This is the practical post in the series introduced in Jewish ancestry and ancient DNA. It is for someone with Jewish ancestry who already has a raw DNA file from 23andMe, AncestryDNA, MyHeritage or FamilyTreeDNA, has looked at the percentages, and is deciding whether a qpAdm analysis would tell them anything more. We sell one, at /qpadm, so read this as an insider's account that tries to be exact rather than persuasive. As everywhere in the series: ancestry is not identity, and nothing in any report is a statement about who is Jewish.
What the percentage tools measure#
A consumer ancestry breakdown compares your genome with reference panels of living people grouped by self-reported origin and reports how much of your genome is best matched by each group. For Jewish customers the reference groups include one or more Jewish categories, and this is where the method does something that is technically correct and practically misleading.
Because of the medieval bottleneck and centuries of endogamy, Ashkenazi genomes resemble one another more than almost any other population's members resemble each other. The panel therefore recognises the Ashkenazi cluster with very high confidence, and a customer with four Ashkenazi grandparents is reported at close to 100%. The number is a statement about resemblance to that reference group. It says nothing about what the group is made of: the Levantine half and the southern European half of Ashkenazi ancestry are both inside the reference before the counting starts, and so the tool cannot report them separately even in principle. The same applies, with smaller clusters, to the Sephardic, Mizrahi and Yemenite categories some panels offer.
Three consequences follow.
- "0% Middle Eastern" is not a finding. It means the Middle Eastern ancestry was assigned to the Jewish category, not that it is absent.
- Partial results are unstable. A customer with one Jewish grandparent often sees the Jewish share drift between updates, because the panels are re-drawn and the bottleneck signal is easy to over- or under-count.
- The tool cannot fail. There is no version of the calculation that reports "this genome is not consistent with the proposed mixture". Every genome gets a breakdown that sums to 100%.
Free calculators built on Global25 coordinates have the same property: a distance to the nearest population averages, and a mixture that always sums to 100%. Useful for exploration, and we offer it ourselves at /g25; the comparison with qpAdm explains which question each method answers.
What a qpAdm model measures#
qpAdm is the method from the ADMIXTOOLS package, developed in David Reich's laboratory and used in most published ancient-DNA admixture studies since 2015, including the Levantine and Ashkenazi studies discussed in this series. It takes your genome as the target and asks whether it can be written as a mixture of chosen ancient source populations, measured against a set of distant outgroups. It returns a weight for each source, a standard error on each weight, and a single p-value for the whole model, which can reject it. The method is in Understanding qpAdm; the reading guide for the three numbers is here.
The differences for a Jewish genome are concrete.
- The sources are ancient, not modern. Instead of a reference group of living Ashkenazi Jews, the sources are the Bronze Age Canaanites of Megiddo and Hazor, the Imperial Romans of central Italy, the Neolithic farmers of the Zagros, the pre-Islamic Arabians, the Iron Age Iberians, the medieval Germanic and Slavic populations, and so on. The Levantine and European halves of an Ashkenazi genome are therefore separate sources with separate weights, and the model reports how confidently it separated them.
- The model can fail. If an Ashkenazi genome is proposed as Canaanite plus Imperial Italian and that mixture is inconsistent with the data, the p-value says so. If a third source is proposed that the genome does not need, its Z-score falls below the bar and the nested-model check removes it.
- Every number is published. The report shows each source's weight, standard error and Z-score, the p-value, and the complete outgroup list; the Reading view shows the full record, down to the chi-square, the f4 rank and the nested-model table, with a written explanation from the analyst of why your genome resolved into these sources. Every number is explained in The model record explained, and the record downloads as a plain-text file at every tier.
Since August 2026 the catalog carries a dedicated Canaanite (2000 - 1200 BC) source, built for exactly this kind of genome; its post explains why. The full directory lists all the sources, each with its own page.
What the four tiers buy#
The qpAdm analysis is sold at four tiers: €29.99, €39.99, €49.99 and €59.99. What changes between them is not what most people assume.
