What Survives, and Why It Isn't Random

Of the roughly one hundred and twenty plays Sophocles is said to have written, seven survive complete. Of Aeschylus, seven of about ninety. The figure that matters is not the loss but the selection: those survivors are not a random sample. They reached us because somebody kept copying them by hand for a thousand years, and people copy what their own culture finds useful, orthodox or beautiful. We are not reading Greek drama. We are reading a medieval anthology of it, chosen by criteria that were not ours and are mostly unrecorded.

This is survivorship bias, and in history it is not an occasional hazard. It is the condition of the whole discipline.

The mechanisms are dull and enormously consequential. Stone survives and wood does not, so we systematically over-read societies that built in stone. Dry climates preserve organic material and wet ones destroy it, so Egyptian daily life is vastly better documented than Gaulish daily life for reasons that have nothing to do with either. Institutions keep records and households do not. Literate people leave writing and illiterate people appear only when somebody literate had a reason to write them down. Cities are excavated more than fields, because cities are where the interesting buildings are and where the archaeologists live.

Put these together and the record over-represents the durable, the official, the urban, the literate and the male, in each case for reasons of physics and administration rather than importance. A historian who simply counts what is in the archive is measuring the archive.

Which brings us to the archive itself, and to the realisation that changed how the profession works. An archive is not a container. It is an actor. Somebody decided what entered it, what was catalogued and under what heading, what was weeded, what was closed and for how long. A colonial archive documents a colonial state's questions. It will tell you a great deal about tax yields and rebellion and almost nothing about what the taxed thought, because nobody was assigned to record that. Studying the shape of a collection frequently tells you as much as reading its contents.

The hardest part of all this is knowing what to make of silence. Absence of evidence is evidence of absence, but the strength of that inference depends entirely on whether you would have expected a trace. In a complete parish register, a birth that is not recorded means something: the record was designed to catch every one. In a body of material that is oral, fragmentary, or kept by an institution uninterested in the thing you are asking about, absence means almost nothing at all. Confusing the two is the most common mistake made with a gap.

Here is the standard demonstration, and it is worth carrying. A study concludes that medieval English women rarely worked outside the home, on the evidence of guild membership rolls. The rolls are genuine, the reading is careful, and the conclusion is wrong, because guild rolls record formal membership and women were largely barred from it. The work is invisible to the source by construction. Coroners' inquests (which record, incidentally and with no interest in the question, where people happened to be when they died) show women at work all over the economy.

Same century, same country, opposite answer, and neither source is lying. One of them was simply never built to see.