The Measured Values

The reference tables — effective size across life, measured selection coefficients, mutation rates, genetic maps, lethal equivalents, and every population founded from twenty individuals or fewer — each row tagged measured, model output, or assumed.

What this page is

Companion to the population-genetics reference.  Every table here carries units, a convention, a source with a locator, and a tag saying whether the number was measured, is a model output, or was assumed.

That last column is the point of the page.  Arguments on both sides routinely treat a model output as though it were a measurement, and the difference decides how much weight a number can carry.  The conventions the tables obey are on the conventions page, and every computed figure is reproducible with the Pop Gen Calculator.

Where a number could not be verified against a primary source, the cell says so rather than guessing.

1. Effective population size across life

From the only dataset with both a directly measured mutation rate and a measured natural-population diversity for the same species — 144 usable species.1

Species Ne
Lowest measured vervet monkey, Chlorocebus aethiops 1.09 × 104
walrus just above
human 2.21 × 104 — fifth from the bottom of 144
Median species — 3.8 × 105
Highest measured Pseudomonas fluorescens SBW25 2.69 × 109

Span:  5.4 orders of magnitude.  Tag: model output — each value divides a measured diversity by a measured mutation rate.  Lynch uses θ/(4μ) for diploids and θ/(2μ) for haploids, with θ = πs/(1 − πs);  the ploidy factor is not cosmetic, because reading the bacterial rows as πs/(4μ) would put Pseudomonas fluorescens at 9.0 × 108 instead of 2.7 × 109.

This is a lower bound on the real span, and the reason is sampling rather than biology.  The bottom of the table is all vertebrates, because vertebrates are what get parent-offspring trio sequencing.  The top is bacteria with mutation-accumulation lines.  Obligate endosymbionts, island endemics and the great majority of eukaryotes are absent entirely.

A second author, fifteen years earlier, using a different method:2

Species Charlesworth 2009 Lynch 2023 Ratio
Human 10,400 22,118 2.13
Drosophila melanogaster 1,150,000 627,824 0.55
Caenorhabditis elegans 80,000 124,155 1.55
Escherichia coli 25,000,000 170,918,080 6.8

They agree on the order of magnitude and not on the digit.  For humans the entire difference is the mutation rate assumed.  That is the right size of error bar to carry on any cross-species Ne claim.

Consequence: the drift barrier moves with Ne

Population Ne Effectively neutral below
Human 2.2 × 104 |s| ≈ 2.3 × 10-5
Median species in the table 3.8 × 105 |s| ≈ 1.3 × 10-6
Pseudomonas fluorescens 2.7 × 109 |s| ≈ 1.9 × 10-10

A mutation costing one part in 100,000 is invisible to a human population and strongly selected in a bacterium.  This is why "is this mutation neutral?" is not a question about the mutation.

2. Ne/N, and what each ratio divides by

Ratio Value Denominator Tag Source
Comprehensive wildlife estimates 0.10–0.11 adult census measured, meta-analysis Frankham 1995, abstract3
Conservation working default 0.1–0.2 adult census assumed Frankham 2014 Table 1, p. 584
All estimates, unfiltered 0.34 average mixed — "estimating different things" measured, meta-analysis Frankham 1995, p. 983
Re-analysis of correctly linked estimates 0.231 median adult census measured Palstra & Fraser 20125
Modern humans ≈ 1/3 (0.34, 95% CI 0.27–0.51) total census measured, from IBD segments Browning & Browning 20156
Modelled from vital rates, 63 species 0.19 to 3.69 adult census model output Waples 20137
The largest of 192 estimates, and it is human 0.994 breeding adults, one model a human figure, but not a comprehensive one Frankham 1995, p. 983

Two things this table exists to prevent.  The 0.994 gets quoted as "the human Ne/N."  It is the top of a range Frankham is about to argue is unreliable, and his very next sentence says the estimates "were estimating different things."  And nine of Waples' 63 species had Ne greater than N — the ratio is not bounded above by one, and it is not a constant of nature.

A fourth "one third" lurks and is a trap:  Frankham also notes that for humans "approximately 1/3 … of the census size" are sexually mature adults.  That is an adults-to-total conversion, not an Ne/N ratio.  Multiplying by it twice is a real error waiting to happen.

