Molecular Clocks

"There may thus exist a molecular evolutionary clock." — Emile Zuckerkandl and Linus Pauling, 1965¹

"Here we show that the precision of molecular clock estimates of times has been grossly over-estimated." — Mario dos Reis and colleagues, Current Biology, 2015²

Summary

A molecular clock converts the number of DNA or protein differences between two species into the time since they split.  This article judges how reliable those dates are by the tests the method can be put to, and tests the young-earth criticisms and the young-earth clocks the same way.  Ages in years below are the mainstream's figures, quoted to show the size of each effect.  The article does not adopt them.

1. The sequences measure rate times time, never time alone. Every date takes its time from a calibration: a fossil, a dated geological event, or the known dates of the samples. For deep nodes the fossil calibrations carry nearly all of the uncertainty, and no amount of sequence lowers it below about ±16 % of the date.³
2. The young-earth circularity charge is correct for fossil-calibrated dates. Such a date spreads the radiometric timescale across a tree and cannot test it, as mainstream authors also write.⁴
3. Calibrations that use no fossils do not give young dates. Rates counted in families put the human-chimp split earlier, not later, and Y-chromosome rates from pedigrees match those from ancient genomes. Across 68 vertebrate species, family rates loosely track long-term rates.⁵
4. The clock is irregular. Substitutions arrive two to six and a half times more raggedly than a steady clock would allow,⁶ and rates differ with generation time, body size and metabolic rate. Variation of this size widens error bars. It is thousands of times smaller than would be needed to fit mainstream dates into a few thousand years, a fit no young-earth model attempts by mutation alone.
5. Rates measured over short spans run fast. Where the cause has been traced, it lies in what was counted: short-lived heteroplasmies in penguins,⁷ protein-changing mutations that selection removes in fruit flies.⁸ Carter's claim that family rates are "several orders of magnitude" faster⁹ fails on nuclear DNA and the Y chromosome and reaches at most about ninetyfold on mitochondrial DNA.
6. Known-answer tests favor the clock inside its sampling window and expose it beyond. It dated the SARS-CoV-2 ancestor to within weeks¹⁰ and recovered the year of a 1960 HIV sample.¹¹ Stretched over centuries it fails toward youth: modern hepatitis B samples gave an ancestor 229 years old,¹² but a strain close to a modern lineage turned up in a skeleton from 1020–1116 AD.¹³
7. Clock dates for the great radiations precede the oldest fossils. The gaps run from about 13 million years for placental mammals to over 100 for animals and flowering plants on the mainstream timescale, and the clock date is almost always the older. The field explains them by late fossils, rate bursts and calibration choices, and its own authors dispute the size in print.¹⁴¹⁵ The comparison is worked group by group in Do Clock Dates Agree with the First Fossils?.
8. Jeanson's mitochondrial clock fits 6,000 years only on mutations that do not last. In fruit flies, keeping only the kinds of mutation that build up between species leaves his prediction 4.7 to 110 times short.¹⁶ His model also needs the flies' nuclear differences to be created, and it sets aside the ancient DNA that would test it.¹⁷
9. Young-earth geneticists disagree about whether the clock was constant,¹⁸ and for differences between created kinds they use no clock at all, holding those differences were made.⁹

Verdict.  A deep clock date is an estimate on the mainstream timescale with an uncertainty of at least ±16 % and a lean toward the old side, not independent evidence for that timescale.  The young-earth case gains nothing from the clock's weaknesses, because the fossil-free rates give long dates too, and the young-earth clock passes only by counting mutations that do not persist.

Part 1: How a clock date is made

The idea.  If changes in DNA or protein pile up at a steady rate, the number of differences between two species measures the time since they split.  Emile Zuckerkandl and Linus Pauling first used the idea in 1962, to date the gene duplication that made the alpha and beta chains of hemoglobin.¹⁹  In 1965 they gave it its name:  "There may thus exist a molecular evolutionary clock."¹  They offered it as a hypothesis.  Two pages later they wrote, "The verdict of the facts will be looked forward to."¹  They also warned that a date taken from a single protein "cannot be expected to be satisfactory, except by coincidence."¹

Their own numbers.  Human hemoglobin chains differ from those of horse, pig, cattle and rabbit at an average of 22 of about 143.5 amino acids.  Zuckerkandl and Pauling assumed those mammals shared an ancestor 80 million years ago, and so got a rate of "about 1 in 1 billion years" for any one site on one line of descent.¹  The whole calculation rests on the assumed 80 million years.  That is the first lesson of the method, and it has never gone away:  the molecules give a count, and something else has to give the time.

