS61 — is stated openly.

An analysis that silently drops the top and bottom of a distribution is doing something reasonable and hiding it. The test is whether the exclusions are stated with their rule and their count, so a reader can disagree with the rule.

On the Porsche 356 set: 1,146 sales, median $120,000, typical range $75,000 to $208,550, average $166,417, 4 high sales set aside and disclosed, 26% of raw rows removed with each rule's count published. Every one of those figures is on the model page, each sale is linked to its auction record, and the whole thing is reproducible.

That is the standard to hold any set of auction results to, ours included: a median with its count, a band that admits its width, exclusions stated with their rule, and a date on the figure. Our method is here, and the model pages apply it to 28 models.