3. Mutation rates

3.1 The human rate, by study

Seven independent whole-genome trio studies across three generations of sequencing technology.  All measured.

Estimate Note
1.1 to 1.29 × 10-8 per site per generation the full span of every serious modern estimate
≈ 70 point mutations per newborn the same number, per diploid genome
98 to 206 events per transmission once indels, tandem-repeat length changes and structural variants are counted
16% of observed de novo SNVs are post-zygotic they arose in the embryo, so strictly they are not part of μ

Most of the remaining disagreement is the callable fraction, and the papers say so.  Rahbari observed 64 mutations per child and reported 76.9 after correcting for regions the pipeline could not reach8 — a 20% correction, which is the whole difference between their 1.28 × 10-8 and Keightley's 1.1 × 10-8.  Porubsky's 2025 study reached 91.9% of the autosomes with five sequencing technologies and near-complete assemblies, and 8% was still invisible.9

3.2 The paternal age effect

Parent Extra mutations per year of parental age Source
Father 1.51 (95% CI 1.45–1.57) Jónsson 2017, 1,548 trios10
Mother 0.37 same
Between-family range, paternal slope 0.19 to 3.24 Sasani 2019, 33 CEPH families11

Tag: measured.  The consequence matters for every date:  because of this effect, the per-generation rate and the per-year rate move in opposite directions with generation time.  Shorter generations mean fewer mutations per generation but more generations per century, so the per-year rate rises.  That is the mechanism behind the measured great-ape "slowdown."

3.3 Across life

Group Per site per generation
Bacteria, unicellular eukaryotes 10-11 to 10-9
Multicellular eukaryotes 10-9 to 10-8
DNA viruses (per nucleotide per cell infection) 10-8 to 10-6
RNA viruses (same unit) 10-6 to 10-4

About 1,000-fold variation per generation across the tree of life, and over 5,000-fold among eukaryotes alone in the largest meta-analysis.  Per cell division the ranking inverts:  the human germline runs at 0.06 × 10-9 per site per division, "approximately one fifth that in unicellular organisms."12

3.4 Do rates change?

Finding Effect Tag
E. coli mismatch-repair knockout 138-fold increase measured
Natural segregating Mutyh allele in mice +50% C>A rate measured
European TCC→TTC burst, ~15,000–2,000 years ago ~+50%, then back down measured
Chernobyl cleanup workers' children, paternal dose to 4,080 mGy no detectable effect: −0.0007 per mGy, 95% CI −0.003 to 0.002, P = 0.56 measured13
Chimpanzee, gorilla, orangutan vs human, per year 1.48 ± 0.08 times faster than humans today measured14

A young-earth reader will want past rates to have been higher.  The measured evidence points the other way for humans — the great-ape comparison is read as a recent human slowdown.  What is genuinely open is whether that slowdown is biochemical or just a change in generation time;  generation time alone can account for the whole gap.

3.5 U, the genomic deleterious rate

Route Range for humans What it measures
Sequence: μ × constrained fraction × 2 2.2 to 22 how many mutations land in sequence that looks conserved
Fitness: mutation-accumulation lines 0.01 to 1.0 how many mutations produce a detectable fitness decline

An order of magnitude apart, and they must not be blurred.  The full audit of what has been measured about mutational effects is at what has actually been measured about mutations.

4. Measured selection coefficients

Every row is model output unless marked otherwise:  a selection coefficient is fitted to a trajectory, not read off an instrument.  The convention column is not optional — see conventions.