Why a steady rate might exist.  Motoo Kimura supplied the reason in 1968 and 1969.  If a mutation has no effect on fitness, its chance of spreading through the whole population equals its starting frequency.  New copies arise in proportion to population size, and each fixes with a chance inversely proportional to it, so population size cancels.  In Kimura's words, "for neutral mutations the rate of gene substitution in a population is equal to the rate of production of new mutations per gamete."²⁰  That predicts a clock that ticks per generation.  Kimura himself found it "extraordinary" that real protein rates "mainly depend on time measured in years but are almost independent of generation time."²⁰  Whether most substitutions really are neutral is argued in Neutral Theory.  For this article neutrality is only the premise the clock stands on.  Zuckerkandl and Pauling's own mechanism was not strictly neutral.  They pictured sites shuttling among a few amino acids, "the ones or the others" being "slightly more advantageous for the organism" according to circumstances.¹

The arithmetic.  A date is the distance divided by twice the rate:  t = d / 2r.  The distance d counts changes per site along both branches back to the common ancestor, which is why the rate r, per branch, is doubled.  Where a site may have changed more than once, the raw count is corrected upward first.  With 1 % divergence and a rate of one change per site per billion years, t = 0.01 / (2 × 10⁻⁹) = 5 million years.²¹

The catch:  the data give only rate times time.  A branch with 1 % of change could be 5 million years at one rate or 50,000 years at a rate a hundred times faster.  Mario dos Reis, Philip Donoghue and Ziheng Yang put it plainly in their 2016 review:  "the molecular sequence data provide information about the molecular distances (the product of times and rates) but not about times and rates separately."¹⁹  Every absolute date therefore comes from outside the sequences, from a calibration.  Part 2 is about where calibrations come from.

Strict and relaxed clocks.  A strict clock gives the whole tree one rate.  Until the late 1990s most studies used one, often "even for distantly related species," with fossil dates entered as exact points.¹⁹  Modern programs relax the clock.  Each branch gets its own rate, drawn either independently from a common distribution or inherited with some drift from its parent branch, and the calibrations enter as probability curves rather than points.  BEAST, the program most used for viruses and ancient DNA, is "Particularly strong for the analysis of serially sampled DNA sequences";  MCMCTree is built for genome-scale alignments.¹⁹  The spread of branch rates is measured as a coefficient of variation, where 0 is a perfect clock.  Alexei Drummond and colleagues found 0.32 in marsupials, 0.39 in dengue virus and 0.51 in influenza.²²  Real lineages differ in rate by a third to a half of the average.

A precision floor that more data cannot lower.  Because rate and time are confounded, adding sequence shrinks the error only down to a floor set by the calibrations.  Dos Reis and Yang dated six primates on 8,708,584 aligned sites.  They found that "for every one million years of divergence, 0.322 million years are added to the CI width," so that "nearly all the uncertainty in the posterior time estimate is due to uncertainties in the fossil calibrations, and adding more sequence data is unlikely to produce more precise estimates."³  A width of 0.322 × t is about ±16 % of any date, at any age.  On the mainstream timescale a node dated to 66 million years carries an interval at least 21 million years wide.²¹  When dos Reis and colleagues ran every calibration strategy, clock model and data partition for the origin of animals, the combined interval for crown animals ran from 833 to 650 million years ago.²  Their verdict was that "the precision of molecular clock estimates of times has been grossly over-estimated,"² and that "the limitations of the fossil record and the confounding effect of times and rates will remain."²

Part 2: Calibration, and the charge of circularity

Every clock date borrows its time from outside the sequences.  There are three places to borrow it:  a fossil, a dated geological event, or the known dates of the samples themselves, which includes a pedigree, where the generations are counted.¹⁹  Each has its own error, and the errors do not point the same way.

