Science & SocietyModels in Anthropology
Pedigree collapse vs endogamy: what each does to your DNA matches
Pedigree collapse repeats a few ancestors in one tree; endogamy links a whole community many times over. How each inflates shared cM, with worked examples.

Pedigree collapse means the same ancestor fills more than one place in a single family tree, usually because of one or a few marriages between relatives that you can find in the records. Endogamy means a whole community married within itself for many generations, so that its members are related to one another by many routes at once, most of them too old to document. Both make DNA matches share more than their known relationship predicts, but collapse inflates a few matches on particular branches, while endogamy adds a little to nearly every match from the community.
The difference at a glance
| Pedigree collapse | Endogamy | |
|---|---|---|
| Scale | one person's tree | a whole population |
| Typical cause | a known cousin marriage, or a few | generations of marrying inside a group, by rule or by geography |
| Where the loops close | often within the records, a few generations up | mostly beyond the records, many generations up |
| Who is affected | matches descending from the repeated ancestors | almost every match from the same community |
| Effect on shared DNA | a sizable addition, up to doubling, on specific matches | a modest addition to many matches, mostly as short segments |
| Can you draw it? | yes, with the loop marked | rarely; the routes are too many and too old |
One way to hold the two together: endogamy is pedigree collapse spread across everyone in a population and pushed back past the paper trail. Every member's tree folds, but the folds are too many and too old to point to.
Two words from two traditions
"Pedigree collapse" is a genealogist's term for a count. In generation g a tree has 2^g places, and when one person fills two of them, the number of distinct ancestors falls below that. Our piece on how many ancestors you have shows the doubling and where it has to give way, and the pedigree collapse calculator turns your own count of distinct people into a share of repeats.
"Endogamy" comes from anthropology, where it names a marriage rule or strong preference: marry inside your caste, your religious community, your village or your descent group. Its opposite is exogamy, marrying out, and one society can practice both at different levels, forbidding marriage inside a clan while expecting it inside the wider community. Many societies permit, and some prefer, marriages between particular cousins, as our explainer on cross and parallel cousins describes. Genetic genealogists use the word more loosely, for any population whose members share many ancestors because they married among themselves for a long time, whether a rule required it or an island, a valley or a migration did.
Why every extra route adds DNA
Sewall Wright's coefficient of relationship, r, adds up paths. Each path from one person up to a shared ancestor and down to the other contributes (1/2) raised to the number of parent and child links on it. Two people related by more than one route simply get more paths, and to a close approximation the expected shared DNA adds up route by route.
The cousin calculator handles one route at a time: one shared couple, or one shared person. Its expected shared DNA for relatives other than full siblings is 2r × 3,500 cM, where 3,500 cM is its rounded length of one copy of the genome. For several routes, run it once per route and add the results.
That is also why DNA so often disagrees with a tree. The expected figures, as Doug Speed and David Balding put it in a 2015 review of relatedness, "vary according to the pedigree that happens to be available." A tree that stops at great-grandparents cannot show routes that close further up, and the DNA counts them anyway.
Worked example: a loop you can see
Imagine Mara, whose parents, Leo and Rose, are first cousins. Leo and Rose share a pair of grandparents, call them the Hales, who are Mara's great-grandparents and fill two places each in her tree. Now take Sam, a grandchild of the Hales through a third child. Sam is a first cousin of Leo and also a first cousin of Rose.
- One route. Through Leo, Mara is 3 generations below the Hales and Sam is 2. Enter 3 and 2 in the cousin calculator: first cousin once removed,
r = 2 × (1/2)^5 = 1/16, expected shared DNA2 × 1/16 × 3,500 = 437.5, shown as about 438 cM. - The second route. Through Rose the counts are again 3 and 2, another 1/16.
- Add them.
r = 1/16 + 1/16 = 1/8, and the expected sharing is2 × 1/8 × 3,500 = 875 cM, which is exactly the calculator's average for first cousins.
So Sam, a first cousin once removed on paper, is expected to match Mara like a first cousin. The Shared cM Project's observed ranges show why that misleads: in version 4.0, first cousins once removed averaged 433 cM (range 102 to 980) and first cousins 866 cM (range 396 to 1,397), so a match near 875 cM fits the closer relationship far better.
Now take a match on one of Mara's other branches, say a descendant of Leo's grandparents on his other side, who are not Hales. Nothing is repeated there, and that match is expected to share the ordinary amount. That is the signature of pedigree collapse: the extra DNA sits on the branches that run through the repeated couple, and nowhere else.
Mara herself carries the loop too. Her inbreeding coefficient from it is 1/16, so about 3,500 / 16 ≈ 219 cM of her genome is expected to lie in stretches where her two copies are identical. Geneticists call these runs of homozygosity.
Worked example: loops you cannot see
Now imagine Ada and Ben from a small community that has married mostly within itself for two centuries. Their documents show one route: they are third cousins, sharing a couple four generations above each of them. Suppose that, unknown to either, they are also fourth cousins in two other ways and fifth cousins in ten more. The numbers are invented for the arithmetic.
| Route | Calculator average per route | Routes | Expected cM |
|---|---|---|---|
| third cousin | 54.7 | 1 | 54.7 |
| fourth cousin | 13.7 | 2 | 27.3 |
| fifth cousin | 3.4 | 10 | 34.2 |
| Total | 13 | 116.2 |
The documented route predicts about 55 cM. All thirteen predict about 116, more than twice as much and close to the calculator's 109 cM for second cousins once removed. No single hidden route matters much; together they do.