System Allele s as published Error bar Convention
Human, European LCT −13910T (lactase persistence) 0.014–0.15 range over assumed starting frequency; no CI s against the non-persister homozygote
Human, Scandinavian LCT −13910T 0.09–0.19 same same
Human, Britain LCT, ancient-DNA time series 0.064 mean over 4,500 y; peak ≈ 0.08; ≈ 0.02 in the last 50 generations ±0.01–0.02 per time point per allele
Human, Tanzania LCT −14010C 0.070 dominant / 0.097 additive 95% CI 0.023–0.143 / 0.040–0.145 homozygote, both models given
Human, Central Asia LCT in herders selection rejected — at s = 0.04 the allele "should have reached 51%"; observed ~5% — homozygote
Human, malaria zones HbS (sickle) 0.092 against AA not given heterozygote-centred
Human, Africa G6PD A− 0.044 vs 0.25 the two published estimates differ 6-fold homozygote
Human, Europe SLC24A5 (light skin) 0.16 dominant / 0.08 additive 95% CI 0.05–0.73 / 0.02–0.55 homozygote
Human, Europe HERC2 (blue eyes), aDNA 0.017 0.02–0.04 in text per allele
Human, East Asia EDAR 370A 0.114 vs 0.0744 95% CI 0.030–0.186 vs 0.0657–0.0831; the two methods differ 40% homozygote
Human, Europe CCR5-Δ32, the 1998 claim 0.23 or 0.37 none — a required-value calculation, not a fit —
Human, Europe CCR5-Δ32, the 2005 retraction no selection detected; allele >5,000 y old EHH p = 0.49 —
Human, Europe height, polygenic — 2012 claim 10-3 to 10-5 per allele none per allele
Human, Europe height — 2019 re-analysis signal "strongly attenuated or absent"; earlier work "confounded by population stratification" — —

Four things this table is designed to show.

The published human range is roughly 0.01 to 0.2, and it is censored at the bottom rather than measured there.  The best current method cannot resolve anything below about s = 0.02:  "we cannot reliably distinguish smaller selection coefficients from zero."15

Two of the most-cited claims were later withdrawn by the same field.  CCR5-Δ32 and the polygenic height signal are both in the table as negative results, at the same weight as the positives.

Where two methods are applied to the same allele they disagree by 40% to 6-fold.  G6PD and EDAR are the clean examples.

And the published values are systematically too large.  Kingsolver's meta-analysis found only 25% of linear selection gradients individually significant, a median sample size of 134, and an explicit file-drawer effect among small samples.16  Morrissey and Hadfield found that about 90% of apparent temporal variation in selection is sampling error.  Hereford and colleagues wrote in print that their median cannot be "representative of selection on all traits."17

5. Recombination

Quantity Value Tag
Human autosomal map, male 2,602 cM measured, summed from the published map files
Human autosomal map, female 4,180 cM same
Human autosomal map, sex-averaged 3,391 cM over 2,754 Mb = 1.23 cM/Mb same18
Female-to-male crossover ratio 1.61 derived
Crossovers counted in 91 single sperm 22.8 ± 3.7 (SD) measured directly19
Crossovers counted in 51 oocytes and embryos 41.6 ± 11.3 (SD) measured directly20
Share of the genome carrying 80% of recombination 10.6% sex-averaged, 6.1% in males, on the refined map; 38.5% on the unrefined pedigree map model output — the refinement is an MCMC fit against HapMap LD21

A Morgan is a count, not a distance:  the total map length in Morgans is the average number of crossovers per meiosis.  So "the human genome is about 34 Morgans" and "a human gamete carries about 34 crossovers" are the same sentence, and the direct counts above confirm the sex-averaged and female maps to within 5%, and the male map to within 13%.

Hotspot positions are set by PRDM9 and are almost completely different between humans and chimpanzees, despite 99% sequence identity.

What linkage actually costs

Holding everything else fixed and changing only the genetic map, in a Wright-Fisher simulation over 3,000 generations:

Genetic map Deleterious fixations
Free recombination 729 ± 38
Human sex-averaged map (33.9 M) 705 ± 38
Human female map 715 ± 46
Human male map 710 ± 25
One tenth of it 994 ± 90
1.1 Morgans about 1,650 across three fresh seeds (1,631–1,695); earlier runs recorded 1,403–1,591, one of them 1,484
No crossing over at all 3,712 ± 233 — about 5×

At a real human map length, linkage costs essentially nothing at the genome scale.  The cost is real in low-recombination regions — where an excess of damaging alleles has been measured in humans — and severe where recombination is absent, which is the mitochondrion and the Y.

Background selection reduces to B = e^(−U/M).  At the whole human autosome that is a 4% reduction in Ne;  the measured genome-average reduction is 19–26%, because deleterious mutations concentrate near genes where the local ratio is much higher.  So Ne is a function of genomic position, not one number for a species.