Fossils give minimum ages.  The oldest fossil of a group shows that the group existed by then.  It does not show when the group began.  Michael Benton and Philip Donoghue, writing the field's reference guide to fossil calibrations, put it this way:  "Fossils can provide good "minimum" age estimates for branches in the tree, but "maximum" constraints on those ages are poorer."²³  If a node is pinned at its oldest fossil, the rate comes out too fast by the ratio of the true age to the fossil age, and every other date on the tree comes out too young by the same ratio.  The same paper reports that the split between mammals and birds was fixed at 310 million years "in some 500 or more publications since 1990," when the fossils allow anything from 330 to 288.²³  On those figures a node fixed at 310 that is really 330 makes every other date on its tree about 6 % too young.²¹  Simon Ho and Matthew Phillips showed in 2009 that treating such minimum ages as exact points makes dates "display illusory precision" and gives "a clear bias toward obtaining younger age estimates."²⁴

The number on the fossil is itself borrowed.  A fossil comes from a rock layer.  The layer gets a number in years from radioisotopic dating, either directly or by matching it to a dated layer elsewhere.  James Parham and 24 coauthors, setting out best practice for calibrations in 2012, noted that "Divergence dating analyses require numeric ages, but paleontologists do not routinely use or report numeric ages."  The step from fossil to number "frequently involves a daisy chain of correlations," and "for the vast majority of calibrations, this translation is not explained."²⁵

The curve you enter is not the curve the program uses.  Modern programs take each calibration as a probability curve.  To keep every ancestor older than its descendants, they trim those curves jointly, so the prior that is actually used differs from the one the user typed.  Rachel Warnock, Ziheng Yang and Philip Donoghue found in 2012 that changing only the shape of one calibration curve can make the mean dates and both ends of their intervals "differ by hundreds of millions of years."²⁶

Islands and continents.  A clock can also be set by a geological event, such as the age of an island or the opening of a seaway.  Robert Fleischer and colleagues used the potassium-argon ages of the Hawaiian islands, from Kauai at 5.1 million years to Hawaii at 0.43 million on the standard dating, to calibrate a rate of 0.016 sequence divergence per million years in the mitochondrial cytochrome b gene of honeycreepers.²⁷  They noted that an island's age gives only "a maximum age for the taxa inhabiting the island."²⁷  Michael Heads argued in 2005 that even that assumption fails, because island chains form above hot spots and subduction zones "where small, individually ephemeral islands are constantly being produced and disappearing, and a metapopulation can survive indefinitely."²⁸  A lineage can be older than the island it lives on.  The error of a geological calibration can therefore run in either direction.

Sample dates and pedigrees.  The third route needs no fossil and no geological event.  If sequences are collected at known dates, from a virus over a decade or from bones dated by other means, the change between early and late samples gives the rate directly.  In a pedigree the generations are counted.  These routes still assume that the samples are related as the tree says, but they do not assume any date for a deep ancestor.  Part 5 shows that they are the routes that pass tests where the answer is known.

The circularity charge, stated

Young-earth writers have made this the center of their case against clocks.  Jeffrey Tomkins and Jerry Bergman, in the Journal of Creation in 2015, wrote that clock methods "assume evolution and deep-time calibrations taken from paleontology," and that "divergence dates commonly disagree with paleontology despite being calibrated by it."²⁹  Their central objection:  "the overriding problem is the illogical circular reasoning surrounding the whole process.  Molecular genetic clocks are calibrated by fossils that are themselves calibrated by their sedimentary rock formations and the sedimentary rock formations are calibrated by the fossils."²⁹  Nathaniel Jeanson draws the same contrast for his own clock.  Evolutionary clocks "calibrate" genetic differences with dates "derived from the evolutionary interpretation of the fossil record," while his "was derived from empirical measurements of mutation rates."¹⁷  Robert Carter of Creation Ministries International writes that when clocks are calibrated on dated events, "the 'deep time' dates are baked into the cake."³⁰

The charge, tested

For fossil-calibrated deep dates, the charge is right, and the mainstream says so.  Kevin Langergraber and colleagues wrote of the standard human mutation rate, which was set from an assumed 7-million-year human-chimp split, that "this approach has an inherent circularity and is subject to possible error because it relies on the accuracy of the ages of fossils."⁴  Dos Reis and Yang's finding that nearly all the uncertainty in a genome-scale date comes from the fossil calibrations says the same thing in numbers.³  A clock calibrated on fossils redistributes the radiometric timescale across a tree.  It cannot test that timescale.  A fossil-calibrated clock date is therefore not independent evidence for deep time, and an article that cites one as if it were overstates its case.

