Same pedigree, different denominator: pairings versus offspring
Research note · Units of analysis
Abstract
The same breeding records can answer different questions depending on what we count. In a published pedigree of 237 sampled dogs, five of 30 distinct recorded sire–dam pairings had a nonzero expected offspring pedigree coefficient of inbreeding (COI): 16.7% of pairings. Those five pairings account for 22 of 207 sampled offspring with both parents recorded, or 10.6% of those offspring. The unweighted mean COI across pairings was 2.38%; weighting each pairing by its sampled offspring count gave 0.87%. Each summary is mathematically valid for its stated unit. Neither estimates a breed-wide rate or an individual dog’s genomic inbreeding.
What exactly is being counted?
A pairing-level summary gives each distinct sire–dam combination one vote, whether it has one sampled offspring or 28. An offspring-level summary gives each sampled dog with two recorded parents one vote. That repeats the same expected pedigree COI for siblings from a recorded pair, because the calculation uses their parents and the available ancestry. The offspring count describes how many samples in this file belong to that pair; it is not a count of independent breeding decisions or litters.
This distinction is useful to a breeder, lab, or registry reading a program-level figure. “Share of pairings with nonzero calculated COI” and “share of sampled offspring from those pairings” describe different sets of records. The number must travel with its denominator before it can be interpreted.
Data and method
We used the pedigree fields released with Campbell et al. (2016). The file contains 237 sample IDs from a Labrador Retriever–Greyhound research colony. Of these, 207 have a recorded sire and dam, representing 30 distinct pairings. We used the full available pedigree to calculate each pair’s expected offspring COI from half the tabular relationship between its parents. The 30-pair source table for this summary lists every pair, its sampled offspring count, and its calculated coefficient; the pedigree calculation method is downloadable.
For the pairing-level mean, we added the 30 coefficients and divided by 30. For the offspring-weighted mean, we multiplied each pair’s coefficient by its number of sampled offspring, added those products, and divided by 207. The nonzero shares use the same two denominators: distinct pairs and sampled offspring. Our short summary script reproduces both rows from the public 30-pair table.
Results
| Measure | Distinct recorded pairings | Sampled offspring with two recorded parents |
|---|---|---|
| Denominator | 30 | 207 |
| Count associated with nonzero calculated COI | 5 | 22 |
| Share associated with nonzero calculated COI | 16.7% | 10.6% |
| Mean expected offspring pedigree COI | 2.38% | 0.87% |
The difference comes from how the 207 samples are distributed across pairs. The pairing PFZ19E09 × PFZ18D01 has a full-record expected offspring COI of 36.33%, but only one sampled offspring. The two pairs with the largest sampled offspring counts—PFZ24C06 × PFZ26A06 with 28, and PFZ26F05 × PFZ26G05 with 26—each have 0.00% within the available pedigree. Their zeros enter the pair-level mean once each, but enter the offspring-weighted mean 28 and 26 times.
There is no contradiction between the two rows of results. A person comparing distinct recorded mating combinations needs the first column. A person describing sampled dogs represented by those combinations needs the second. Neither column says how many total litters were produced, how many unsampled dogs were born, or whether a mating was healthy or suitable.
Why this matters for AnimalTrace
Search, saved updates, and Match operate on records that can be grouped in several ways: animals, parent links, pairings, and lines. A program summary should name the unit it counts. A newly imported group of siblings can raise an animal count without adding a new sire–dam pairing. A newly documented pairing can change a pairing count even before many offspring appear in the available records. The underlying records need to remain inspectable so those changes have an explanation.
For an individual prospective match, the pair-specific result and its evidence are more relevant than either colony-wide average. Our worked AnimalTrace example shows one pairing calculation and its recorded ancestor; our parent-link study shows why a later record correction can change that calculation.
Limits and reproducibility
This is a descriptive reaggregation of a selected 237-sample research pedigree, not a population study. Siblings and other relatives are present, and the two summary rows are calculated from the same records; we report no significance test. The source has founders with ancestry unrecorded in this file, so a zero means no overlap found in the available pedigree. The .fam fields used here contain no dates, litter IDs, clinical outcomes, or population sampling weights. We cannot infer a time trend, a breed-wide mating rate, or genomic inbreeding from these summaries.
- Source study: Campbell et al., 2016, A Pedigree-Based Map of Recombination in the Domestic Dog Genome.
- Source data: Authors’ public repository,
new_pedigree_dogs.faminsidedog_genotype_data.tar.gz. - Pair-level data: All 30 recorded pairings and their full-record COI values.
- New summary: Download the two-unit results and the standard-library Python summary script.
This AnimalTrace research note has not undergone external peer review.