Preservation bias
A Socratic walk-through of preservation bias — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does the archaeological record hold far more pottery than the clothing everyone actually wore?
Everybody wore clothes every day, and almost nobody owned as many pots as garments. Yet dig a settlement of almost any period in a temperate country and you lift sherds by the crate and textile by the thread, if at all. The tempting reading is that these were pot-rich, cloth-poor people. Before accepting it, ask a duller question: what happens to a tunic and a pot after they are thrown away?
Reasoning it through
REASONING #The ground is not a container. It is a chemical process with things in it. Wool, linen, leather, wood and basketry are polymers that something can eat — fungi and bacteria have enzymes for exactly them, and the reaction needs water, oxygen, a workable temperature and time. Fired pottery is in a different situation entirely. It has already been through the fire; its clay minerals have been reorganised into a silicate ceramic offering no chemical energy to any organism and dissolving at no ordinary soil pH. Nothing in the soil wants it.
That is most of the answer, and it has a second stage that quietly doubles the effect. Break a pot and you get thirty durable countable objects where there was one. Break a tunic and you get rags, then lint, then nothing. Ceramic does not merely survive; it multiplies into the very units archaeologists count.
But a filter operates even before the ground is involved. What entered it in the first place? A cracked cooking pot is worthless and is thrown out that week; cloth was expensive in labour, so garments were mended, handed down, cut for children, then for cleaning cloths, and finally sold on for rag. Very little cloth was ever discarded intact. Two independent filters — a low rate of entry and a high rate of destruction — multiply, which is why the disparity is extreme rather than merely noticeable.
Now the part that matters most. Is this filter, unlike others, something we can measure? Decay stopped everywhere one of its requirements was missing. Permanently waterlogged, oxygen-starved deposits — bogs, wells, ships' bilges, latrine pits — preserve wood and cloth. So do genuinely arid sites, and permafrost. Charring converts grain and wood into a form nothing eats. And where cloth lay against corroding copper alloy or iron, the metal's salts replaced the fibre and left a mineral cast of the weave. Each is a window onto part of what the ordinary record destroyed.
That is the load-bearing claim: the filter is not indifferent noise but a systematic function of material chemistry and depositional setting, and because those windows exist, its shape can be estimated rather than merely lamented. The refuting observation is clean and available: if waterlogged, arid and frozen sites, once dug, showed roughly the same proportions of material classes as ordinary dryland sites of the same culture, the scarcity of cloth on dryland sites would be a real scarcity and this whole account would be wrong. They do not.
The analogy
THE ANALOGY #Think of photographing a room lit only by red light. What you get is sharp and entirely truthful about the red objects; the blue ones are simply not in the picture. Nothing in the photograph is a mistake, and no amount of studying it harder will bring the blue objects out. Change the lamp — walk into a waterlogged site — and a different set of objects appears.
relighting a room is reversible, and the blue objects were there all along. The ground destroys, so no later technique recovers a decayed tunic; the wet site is not the same room relit but a different room, and one people used differently, which is the difficulty the next section turns on.
Clarifying the model
THE MODEL #It is worth being exact about how this differs from two neighbouring ideas, because the three get run together. Historical silence concerns what somebody had a reason to record: a filter applied by human intent at the moment of creation. Survivorship bias concerns a dataset whose entry condition is correlated with the outcome under study — the returning aircraft, the surviving fund. Preservation bias is neither. Its filter runs after deposition, is applied by soil chemistry, and is wholly indifferent to whether an object was important, typical or interesting. That indifference is the fixed point of difference, and it is what makes the bias tractable: a filter that does not care about your research question can be characterised once and applied everywhere, which is never true of an archive curated by an interested party.
Now let me undercut my own optimism, because "we have controls" is too comfortable. Bogs, deserts and permafrost are unusual places where people did unusual things — a bog deposit may be votive rather than domestic rubbish, so its contents are not a fair sample of daily life. Charred grain is the grain that happened to burn, over-representing accidents and hearths. Mineral-replaced textile survives precisely where metal was buried with it, meaning graves furnished with metalwork, which skews toward wealth and particular rites. Every correction for a taphonomic bias is drawn from a sample carrying a locational and behavioural bias, and the corrected estimate inherits both.
And I will not give you a survival percentage for organics: published figures depend so completely on material, soil, period and what counts as "recovered" that a single number would be decoration rather than measurement.
A picture of it
THE PICTURE #How to readThe five spokes are depositional settings, not places on a map, and the three rings are material classes — a curve far from the centre on a spoke means that material usually survives there. The values are my ordinal judgement of relative survival, not measurements; read the shapes, never the numbers. Pottery's ring is nearly circular, which is the point: it barely cares where it lands. Cloth collapses to almost nothing on acid and ploughed ground, and the gap between the outer ring and the inner one, spoke by spoke, is the bias.
What became clearer
WHAT CLEARED #The pot-rich, cloth-poor past is an artefact of soil chemistry compounded by discard behaviour: ceramic cannot be eaten and fragments into countable units, while cloth was too valuable to throw away and too edible to last once it was. What separates this bias from the human filters it is usually grouped with is that it is indifferent to what it destroys, and therefore partly measurable — through wet, arid, frozen and charred windows. But those windows are peculiar places used in peculiar ways, so the correction improves on ignoring the problem rather than returning us to an unbiased record.
Where to go next
ONWARD #- How residue analysis and mineral pseudomorphs let a vanished object be studied through traces it left in something durable.
- Why the same reasoning distorts the recovered range of tools, since stone survives everything and wood almost nothing.
Key terms
TERMS #| Term | What it means |
|---|---|
| Taphonomy | the study of what happens to material between its use and its recovery, treated as a process with its own laws. |
| Pseudomorph | a cast of a decayed organic object formed in the corrosion products of adjacent metal, preserving weave or grain but not the material. |
Every term the collection defines is gathered in the glossary.