The circle Tomkins and Bergman draw is wrong in its details.  The numbers on calibration fossils come from radioisotopic dates on rocks, as Parham's checklist requires, not from the fossils.²⁵  Fossils give the order of the layers and let layers be matched between regions.  That matching is older than radiometric dating, and young-earth geologists accept it.  Kurt Wise and Donna Richardson concluded in 2023 that "the global biostratigraphic column seems to be real."³¹  So the charge reduces to the dispute over radiometric dating, which has its own articles on this site, starting at Dating the Earth.  A clock neither adds to that evidence nor takes from it.

Taking the fossils out does not make the dates young.  The non-circular routes exist and have been run.  The human mutation rate counted in families is about half the old fossil-calibrated rate, so it roughly doubles the age of the human-chimp split rather than shrinking it;  that case is worked in The Human-Chimp Split and the Mutation Rate.  On the Y chromosome, a rate taken from pedigrees alone puts the root of the human tree at about 163,000 years, as worked in Y-Chromosome Adam and Noah.  Across 68 vertebrate species sequenced as parent-offspring trios, Lucie Bergeron and colleagues found that yearly rates from families track the long-term substitution rates read from genomes, though loosely:  "adjusted r2 = 0.23, P = 0.002" on ultraconserved regions and their flanks.⁵  In single species the fossil-free date can come out younger as well as older.  Peter Keightley and colleagues counted nine new mutations in families of Heliconius butterflies, 2.9 × 10⁻⁹ per site per generation, and dated the split of H. melpomene from H. erato to "approximately 6 Ma, which is somewhat more recent than that estimated from a fossil-calibrated phylogeny of 10–13 Ma."³²  Within about a factor of two either way, the counted rates and the fossil-calibrated rates tell the same story.  The circularity charge is a fair objection to fossil-calibrated dates.  It is not a route to a young timescale, because the fossil-free rates for nuclear DNA give dates as old or older.  Part 7 shows how young-earth geneticists meet this:  they hold that most differences between created kinds were never produced by mutation at all.

Part 3: A clock that runs at different speeds

Even a single protein ticks unevenly.  If substitutions arrived like the clicks of a Geiger counter, the variance in their number across lineages would equal the mean.  The ratio of the two, the index of dispersion R, would be 1.  John Gillespie measured it in 1984 for five proteins across mammals and got 1.39 for hemoglobin alpha, 3.47 for hemoglobin beta, 1.87 for myoglobin, 3.62 for cytochrome c and 2.85 for ribonuclease.³³  He concluded that Kimura's claim that the observed values argued for indices "close to 1 and, hence, neutrality, is vacuous."³³  Trevor Bedford, Ilan Wapinski and Daniel Hartl repeated the test genome-wide in 2008.  The mean index was 2.10 across 3,081 yeast proteins, 4.19 across 7,174 fruit-fly proteins and 6.48 across 8,065 mammal proteins.⁶  They found "the extent of overdispersion shows a strong negative correlation with the effective population size."⁶  Mammals, with the smallest populations, have the least regular clock.  An index of 6.5 means a date from one gene carries about two and a half times the error a simple clock would assign it.²¹

Different genes can give wildly different dates for the same split.  The evolutionary geneticist Francisco Ayala set two enzymes against each other in 1997.  Calibrated in fruit flies and applied to the split between plants and animals, which he took to be about 1,100 million years old, glycerol-3-phosphate dehydrogenase gave 3,990 million years and superoxide dismutase gave 224 million.³⁴  The two estimates differ eighteenfold.  It "remains unsettled how often proteins evolve in such erratic fashion," he wrote.³⁴