6. Inbreeding and lethal equivalents

Quantity Value Units Tag
Morton, Crow & Muller 1956 1.5–2.5 per gamete ("or 3–5 per zygote") measured22
Bittles & Neel 1994; Bittles & Black 2010b 0.56–0.7 per haploid genome measured, but not read here — the range is quoted from Gao 2015 p. 1249; neither primary was obtained23
Gao 2015, Hutterite pedigrees 0.29 recessive lethals (95% CI 0.10–0.84) per haploid set measured24
Ralls, Ballou & Templeton 1988 median +1.57 per gamete; per-individual mean 4.6, median 3.1 40 captive mammal populations, 38 species measured25

The measured human value has fallen by a factor of three since 1956 — a range this page takes from Gao 2015's citation of Bittles & Neel 1994, not from the primary, which has not been read here.

And a note the ark-bottleneck argument needs:  Ralls et al. never print "3.14."  The widely quoted figure is 2 × 1.57.  In their sample B runs from −0.68 to +15.16 per gamete — −1.4 to 30.3 per individual — and only 9 of the 40 slopes (23%) were significantly greater than zero.

The measured cost of first-cousin marriage

Outcome Effect Sample Tag
Excess mortality, ~6 months' gestation to age 10 3.5% meta-analysis of 69 populations measured23
Odds of having children −55%, 95% CI 44–66% 1.4 million people, confirmed within full-sibling pairs measured26
Height −1.2 cm same programme measured
Education −10 months same measured

The mortality figure is the small one and it is the one everybody quotes.  Any bottleneck argument citing only mortality is cherry-picking.

7. Populations founded from twenty individuals or fewer

The strongest real-world evidence about very small founder populations.  Empty cells mean the number was not found in a primary source and should not be invented.

Population Founding number and date Generations Census now Genetic cost
Chatham Island black robin fell to 7 in 1976, then 5; recovery from one breeding pair ~20 two populations heterozygosity halved; FROH 0.63 vs 0.17 historical; "arguably the most inbred bird species in the world"; ~20% still show the maladaptive rim-laying trait27
Mauritius kestrel 4 known wild birds, 1974;28 one breeding pair founded the recovery ~30 ~250 adults allelic diversity −55%, heterozygosity −57%29
Pink pigeon ~10 by 1990 ~15–20 ~500 ~15 lethal equivalents for longevity; simulations predict wild extinction within 100 years without intervention
Mauritius parakeet ~20 in 1986 ~15 ~650 —
Northern elephant seal ~20 in 1892 22 >220,000 heterozygosity 8-fold down and unrecovered, 0.00142 → 0.000176 per bp; measured fitness reductions in breeding and foraging30
Whooping crane ~20 by 1944; recovery stock of 16 ~10–15 ~840 70% of historical diversity lost; authors argue against downlisting31
European bison 12 founders, 7 in the Lowland line ~15–20 ~9,000+ pedigree inbreeding 44% in the Lowland lineage; skull deformities reported
Przewalski's horse 12 wild-caught founders ~12–15 2,109 reduced heterozygosity; domestic-horse introgression to 31.1% in individual genomes
Norfolk Island boobook a single female in 1986, plus two introduced males of another subspecies ~10 unknown pedigree F reached 0.375; 54–55% of original neutral alleles may eventually be lost

Eight of the nine are recovered demographically and all nine are impaired genetically;  the Norfolk Island boobook's current census was not found in a primary source.  That is the honest summary, and it cuts both ways:  a founding pair is survivable, and the cost is permanent on any timescale that matters.

The counter-cases

Population Founders Outcome
Isle Royale wolves 2–3, closed system down to two animals in about 70 years; 23–48% of the genome in runs of homozygosity; 33% vertebral malformation32
Ashkenazi Jews Ne ≈ 250–420, 25–32 generations ago heterozygosity 2.4% higher than the European comparison group33

The second row belongs in the same table as the first.  A bottleneck of a few hundred, followed by growth, leaves almost no heterozygosity signature at all.