What testing companies report will differ from that total, because most of the hidden routes deliver their DNA in short pieces. AncestryDNA's matching white paper (Ball and colleagues, 2020) keeps a segment only if it is at least 6 cM long and the match totals at least 8 cM, and it notes that longer segments point to more recent inheritance. For matches under 90 cM in total it also runs a step called Timber, which shrinks the measured length of segments in regions where a person matches unusually many others, on the reasoning that such pile-ups seldom reflect recent shared ancestors. Short background sharing is therefore partly trimmed away, but not all of it.
Real populations show the effect. Brenna Henn and colleagues (2012), working with 23andMe data, found that average sharing within ethnolinguistic populations such as Finns, Ashkenazi Jews and Native American groups differed from that of continent-wide samples by several orders of magnitude. In their pedigree simulations built from Ashkenazi genomes, many pairs shared segments that did not come through the common ancestor the simulation had specified, and some pairs looked more closely related than the simulated cousinship. Their reading was that many people in the sample were already related as second to ninth cousins before the simulation began.
What the Shared cM Project says
Blaine Bettinger's Shared cM Project collects shared cM totals from testers who know their relationship. Version 4.0 (March 2020) was built from nearly 60,000 submissions, and its notes address both problems directly: endogamy and pedigree collapse raise the DNA shared at a given relationship, and although the submission form asked about known endogamy or cousin marriage, many contributors will not know about it, and some may have reported only one of several relationships. Bettinger runs a separate collection for pedigree collapse, double and multiple cousins, and runs of homozygosity.
Set the project's averages beside the calculator's, as our guide to shared cM by relationship does relationship by relationship, and the gap grows with distance:
| Relationship | Calculator average (cM) | Shared cM Project 4.0 average (cM) |
|---|---|---|
| second cousin | 219 | 229 |
| third cousin | 55 | 73 |
| fourth cousin | 14 | 35 |
| fifth cousin | 3 | 25 |
Not all of that gap is hidden kinship. For relationships where some pairs share nothing, the project averaged only the pairs who shared a detectable amount, and what counts as detectable depends on each company's threshold. The averages for distant cousins therefore describe the matches that were found, not all cousins, and that alone lifts them above the simple expectation. Unrecognized collapse and endogamy add to it. Version 4.0 also dropped the earlier split of results by reported endogamy, saying it was rarely used and made little difference to relationship predictions, which is a reminder that a yes or no box captures only what testers know.
Reading your own matches
- Look at where the inflated matches sit. If they cluster on one branch and that branch has a documented loop, you are looking at pedigree collapse. Count the routes, run the cousin calculator for each, and add.
- If almost every match with roots in one community comes in high, and that community married within itself for generations, expect endogamy. Henn and colleagues found that 99% of the Ashkenazi customers in their sample had at least one detectable relative in the database.
- Treat a predicted relationship as a likely overestimate of closeness. Under endogamy the true relationship can be more distant than the prediction, as Henn's simulations showed.
- Check segment sizes. A total built from one or two long segments points to a recent route; the same total built from many short ones points to many old routes.
Back to Ada and Ben. A report of about 110 cM spread over many short segments, between two people from the same small community, is as consistent with third cousins plus a crowded background as with second cousins once removed. Only the documents can choose between them.
Sources and further reading
- Bettinger, Blaine T. 2020. "The Shared cM Project, Version 4.0 (March 2020)." The Genetic Genealogist. https://thegeneticgenealogist.com/2020/03/27/version-4-0-march-2020-update-to-the-shared-cm-project/
- Henn, Brenna M., Lawrence Hon, J. Michael Macpherson, Nick Eriksson, Serge Saxonov, Itsik Pe'er, and Joanna L. Mountain. 2012. "Cryptic distant relatives are common in both isolated and cosmopolitan genetic samples." PLoS ONE 7 (4): e34267. https://doi.org/10.1371/journal.pone.0034267
- Ball, Catherine A., Mathew J. Barber, Jake Byrnes, Peter Carbonetto, Kenneth G. Chahine, Ross E. Curtis, Julie M. Granka, and others. 2020. "AncestryDNA Matching White Paper: Discovering genetic matches across a massive, expanding genetic database." AncestryDNA, updated 15 July 2020. https://www.ancestrycdn.com/support/us/2020/08/matchingwhitepaper.pdf
- Speed, Doug, and David J. Balding. 2015. "Relatedness in the post-genomic era: is it still useful?" Nature Reviews Genetics 16 (1): 33-44. https://doi.org/10.1038/nrg3821
- Wright, Sewall. 1922. "Coefficients of inbreeding and relationship." The American Naturalist 56 (645): 330-338. https://doi.org/10.1086/279872
- "Endogamy." Wikipedia, for general background on the term. https://en.wikipedia.org/wiki/Endogamy