Lineages differ too, and the differences follow life history.  Short-lived animals copy their germ lines more often per year, so a clock that ticks per generation runs faster per year in mice than in men.  Tomoko Ohta found the rodent-to-primate rate ratio "about 3 for synonymous substitutions but only 1.6 for nonsynonymous substitutions."³⁵  Sudhir Kumar and Sankar Subramanian, on 5,669 mammal genes, found instead that the rate is "approximately constant per year and largely similar among genes."³⁶  The family studies settle part of the dispute.  Across 68 vertebrate species the per-generation rate varies about fortyfold, and "generation time, age at maturity and species-level fecundity are the key life-history traits" behind the variation.⁵  Within primates, Priya Moorjani and colleagues found substitution rates "up to 64% higher" on the branch to New World monkeys than on the branch to apes.  Only one class of change, transitions at CpG sites, which arise by chemical decay rather than copying errors, behaves close to a clock.³⁷  In birds, Jacob Berv and Daniel Field found substitution rates "varying up to 20-fold" across a genome-scale dataset, "strongly correlated with body size and metabolic rate."³⁸  In flowering plants, Stephen Smith and Michael Donoghue found rates "consistently low in trees and shrubs" and much higher, and more variable, in herbs.³⁹

The apes show how much this matters.  Applied to apes, the human rate counted in families puts the split from orangutans at 34 to 46 million years, against fossils of the orangutan line under 18 million on the standard dating.  The usual repair is a slowdown in the human yearly rate as generations lengthened.  That case is worked in The Human-Chimp Split and the Mutation Rate.

Not every change is free to accumulate.  The clock counts synonymous changes, which alter DNA without altering protein, as if they were all neutral.  About a fifth to two fifths of synonymous sites in apes are under some constraint, which lowers the rate at which they change.  A family-counted rate applied to synonymous divergence therefore makes dates too young, by 21 to 37 % on the estimates worked in Synonymous Mutations and Codon Usage.  A fossil-calibrated clock cancels any constant bias of this kind, because the same constraint is in the calibration.

What relaxed clocks do with this.  The answer of the field is the relaxed clock of Part 1, which lets each branch have its own rate.  That handles variation only as well as the model of variation is right.  Dos Reis and colleagues showed that "at this level of divergence the molecular clock is significantly violated," and that the choice of relaxed-clock model moved animal dates substantially.²  The irregularity is not a scandal the field hides.  It is the main subject of the field's own methods literature.

The young-earth reading

Young-earth writers make this point too, and they are right about the facts.  Tomkins and Bergman list "different genes/sequences give widely different evolutionary rates" and "different taxa exhibit different rates for homologous sequences" among the problems that "plague" the method.²⁹  Carter asks of human mitochondrial DNA, "Should we even expect a mutation 'clock' to exist?"³⁰  The mainstream sources quoted above say the same, with numbers.

What rate variation cannot do is move dates by orders of magnitude.  The largest measured spreads are about fortyfold per generation across vertebrates and twentyfold across birds.  For scale only, since no young-earth model claims that mutation produced the differences between kinds:  compressing a node the mainstream dates to 66 million years into 6,000 years would need rates about 11,000 times faster along every branch below it.²¹  Irregularity widens the error bars on clock dates.  It does not turn millions of years into thousands.

Part 4: Rates that depend on the timescale

The pattern.  A rate measured over a short span comes out faster than a rate measured over a long one, in the same kind of DNA.  Simon Ho and colleagues documented it in 2005 for bird and primate mitochondrial DNA, finding "a measurable transition between the high, short-term (<1–2 Myr) mutation rate and the low, long-term substitution rate," a curve they proposed "may be used for correcting molecular date estimates."⁴⁰  Ten years later Agnieszka Molak and Ho compiled 239 published rate estimates from animal mitochondria and "found evidence of time-dependent rates in both coding and non-coding mitochondrial markers, in every group of animals that we studied."⁴¹  In viruses, where every rate comes from dated samples and no fossil is involved, Sebastián Duchêne, Edward Holmes and Ho found rates fall with the timescale of measurement in both DNA and RNA viruses, with a mean log-log slope of −0.38 in DNA viruses.  One papillomavirus rate set by assuming the virus split along with its hosts was "four orders of magnitude lower" than another, 4.58 × 10⁻⁵ per site per year, measured from dated human samples.⁴²

Outside humans, the pedigree and the phylogeny sometimes agree.  The best single test is in Adélie penguins.  Their bones lie under living and abandoned colonies in Antarctica, so Craig Millar and colleagues could count mutations in 508 living families and measure change in the same 344 letters of mitochondrial DNA from radiocarbon-dated bones.  Counting every new heteroplasmy, a mutation present in only some of a bird's mitochondria, gave 54.9 changes per site per million years.  Modelling how many such variants survive the bottleneck between mother and egg cut that to 0.55.  The ancient bones gave 0.86, and "they cannot be distinguished statistically (p = 0.13)."⁷  On their model, counting each heteroplasmy as a mutation "over-estimates m by a factor of 5."⁷  In this species the fast pedigree rate was an accounting choice, and when it was corrected the gap closed.

