Genetic Drift: History by Coin Flip
The question Mendel left on the table
Last chapter ended with a gift. Mendel proved that inheritance is discrete — an allele is a coin, not a splash of paint. It can hide for a generation and come back whole. And because it's a coin, you can do the single most important thing in all of population genetics: count it.
So let's count. Take a population — say, ten thousand people — and pick one gene where two versions float around. Some people carry version A, some carry version a. You tally up every copy in the whole group and find that 40% of the copies are A.
That number — 40% — is the allele frequency. It is the number Reich's entire field is built on. Every sentence he says like "this variant is common in Europeans and nearly absent in East Asians" or "this population lost this allele entirely" is a statement about an allele frequency.
Now the question that opens the rest of this series:
What makes that 40% change?
Because it does change. Wait a thousand generations and it might be 60%. Or 5%. Or 0%. Something moves it. Understanding what moves it — and this is not an exaggeration — is population genetics.
The famous answer, and the quiet one
Ask anyone who half-remembers school biology what changes allele frequencies and you'll get one answer: natural selection. The A version helps you survive, so A-carriers have more children, so A becomes more common. Survival of the fittest. Frequencies move because the allele is good.
That's real, and it happens. But it is not the answer we need first, and here's why: selection needs the allele to actually do something. It has to change a protein, which changes a trait, which changes how many kids you have.
Go back to T4 for a second. How much of your genome is actually genes? About 1–2%. The other ~98% is the non-coding ocean. And even inside genes, loads of mutations swap one letter for another and change nothing about the protein at all.
So for the overwhelming majority of the genome, selection has nothing to grip. These variants are neutral — they don't help you, don't hurt you, don't do anything. Natural selection is completely blind to them.
And yet their frequencies still move. A neutral variant sitting at 40% does not politely stay at 40% forever. It wanders. It can climb to 100% or crash to zero, having never once made anyone healthier, taller, or better at anything.
Something is pushing it. Not selection. Luck.
That's genetic drift — allele frequencies changing by pure chance. And here's the part that matters for where we're going: because drift works on the neutral 98% that selection ignores, drift is doing most of the work across most of your genome. The famous force is the minority shareholder.
The mechanism: every generation is a random sample
Here's the whole thing, and it's almost insultingly simple.
Not everyone has children. Of the copies of a gene present in one generation, only some get passed on. Meiosis (T6) already told you it's a coin-flip within a person — each parent hands over one of their two copies at random. Now zoom out and run that coin-flip across a whole population.
The next generation isn't a copy of this generation's gene pool. It's a random sample of it.
And random samples don't come out exactly right.
Flip a fair coin 10 times. You expect 5 heads. Do you get exactly 5? Often not — you get 7, or 3. Nothing is wrong with the coin. That's just what small samples do. Flip it 10,000 times and you'll land very close to 50% — not because the coin changed, but because big samples smother luck and small samples amplify it.
That's drift, entirely. If A is at 40% in the parents, the children should be at 40%. But they're a sample, so they come out at 38%. Or 43%. Then that number is the new starting point, and the next generation samples from that, and drifts again, and again, for a thousand generations.
Nobody is choosing. Nothing is being optimized. The frequency just... slides around, because sampling is noisy and there's no force holding it in place.
Population size is the entire dial
If drift is sampling noise, then one thing controls how violent it is: how big the sample is. In other words, how many people are in the population.
- Small population → tiny sample each generation → wild swings → fast, brutal drift. A frequency can lurch from 40% to 15% in a few generations for no reason at all.
- Huge population → enormous sample each generation → luck averages out → slow, gentle drift. The frequency barely twitches.
This is the single most useful intuition in the chapter, so let it land properly: drift is strong in small populations and weak in big ones. Roughly, the strength of drift scales with 1/(2N) — one over twice the population size. Double the population, halve the drift.
