Out of 60 advertisements noted on August 30, 2026, only one weighs 47% of the total sum of the asking prices.Removing it drops the average from €21.79 to €11.73, a fall of half - while the median moves from €9.88 to €9.86, that is to say by two cents. And this unique announcement does not concern a booklet: it is a bound work in the German language. The top of a distribution of offers is not “the top of the market”, it is very often an object which does not belong to the corpus that we thought we were measuring.
The articles in this series have talked a lot about the form of ads. This speaks of the form of their distribution - that is to say of the way in which the prices asked are distributed, and of what this distribution allows or prohibits to conclude.
The question is not academic. A collector who wants to know “how much is worth” a set looks at aggregated figures: an average, a maximum, sometimes a range. These three indicators behave very differently depending on the shape of the distribution, and two of them allow themselves to be overturned by a single line.
What a single line weighs
I surveyed the French market on August 30, 2026 and retained 60 advertisements after discarding the lots. The amounts aresums claimedthat day: no sales are observed, and an observation of a single day does not measure any development. What follows does not relate to the level of prices, but to their distribution.
The median is €9.88. Fifty-two percent of ads are under €10, eighty-five percent under €20. The maximum reaches €615.85 – or sixty-two times the median, and ten times the second most expensive ad in the survey.
This maximum is not a booklet. It is a bound work, published in German. It appears in the results because it has the right words, and it alone brings 47% of the sum of all the prices asked in the statement. One line out of sixty, and almost half of the total.
The average collapses by half
€21.79 with this ad, €11.73 without it. An indicator that a single line halves doesn't describe a market, it describes that line.
The median does not move
€9.88 with, €9.86 without. Two cents difference. Removing even the three most expensive listings only brings it down to €9.10.
The maximum is a foreign object to the corpus
A bound book in German, in a statement intended to describe fascicles. It says nothing about the price of a booklet, but it governs the range displayed.
A sudden dropout
€615.85 for the first, €61.85 for the second, €45.53 for the third. A factor of ten between the first two: it's not a slope, it's a break.
This last point is the simplest test to apply, and I recommend it before reading any reading. Compare the first value to the second. If the gap between them is of the same order as that between the second and third, the distribution has a regular slope and its peak is probably composed of comparable objects. If the first drops by a factor of ten, it belongs to another population, and the question is no longer “why is it so expensive” but “what is it”.
We must measure what this implies for any price range displayed somewhere. A range “from €2 to €616” is arithmetically exact and practically misleading: it suggests a continuum between the two limits, while 85% of the advertisements are below €20 and a gap separates the top from the rest.
A fork only has meaning if you know what it contains. Without the number of observations and without the median next to it, two limits provide no information about anything.
The same measurement on two other characters
To find out if this behavior is specific to this character or general, I applied the same measurement to two other readings taken on the same day. The results are instructive because they diverge, while the medians are very similar.
First comparison: 56 ads, median €9.38, maximum €320.35. The ratio between the two is 34. The most expensive ad weighs 28% of the sum, and it is a Japanese illustrated book from 1979. The second most expensive is an embossed edition accompanied by glasses, the third a work in Italian. The top of this distribution is composed, again, of objects which are not fascicles.
Second comparison: 42 ads, median €8.00, maximum €55.00. The ratio drops to 7. The most expensive ad weighs only 10% of the sum and is worth 1.2 times the second. And above all, the six most expensive announcements are all booklets consistent with the corpus. This distribution is healthy: its peak really describes the top of its market.
Three very close medians — €9.88, €9.38, €8.00 — for three radically different forms of distribution. The median alone does not distinguish them. What distinguishes them is the relationship between the maximum and the median, and the nature of the objects which occupy the vertex.
There is a basic reason why these non-corpus objects are located precisely at the top, and not elsewhere in the distribution. An ordinary booklet sells for a few euros because the market offers a lot of them; a bound work in a foreign language, a particular edition, a derived object do not have this point of comparison. Their price is set by a seller who has no neighbors, and nothing attracts him to the rest. They therefore land mechanically at the top of the table, regardless of the character questioned.
This is why checking the summit is not an occasional check but a reflex to be applied systematically. The top of a distribution of offers is the place where, by construction, objects which have no equivalent – therefore those whose price is the least comparable – gather together. Finding the price of a booklet by looking at the maximum of a statement is like finding the average size of a class by measuring the tallest student.