In fruit flies the gap is real, and its cause is visible.  Cathy Haag-Liautard and colleagues measured the mitochondrial mutation rate in laboratory lines kept for hundreds of generations:  28 point mutations, 6.2 × 10⁻⁸ per site per generation.⁸  That is more than ten times their nuclear rate, yet the two genomes show the same synonymous divergence, about 0.12, between Drosophila melanogaster and D. simulans.⁸  Of the 24 mutations in genes, "all but one were nonsynonymous."⁸  In the authors' words, the high rate "largely comes from mutations at nonsynonymous major-strand G sites; these are subject to strong purifying selection in nature, and this contribute little to between-species divergence."⁸  The mutations that make the short-term rate high are the ones that do not last.

Why it happens, by the mainstream's account.  Four causes are proposed, and the field has not measured their shares.

1. Purifying selection. Slightly harmful mutations are counted in families but removed over longer spans. Sankar Subramanian and colleagues, using penguin mitochondrial genomes up to 44,000 years old by radiocarbon, found the rate at constrained sites "varies more than twofold with time," while "the rate of evolution at synonymous sites remains the same."⁴³
2. Ancestral variation. Two lineages that split recently already differed by the variation in their common ancestor. Grant Peterson and Joanna Masel showed this inflates an apparent rate by a factor of 1 + Ne/t for mitochondrial DNA, where Ne is the effective population size and t the time in generations. For human mitochondria they estimated Ne of about 300. The apparent rate falls to twice the true rate "in approximately Ne generations," which at 28-year generations "corresponds to 8,400 years."⁴⁴ The same formula gives a factor of 2.4 at 6,000 years and 1.05 at 168,000.²¹
3. Saturation. Over long spans some sites change twice or more, and imperfect corrections undercount deep change.
4. Calibration error. If deep calibrations were too old, deep rates would come out too slow. This cause would make the whole pattern an artefact of the deep timescale. It is the one a young-earth reader will favor. It is weakened, but not excluded, by the fact that the same decline appears in viruses dated only by their sampling dates.

The pattern itself is disputed in print.  Brent Emerson and Michael Hickerson argued in 2015 that their assessment "reveals a lack of theoretical support and empirical evidence for hypothesized magnitudes of time-dependent rates," with "much of the apparent rate changes coming from artefacts and biases inherent in the methods of rate estimation."⁴⁵  Ho and colleagues replied the same year, disputing the claim "that the time-dependent pattern in rate estimates can be entirely explained as an artefact of data analysis."⁴⁶  Both sides of that dispute work on the standard timescale.  Neither is arguing about whether the true elapsed time is thousands of years.

The young-earth reading

Carter states the creationist reading most fully.  "Genealogical mutation rates are several orders of magnitude faster than phylogenetic estimates," he writes, and the slow rates are "clearly influenced by evolutionary assumptions."⁹  He grants a role for selection:  "due to selection, the long-term, phylogenetic mutation rate will be slightly less than the short-term, genealogical mutation rate."⁹  But he argues that "if most mutations are selectively neutral, it is impossible to create a significant separation between the short-term mutation rate and the long-term accumulation of mutations,"⁹ and that "There is a time-dependency to the data, so clearly one cannot assume a constant mutation or fixation rate through all human history."⁹  The creationist biologist Todd Wood took a different route, reading ancient mitochondrial DNA from horses, cats and dogs as evidence of "an accelerated mutation rate near the time of the Flood."⁴⁷

"Several orders of magnitude" does not survive the numbers.  On the nuclear genome, the human rate counted in families is about 0.43 times the old fossil-calibrated rate, slower, not faster.  On the Y chromosome the pedigree rate is 1.15 times the rate from ancient genomes.  On mitochondrial DNA the ratio depends entirely on what is counted.  Changes that reach fixation in a child run 0.5 to 0.9 times the phylogenetic rate.  All changes seen across the Icelandic pedigrees of deCODE genetics run 8.5 times.  New heteroplasmies above 1 % in blood run 38 times, and those found in single egg cells 90 times.⁴⁸  The largest of these is under two orders of magnitude, and it counts the very variants that the penguin study showed are mostly lost.