8. What a founding pair keeps

Generations at Ne = 2 Heterozygosity kept Inbreeding F
1 100% 0
2 75% 0.250
3 62.5% 0.375
4 50% 0.500
5 40.6% 0.594
10 14.1% 0.859
20 1.7% 0.983

Exact dioecious recursion, F(t) = ¼(1 + 2F(t−1) + F(t−2)).  Not Wright's shortcut (1 − 1/2Ne)^t, which decays at a flat 25% per generation where the exact series settles to about 19.1%, and understates the surviving heterozygosity by 42% by the fifth generation — 0.2373 against the exact 0.4063.  The Pop Gen Calculator prints both series side by side in its bottleneck-two table, and treats the shortcut as a negative control it is required to reject.

Depth is cheap;  duration is expensive.  And the baseline matters:  every figure above is against the source population's heterozygosity.  Four different correct answers circulate for "what a founding pair keeps" — 100%, 75%, 65% and 50% — and they differ only in what they are a fraction of.  A percentage without its baseline is unusable.

Allelic richness behaves completely differently, and the amount lost depends on how diverse the source locus was:

θ of the source locus Alleles retained by two founders
0.16 36.2%
0.30 23.4%
1 12.6%
4 5.2%
10 2.6%

The first two rows are Nei, Maruyama and Chakraborty's own published values;34  the rest is the same Ewens sampling formula at modern diversity levels, against a source population of 8.1 × 106 gene copies — the size implied by Nei's own E(K) = 3.599.  The answer moves by a factor of two across plausible source sizes, so that number has to be printed with the θ.  A retained-allele percentage without its θ is meaningless — and the high-θ end is where a young-earth model is in most trouble, because those are the HLA-like loci.

9. Response to selection

Quantity Value Source
Heritability of fitness itself 2.99%, 95% CI [0.80, 6.60] Bonnet 2022, 19 wild populations35
Additive genetic variance in fitness VA(w) = 0.185 same
Maximum response for a trait at h² = 0.3 0.24 phenotypic SD per generation same
Observed phenotypic change in the wild "rarely above 0.03 standard deviations" same
Life-history h², mean of 341 estimates 0.262 Mousseau & Roff 1987, from a 1,120-estimate compilation36
Morphology h², same study 0.461 same
Illinois maize, realised h² for kernel oil 0.34 → 0.31 → 0.14 → 0.08 across four segments Dudley 1974, Table 9, p. 19437

The measured ceiling on selection is about eight times the observed rate of change.  That cuts against genetic-entropy-style arguments and belongs in the table at the same weight as everything else.

The Illinois experiment has run since 1896.  The trait kept moving while the heritability behind it shrank four-fold — and the high-oil line beat Robertson's predicted ceiling, which mainstream quantitative genetics attributes to new mutation:  more than 100 generations at Ne under 30, "yet there is a continuing near linear increase in percentage of oil in the kernel that can be explained only by mutational input."38  The low-oil line did hit a hard limit, because it has almost no oil left in its kernel.

Sources

The full source register, with every locator and the route used to obtain each item, is in the eighteen research dossiers under html/unpublished-ai/population-genetics/research/.  The principal sources for the tables above.  Entries without an inline marker are named in their table's row or caption instead.

  1. 1.

    Lynch, Michael, Farhan Ali, Tongtong Lin, Yaohai Wang, Jiahao Ni & Hongan Long.  "The divergence of mutation rates and spectra across the Tree of Life."  EMBO Reports 24(10):e57561.  2023.  Dataset EV1.  Rebuilt independently here:  the Ne column is a pasted value, not a formula, and it is θ/(4μ) for diploids and θ/(2μ) for haploids with θ = πs/(1 − πs).  On that reading 113 of the 144 usable rows reproduce to better than 0.05%;  the 31 that do not are the species blocks where Lynch averaged two or more published πs or u values, which is a property of the spreadsheet rather than an error in it. ↩

  2. 2.

    Charlesworth, Brian.  "Effective population size and patterns of molecular evolution and variation."  Nature Reviews Genetics 10(3):195-205.  2009.  Table 1, print page 199.  Also on that page, the field's own admission:  "There is, however, no satisfactory treatment of populations in which individuals reproduce more or less continuously, such as humans and many tropical species." ↩

  3. 3.