This remark applies symmetrically to the bottom of the distribution, with a nuance. There are also objects outside the corpus - incomplete batches, very damaged booklets, sometimes articles which are not publications. But as they weigh little in the sum, they do not overturn any indicator. The bottom distorts the ad count, the top distorts the amounts: both deserve a look, not the same vigilance.
Six reading reflexes
Look at the second value
A maximum that is worth ten times the next does not belong to the same population. A regular slope can be read on three or four values in a row.
Be wary of any average
Here it is worth 2.2 times the median, and a single line divides it by two. On offers, it mainly measures the presence or absence of an aberrant object.
Open the most expensive ad
Thirty seconds is enough to see that it is not a booklet. This is the most cost-effective check that can be done on a reading.
Check language and format
The peaks observed here are a German hardcover, a Japanese book, an Italian work. The query returns what has the right words, not what belongs to the corpus.
Count observations
A median out of 60 ads and a median out of 8 do not read the same way. The number must accompany the value, always.
Prefer quantiles over limits
“85% under €20” provides information; “from €2 to €616” do not inform. Both describe the same reading.
What this statement does not establish
He is not saying that the most expensive item is overvalued. A bound work in a foreign language may be worth what you ask for it; simply, it does not compare to a booklet, and its presence in the same table as booklets creates category confusion, not price information.
Nor does he say that the median would be the “true” one. value. A median is a position in an offer list, not an estimate. It has the merit of being insensitive to extreme values – this is the property that this survey illustrates – but it remains a measure of what sellers ask, not what buyers pay.
Finally, the comparison between the three characters says nothing about their respective popularity or the value of their publications. It concerns the form of three distributions of offers on a given day. Another day, a single put on sale or a single withdrawal would be enough to tip one into the configuration of the other.
Translate this into a file
The instruction:never note a reference price without noting next to it how many observations base it and how they are distributed.A single figure, in a field of estimated value, loses in a few weeks everything that allowed it to be judged.
Concretely, three pieces of information are better than one: the median, the number of advertisements to which it relates, and the date of the statement. All three fit into a line of notes and transform an unverifiable figure into a dated observation. Add, when the distribution is unbalanced, a statement such as “summit not comparable” — you will avoid reopening the question in six months.
Above all, avoid recording a maximum as a reference value, whatever the temptation. The maximum of a statement is the amount that a seller hopes for, on an object that you have not verified is similar to yours. In this case, registering it would have given a booklet the value of a bound book in another language.
Finally, a word on the forks that we are tempted to write down ourselves in a file. Two terminals suggest a continuum between them; when 85% of the observations fall into the first twentieth of the interval, this continuum does not exist. A quantile — “most ads under twenty euros” — says the same thing without misleading, and can be reread without having to repeat the reading.
Because a single extreme value moves it entirely. Of the 60 advertisements noted on August 30, 2026, one line represents 47% of the total sum: removing it reduces the average from €21.79 to €11.73. The median, on the same gesture, goes from €9.88 to €9.86. The difference between the two indicators is not precision, it is nature.
By comparing the first value to the second. Here, €615.85 versus €61.85 — a factor of ten. Such a drop indicates another population of objects, while a regular slope is read over three or four successive values. This test is done in one second and avoids most reading errors.
Not necessarily booklets. Of the three surveys compared here, the summits are a German bound work, a Japanese illustrated book, an embossed edition with glasses and an Italian work. A query returns what contains the right words, not what belongs to the corpus you thought you were measuring.
No, and the comparison shows it: three readings from the same day give very close medians — €9.88, €9.38, €8.00 — for distributions of radically different shapes. One has a vertex fifty times greater composed of non-corpus objects, the other a vertex seven times greater composed of coherent fascicles. The median alone does not distinguish them.
A range suggests a continuum between its two limits. When 85% of the observations fall within the first twentieth of the interval, this continuum does not exist and the range is misleading while being accurate. Prefer a quantile — “most ads under twenty euros” — accompanied by the number of observations and the date.
A dated number is better than a single number
My Comics Collection keeps your estimates with their date and the basis for them, so that a noted price remains judgeable in six months.