Which means population size isn't a boring demographic footnote. It's the volume knob on chance itself. A population of 50 is being pushed around by luck constantly. A population of 50 million is barely nudged.
Hold onto that. It's about to explain something big.
The trap to avoid. Drift is not "the population shrinks so the allele gets rarer." Drift has no preferred direction — a small population makes the frequency jump further each generation, up or down. Small size doesn't push the allele anywhere in particular. It just makes the wandering wilder.
The walk that always ends
So a neutral allele wanders. Up a bit, down a bit, no direction, no memory of where it's been. Mathematicians call this a random walk — a drunk stumbling along a street with no destination, each step independent of the last.
But this street has two walls, and they're sticky.
If the frequency ever wanders all the way down to 0%, the allele is lost. Not rare — gone. Nobody carries it, so nobody can pass it on. There is no way back. (A new mutation could reinvent it, but that's a fresh coin, not the old one returning.)
If it wanders all the way up to 100%, the allele is fixed. Everyone carries it, there's no alternative left in the population, and it can't drift anymore because there's nothing to drift against.
Both walls are absorbing — once you touch one, the game is over forever.
Now here's the conclusion that should genuinely unsettle you. A random walk, given enough time, always hits a wall. Wander long enough with no force holding you in the middle, and you will eventually stagger into 0% or 100%. It's not a possibility. It's a certainty.
Every neutral allele is on death row or headed for a throne. Given enough generations, drift alone guarantees that every variant either disappears completely or takes over completely. The only question is which — and which one it is, is decided by luck, not merit.
There's an elegant little result here: the probability that a neutral allele eventually takes over the whole population is exactly its current frequency. Sitting at 40%? It has a 40% chance of one day being universal and a 60% chance of vanishing without a trace. That's it. That's the whole rule. No fitness term. No merit. Just where the coin happens to be standing right now.
The founder event: drift's greatest hit
Now combine the two things you just learned — drift is savage in small populations, and drift is just sampling — and push them to the extreme.
A handful of people leave. They cross a strait, walk out of a valley, survive a collapse, split off over a doctrine. Say twenty of them. And their descendants become millions.
Those twenty were a random sample of the population they left. Not a representative delegation — a sample. Whatever alleles they happened to be carrying, in whatever weird proportions luck dealt them, that sample is now the entire genetic starting point of every person who will ever descend from them.
A variant that was 2% back home might be 20% among the twenty, purely by accident. It's now 20% in a population of millions. An allele that was 30% back home might be carried by none of the twenty at all — and it's now gone from that lineage forever, without a single thing wrong with it.
This is a founder event. And it's Reich's own vocabulary — here he is on the podcast, when Dwarkesh asks about the small population that seeded everyone outside Africa:
"By bottleneck we mean founder event, a relatively small number of people giving rise to a large number of descendants today." — David Reich
Note that he treats "bottleneck" and "founder event" as the same thing, and now you can see why: whether a small group walked away (founder event) or a big group got crushed down to a few survivors and rebounded (bottleneck), the genetics is identical. A tiny number of people; a random sample of alleles; a huge future built on that sample. Same mechanism, different story.
And this is what happened to everyone reading this outside Africa. The group that expanded into Eurasia was small — on the order of thousands — and it carried a random, thinned-out sample of Africa's genetic variation. That's why African populations today hold far more genetic diversity than everyone else on Earth combined. Everybody else is descended from the sample.
Predict-your-confusion sidebar. You might now assume that founding group was strange or special — a genetically impoverished band of oddballs. Reich explicitly says no: "this was not unusual to have a group with low diversity. The great majority of African groups would have had very low diversity. The one that started expanding into Eurasia also had low diversity, but it was so successful it didn't mix with very many other groups and recharge its diversity." It wasn't unusual. It was lucky. That's the whole theme of this chapter.
Why drift is the engine under Reich's entire field
Here's where this chapter cashes out, and it's bigger than it looks.