    Frankham, Richard.  "Effective population size/adult population size ratios in wildlife:  a review."  Genetical Research 66(2):95-107.  1995.  Abstract for the 0.10-0.11 comprehensive figure;  p. 98 for the range, the 0.34 average and the 0.994 maximum;  p. 101 for the adults-to-total conversion. ↩ ↩ ↩

  4. 4.

    Frankham, Richard, Corey J. A. Bradshaw & Barry W. Brook.  Biological Conservation 170:56-63.  2014.  Table 1, p. 58. ↩

  5. 5.

    Palstra, Friso P. & Dylan J. Fraser.  "Effective/census population size ratio estimation:  a compendium and appraisal."  Ecology and Evolution 2(9):2357-2365.  2012. ↩

  6. 6.

    Browning, Sharon R. & Brian L. Browning.  American Journal of Human Genetics 97(3):404-418.  2015. ↩

  7. 7.

    Waples, Robin S., Gordon Luikart, James R. Faulkner & David A. Tallmon.  Proceedings of the Royal Society B 280:20131339.  2013. ↩

  8. 8.

    Rahbari, Raheleh, et al.  "Timing, rates and spectra of human germline mutation."  Nature Genetics 48:126-133.  2016. ↩

  9. 9.

    Porubsky, David, et al.  Nature, 2025.  Five sequencing technologies and near-complete assemblies;  91.9% of the autosomes callable. ↩

  10. 10.

    Jónsson, Hákon, et al.  Nature 549:519-522.  2017.  1,548 Icelandic trios. ↩

  11. 11.

    Sasani, Thomas A., et al.  "Large, three-generation human families reveal post-zygotic mosaicism and variability in germline mutation accumulation."  eLife 8:e46922.  2019. ↩

  12. 12.

    Lynch, Michael.  "Rate, molecular spectrum, and consequences of human mutation."  PNAS 107(3):961-968.  2010. ↩

  13. 13.

    Yeager, Meredith, et al.  Science 372(6543):725-729.  2021. ↩

  14. 14.

    Besenbacher, Søren, Christina Hvilsom, Tomas Marques-Bonet, Thomas Mailund & Mikkel Heide Schierup.  "Direct estimation of mutations in great apes reconciles phylogenetic dating."  Nature Ecology & Evolution 3:286-292.  2019. ↩

  15. 15.

    Mathieson, Iain & Jonathan Terhorst.  "Direct detection of natural selection in Bronze Age Britain."  Genome Research 32:2057-2067.  2022.  Page 2062 for the detection floor;  Table 1 for the ancient-DNA selection coefficients. ↩

  16. 16.

    Kingsolver, Joel G., et al.  "The strength of phenotypic selection in natural populations."  The American Naturalist 157(3):245-261.  2001. ↩

  17. 17.

    Hereford, Joe, Thomas F. Hansen & David Houle.  "Comparing strengths of directional selection:  how strong is strong?"  Evolution 58(10):2133-2143.  2004. ↩

  18. 18.

    Halldórsson, Bjarni V., et al.  "Characterizing mutagenic effects of recombination through a sequence-level genetic map."  Science 363(6425):eaau1043.  2019.  The autosomal map lengths above were summed here from Data S1-S3 rather than quoted. ↩

  19. 19.

    Wang, Jianbin, H. Christina Fan, Barry Behr & Stephen R. Quake.  "Genome-wide single-cell analysis of recombination activity and de novo mutation rates in human sperm."  Cell 150(2):402-412.  2012. ↩

  20. 20.

    Ottolini, Christian S., et al.  "Genome-wide recombination and chromosome segregation in human oocytes and embryos reveal selection for maternal recombination rates."  Nature Genetics 47(7):727-735.  2015. ↩

  21. 21.

    Bhérer, Claude, Christopher L. Campbell & Adam Auton.  "Refined genetic maps reveal sexual dimorphism in human meiotic recombination at multiple scales."  Nature Communications 8:14994.  2017.  The 10.6%, 6.1%, 8.8% and 38.5% figures are all in the same paragraph:  "In the sex-averaged refined map, 80% of recombination is localized in 10.6% of the genome, which is comparable to 13.5% in the HapMap, and much more concentrated than the 38.5% observed for the unrefined map. … Male recombination occurs in a narrower portion of the genome than female, with 80% of male recombination occurring in 6.1% of sequence compared to 8.8% for female." ↩

  22. 22.

