---
title: "Dating admixture with DATES and ALDER: when did the mixing happen?"
description: "qpAdm says how much; linkage-disequilibrium decay says when. How DATES and ALDER read generation counts out of chromosome fragment lengths, what the dates mean, and how a date corroborates or breaks a qpAdm model."
canonical: https://www.ancestrify.io/blog/dating-admixture-dates-alder
date: 2026-09-01T09:24:00+00:00
updated: 2026-09-24
author: "Andi Thomaj"
---

# Dating admixture with DATES and ALDER: when did the mixing happen?

A [qpAdm weight](/blog/understanding-qpadm) is timeless: 47% steppe ancestry says nothing about
whether the mixing happened in 3000 BCE or 300 CE. But the genome keeps time in a second channel
— the *lengths* of the ancestry fragments recombination has been chopping since the admixture —
and two related methods, **ALDER** and **DATES**, read a date out of it. A date is the cheapest
strong corroboration a qpAdm model can get, which is why the two travel together in the
literature and in [our own reports](/blog/qpadm-model-record-explained).

## The clock: recombination grinds fragments down

At the moment of admixture, a first-generation offspring carries whole chromosomes from each
parent population. Every generation after, recombination breaks parental blocks at roughly one
crossover per Morgan per generation — so source-population fragments shorten, generation by
generation, at a known statistical rate. Fragment length is therefore a clock: long intact blocks
mean recent admixture, confetti means ancient.

Rather than calling fragments explicitly (hard in low-coverage data), both methods measure the
statistical shadow of block structure: **admixture linkage disequilibrium** — the correlation
between ancestry-informative alleles at pairs of sites — as a function of genetic distance *d*.
That correlation decays as **e^(−n·d)** for admixture *n* generations ago: an exponential whose
decay constant *is* the answer. Fit the curve, read off *n*, multiply by ~28–29 years per
generation, subtract from the sample's own date (for ancients), and you have a calendar estimate
with a standard error — jackknifed, [as usual](/blog/how-to-read-qpadm-p-value-z-score-standard-error).

**ALDER** (Loh and colleagues 2013, descending from Moorjani's ROLLOFF) established the weighted
form and its significance test for whether admixture occurred at all. **DATES** (Narasimhan and
colleagues 2019; formalised by Chintalapati, Patterson and Moorjani 2022) adapted the machinery
for exactly the regime ancient-DNA work lives in: it needs only the target plus two reference
populations, tolerates pseudohaploid single samples, and was validated across the simulation
range that matters for [AADR-era datasets](/blog/aadr-sample-labels-sg-dg-ho-explained). In the
Holocene survey work it dated the steppe-related admixture arriving in Central and South Asia
and, in our own pipeline, the method behind era-dating work like the Çinamak analyses.

## What a date adds to a weight

A qpAdm model and a date check each other from independent channels — frequencies versus block
lengths:

- **Corroboration**: a passing model whose implied timing matches the LD date is the strongest
  everyday evidence package available. On simulated admixture-graph histories (Flegontova et al.
  2025), [checking distal results with independent methods](/blog/qpadm-distal-vs-proximal) such
  as PCA and unsupervised ADMIXTURE cut false discoveries further, and a date is an independent
  check in the same spirit; on the same study's simulated stepping-stone landscapes the qpAdm
  protocols did not differ significantly, so no single check makes a model safe.
- **Contradiction**: [a passing model](/blog/qpadm-best-practices) whose source was extinct
  by the LD date — or a "recent" source against a 100-generation decay — is a proxy wearing the
  wrong era's clothes: the ancestry is real, the historical story is not.
  [Choose a source consistent with the date](/blog/how-to-choose-qpadm-sources-and-outgroups).
- **Discrimination**: two sources qpAdm cannot split
  ([below the resolution floor](/blog/is-qpadm-still-reliable)) sometimes imply different mixing
  eras — and the clock votes where the frequencies abstain.

## Reading the caveats

The clock's assumptions earn their own honesty box. It dates **pulses**: continuous mixing over
centuries returns one intermediate date, not the interval's edges (extensions like DATES'
multiple-pulse fitting help, within limits). Reference choice matters less than qpAdm's source
choice but still matters. Very old admixture (hundreds of generations) decays into noise within a
few centimorgans; very recent admixture in small samples is dominated by pedigree luck. And a
generation time of 28–29 years is a convention — quote dates with their errors *and* that
multiplier visible.

None of which dents the headline: for the price of one more run on data you already
[merged](/blog/qpadm-data-preparation-merge-aadr), the *when* column gets filled in beside the
*how much* — and a model carrying weight, error, p-value **and** a compatible date is as close to
closed as this field gets.

Terms used here are defined in the [glossary](/glossary).

## References

- Loh, P.-R. et al. (2013). Inferring admixture histories of human populations using linkage
  disequilibrium. *Genetics*, 193(4), 1233–1254. (ALDER.)
- Moorjani, P. et al. (2011). The history of African gene flow into Southern Europeans,
  Levantines, and Jews. *PLoS Genetics*, 7(4), e1001373. (ROLLOFF.)
- Narasimhan, V. M. et al. (2019). The formation of human populations in South and Central Asia.
  *Science*, 365, eaat7487. (DATES in the field.)
- Chintalapati, M., Patterson, N. & Moorjani, P. (2022). The spatiotemporal patterns of major
  human admixture events during the European Holocene. *eLife*, 11, e77625. (DATES formalised.)
- Flegontova, O. et al. (2025). [Performance of qpAdm-based screens for genetic admixture on graph-shaped histories and stepping stone landscapes](https://doi.org/10.1093/genetics/iyaf047). *Genetics*, 230(1), iyaf047.