1. Drift is what creates populations in the first place
Take one population and split it in two — a mountain range rises, a group migrates, a social barrier hardens. Now you have two groups that no longer trade DNA.
Each one drifts. Independently. Different random samples, different luck, different walks. Allele frequencies that started identical slowly, inevitably pull apart. 40% here becomes 55%; 40% there becomes 25%. Nobody was selected for anything. They just stopped sharing coins and let chance do its work.
Do this for enough generations across enough genes and the two groups become genetically distinguishable — not because either is better, but because they drifted apart.
This is the signal Reich reads. A "population" in his work is a group with a characteristic set of allele frequencies, and those frequencies are characteristic because of drift. Without drift, every human group would look identical at the genome level and there would be no history written in DNA at all — no signal, no field, no book. Drift is what makes the record legible. The noise is the data.
2. It explains what's hiding in your genome
Remember T6's payoff — that shared DNA segments get chopped shorter by recombination each generation, so segment length dates the past. Now watch Reich point that tool at India.
He and his colleagues found something odd: Indian groups are far more genetically different from each other than European groups at similar distances — at least three times more. Why?
The answer is founder events, and they found the fingerprint: long identical stretches of DNA shared between two people in the same group — the mark of a recent common ancestor. Segment length gave the date (T6's clock, doing real work).
The finding: around a third of Indian groups have experienced founder events as strong or stronger than the ones behind Finns or Ashkenazi Jews — populations famous in medical genetics for exactly this. The most striking case Reich found was the Vysya of Andhra Pradesh, roughly five million people, whose founder event dates to between three thousand and two thousand years ago — and who have maintained strict endogamy essentially ever since.
That's the jati system, written into the genome. India is not one gene pool; it's thousands of small ones, each drifting on its own for millennia. Every one of those groups has been playing this coin game in a small population — which, as you now know, means luck has been shouting.
3. It rewires how you hear the whole podcast
The deepest thing drift gives you isn't a mechanism. It's a posture. Listen to Reich describe which human groups spread and which vanished:
"It's not a triumphal march of superiority and inferiority with the group that now comes in having advantages and somehow establishing itself permanently... It all results in an almost random process of who spreads or ends up on top." — David Reich
"An almost random process of who spreads or ends up on top." That is a population geneticist telling you that the winners of human history mostly weren't better. They were luckier. Reich's picture of the past is extinction after extinction — Neanderthal groups, Denisovan groups, and modern human groups alike — with survival looking a lot more like a coin landing than a trophy being earned.
You cannot really hear that sentence until you understand drift. With drift, it stops being modesty and becomes the mathematics of the situation.
The challenge
Three questions, escalating:
Warm-up. A neutral allele sits at 30% in a population of 200 and at 30% in a population of 2 million. In a hundred generations, which one's frequency has probably moved further from 30% — and why? (Careful: the question is how far it moved, not in which direction.)
The real one. A variant that does absolutely nothing — no effect on any protein, any trait, any survival advantage whatsoever — is found at 100% in a population. Natural selection cannot have put it there; there was nothing to select. Walk through how it got to 100% anyway, using the random walk and the two sticky walls. Then answer the follow-up: was there anything special about that allele compared to the variants that vanished?
Think ahead. Someone tells you: "Two populations having different allele frequencies proves they adapted to different environments." Dismantle this using everything above. Then push further — explain why, if drift were somehow switched off, David Reich would have no field at all. (You've just worked out that the randomness isn't noise contaminating the historical signal — the randomness is the signal.)
Next: drift takes one population and splits it into two that slowly wander apart — a branching tree. That's the tidy story, and it's **wrong. Reich's central claim is that human history isn't a tree at all: the branches keep crashing back together. Populations that drifted apart for tens of thousands of years meet and **mix, and everyone alive is the product of mixtures upon mixtures. Next chapter we break the tree and build the **trellis* — admixture, and how you catch two drifted populations in the act of merging.*