    Morton, N. E., J. F. Crow & H. J. Muller.  PNAS 42(11):855-863.  1956.  Page 859 for the per-gamete definition. ↩

  23. 23.

    Bittles, Alan H. & Michael L. Black.  "Consanguinity, human evolution, and complex diseases."  PNAS 107(suppl 1):1779-1786.  2010.  Page 1781 for the 3.5% excess-mortality meta-analysis.  This paper prints no lethal-equivalent value;  the 0.56–0.7 range in the table above is quoted by Gao et al. 2015 (Genetics 199:1249) from Bittles & Neel 1994, Nature Genetics 8:117-121, and Bittles & Black 2010b, Early Human Development 86:737-741, neither of which has been obtained here. ↩ ↩

  24. 24.

    Gao, Ziyue, Darrel Waggoner, Matthew Stephens, Carole Ober & Molly Przeworski.  "An estimate of the average number of recessive lethal mutations carried by humans."  Genetics 199(4):1243-1254.  2015. ↩

  25. 25.

    Ralls, Katherine, Jonathan D. Ballou & Alan Templeton.  Conservation Biology 2(2):185-193.  1988. ↩

  26. 26.

    Clark, David W., et al.  Nature Communications 10:4957.  2019. ↩

  27. 27.

    von Seth, Johanna, et al.  BMC Genomics 23:747.  2022. ↩

  28. 28.

    Wang, Xuejing, et al.  "Conservation rescued the Mauritius kestrel from extinction but not from genomic erosion."  bioRxiv preprint, 2026.  doi:10.64898/2026.06.23.733933.  Preprint;  not yet peer reviewed.  Abstract:  "The Mauritius kestrel (Falco punctatus) declined to four known wild birds in 1974 before intensive management recovered the population." ↩

  29. 29.

    Groombridge, Jim J., Carl G. Jones, Michael W. Bruford & Richard A. Nichols.  "'Ghost' alleles of the Mauritius kestrel."  Nature 403(6770):616.  2000.  Page 616:  "Across all loci, allelic diversity fell by 55%, and heterozygosity by 57% (Table 1)." ↩

  30. 30.

    Hoelzel, A. Rus, et al.  Nature Ecology & Evolution 8:686-694.  2024.  The "132 years (~22 generations)" figure is on p. 687, the last paragraph of the Introduction — not p. 686 as it was first recorded here. ↩

  31. 31.

    Fontsere, Claudia, et al.  Molecular Ecology 34:e70088.  2025. ↩

  32. 32.

    Robinson, Jacqueline A., et al.  "Genomic signatures of extensive inbreeding in Isle Royale wolves, a population on the threshold of extinction."  Science Advances 5:eaau0757.  2019. ↩

  33. 33.

    Carmi, Shai, et al.  "Sequencing an Ashkenazi reference panel supports population-targeted personal genomics and illuminates Jewish and European origins."  Nature Communications 5:4835.  2014. ↩

  34. 34.

    Nei, Masatoshi, Takeo Maruyama & Ranajit Chakraborty.  "The Bottleneck Effect and Genetic Variability in Populations."  Evolution 29(1):1-10.  1975.  Table 3, p. 8. ↩

  35. 35.

    Bonnet, Timothée, et al.  Science 376(6596):1012-1016.  2022.  Pages 1014-1015. ↩

  36. 36.

    Mousseau, Timothy A. & Derek A. Roff.  "Natural selection and the heritability of fitness components."  Heredity 59:181-197.  1987.  Table 2, p. 186. ↩

  37. 37.

    Dudley, J. W., R. J. Lambert & D. E. Alexander.  Seventy generations of selection for oil and protein in maize.  1974.  Table 9, p. 194. ↩

  38. 38.

    Hill, William G.  Genetics 202(3):877-881.  2016.  Page 880. ↩

  39. 39.

    Joshi, Peter K., et al.  "Directional dominance on stature and cognition in diverse human populations."  Nature 523:459-462.  2015.