The tiers do not buy a different standard. Every model we publish, at every tier, must clear the same bar: p above 0.05, and for every source in every era a Z-score above 3 in magnitude and a standard error below 0.10. They do not buy more content. The report, its maps and its videos are identical at every tier.
What they buy is search effort: how far the analyst keeps going past the first model that clears the bar. For a Jewish genome that matters more than for most, because there are several plausible sources per era and the interesting questions are the marginal ones. Is the eastern European share in an Ashkenazi genome real, or noise? Does a Moroccan Sephardic genome need the Berber source, or does the Iberian one absorb it? Does a Yemenite genome prefer Canaanite or the Arabian Peninsula for its Levantine side, and by how much? Each is a separate model with its own p-value, and a deeper tier means more of them are composed, run and rejected before one is published. Deeper tiers buy certainty, not complexity; the models that survive the longest search are often the simplest.
If your question is "will the top tier give me a lower standard error?", the answer is no. Standard error is set by your file's coverage. If your question is "will it give me a more defensible model?", the answer is yes.
The honest limits#
Three limits are specific to Jewish genomes, and they belong in every buyer's head before the report opens.
Endogamy and drift. After a founder event, a population drifts a little away from every ancient source at once, in a direction no source supplies. A well-composed model absorbs this and the p-value stays comfortable; a poorly composed one produces a source with an inflated weight or a rejection with no visible cause. This is the reason every model we publish is composed by one person rather than by rotating combinations until one passes. We built automated rotation, measured its false-discovery rate on exactly this kind of genome, and removed it.
Closely related sources. Canaanite versus Phoenician, Imperial Italian versus Aegean, Canaanite versus Arabian: these pairs are close in genetic space, and how well a model separates them is the standard error, which is set by coverage. A low-coverage file can return a model in which the p-value passes but one source's standard error is above the bar, and that model is not published. Check your file's coverage for free in the file check before you pay; if it fixes the standard errors above the bar, the honest answer is to say so before purchase.
Identity. A source population is a statistical proxy, not a family. A passing model is one the data did not refute, not one that was confirmed. And a result describes your genome, not your identity: it says nothing about Jewish status, conversion, tribe, lineage or belonging, and it is not evidence for or against any of them.
After the report#
Once a report is published you can unlock the Model Lab for €10, one unlock, one time, and run your own qpAdm models on your own merged sample with the same panel of ancient genomes, including the medieval Erfurt Jewish genomes and every Levantine group in the AADR, up to 100 runs per rolling 24 hours. Expect rejections; they are the method working. The walkthrough is in Run your own qpAdm models.
A checklist before you buy, anywhere#
- Does the product show a p-value, a standard error and Z-score per source, and the complete outgroup list? If not, it is not a qpAdm result you can check.
- Are the sources ancient populations, or modern Jewish reference groups relabelled?
- Can you download the full record and hand it to another analyst?
- Does it tell you what your file's coverage supports before you pay?
- Does it claim any population, tribe or person as your ancestor? It should not.
- Does it claim to measure who is Jewish? It cannot, and it should say so.
Ours is at /qpadm. The community posts, Ashkenazi, Sephardic, Mizrahi, Yemenite and Mountain, Georgian and Bukharian, describe what the model looks like for each.
References#
- Behar, D. M. et al. (2010). The genome-wide structure of the Jewish people. Nature, 466, 238–242.
- Carmi, S. et al. (2014). Sequencing an Ashkenazi reference panel supports population-targeted personal genomics and illuminates Jewish and European origins. Nature Communications, 5, 4835.
- Harney, É., Patterson, N., Reich, D. & Wakeley, J. (2021). Assessing the performance of qpAdm: a statistical tool for studying population admixture. Genetics, 217(4), iyaa045.
- Waldman, S. et al. (2022). Genome-wide data from medieval German Jews show that the Ashkenazi founder event pre-dated the 14th century. Cell, 185(25), 4703–4716.
- Xue, J. et al. (2017). The time and place of European admixture in Ashkenazi Jewish history. PLoS Genetics, 13(4), e1006644.