Carter's neutral premise is right, and it does not apply to what is being counted.  For strictly neutral mutations, Kimura's result holds and the short-term and long-term rates are the same.  But the fast mitochondrial counts are dominated by variants that never reach fixation, in penguins, or by protein-changing mutations that selection removes, in fruit flies.  Those are not neutral substitutions in waiting.  Where the counted changes are mostly neutral, as on the nuclear genome and the Y chromosome, the family rate and the long-term rate already agree, or the family rate is slower.

The synonymous-site test cuts against the young-earth reading.  If the long-term rate looked slow only because the deep timescale was wrong, every kind of site would show the same inflation in recent samples.  In the penguins, the constrained sites slowed with time and the synonymous sites did not.⁴³  That is what selection predicts and what a timescale error does not.

What remains open.  The share of each cause is not measured.  Peterson and Masel's factor and Millar's heteroplasmy correction are models, and the penguin result is one species and 344 letters.  Calibration error cannot be excluded for the deepest nodes, which is the point of Part 2.  But the evidence that the fast rates are fast because of what they count, rather than because the world is young, comes from tests that need no deep calibration.

Part 5: Tests where the answer is known

A clock can be graded only by dating something whose age is known another way.  The tests below are sorted by depth, because depth decides how they come out.

Days to decades, dated by sampling:  the clock hits.  Jonathan Pekar and colleagues applied a strict clock to 583 SARS-CoV-2 genomes and put their common ancestor "within a 34-day window with a mean of 9 December 2019 (95% HPD: 17 November to 20 December)."¹⁰  The first cases were recognized in December 2019.  For HIV, Michael Worobey and colleagues recovered a viral sequence from a 1960 lymph-node biopsy in Kinshasa and asked the clock when it had been sampled, without telling it.  It answered 1959, "corroborating the authenticity of the DRC60 sequences and the existence of clock-like signal in our data set."¹¹  In a Libyan hospital where foreign nurses were accused of infecting children, Tulio de Oliveira and colleagues showed the HIV and hepatitis C strains "were already circulating and prevalent in this hospital and its environs before the arrival in March 1998 of the foreign medical staff."⁴⁹  In Richard Lenski's long-term experiment with E. coli, where the generations are counted, Olivier Tenaillon and colleagues found that in populations with normal mutation rates "synonymous mutations accumulated at an approximately constant rate" for 50,000 generations.⁵⁰  The same paper shows the caution:  populations that evolved a high mutation rate later lowered it, so rates can change within a lineage.⁵⁰

Centuries, dated by history:  the short-span rate makes old lineages look young.  These are tests a young-earth reader can accept, because the dates come from documents and from radiocarbon inside the historical period.

1. Smallpox. Ana Duggan and colleagues read a variola genome from a child mummy in Vilnius, radiocarbon-dated to "1643–1665 AD."⁵¹ The clock, asked when the sample lived, gave "mean sampling time estimates of 1691 and 1665."⁵¹ That is a near hit.
2. Hepatitis B. From 76 modern genomes, Yang Zhou and Edward Holmes estimated a rate of "7.72 · 10–4 subs/site/year" and a common ancestor of all sampled human strains "229 years (upper 95% HPD = 580 years)" ago.¹² Then hepatitis B genomes were read from skeletons. One came from a medieval cemetery at Petersberg in Germany, dated "1020–1116 cal AD," and showed "only low divergence to genotype D HBV genomes," one of the modern lineages.¹³ If, as its low divergence indicates, that genome belongs inside genotype D, then a modern lineage already existed about 900 years ago, and the common ancestor of all modern strains cannot be 229 years old, or even 580. Using ancient genomes up to about 4,500 years old as calibrations, Barbara Mühlemann and colleagues projected the root of the tree to "between 8.6 and 20.9 thousand years ago" with a rate of "8.04 × 10 −6 –1.51 × 10 −5" per site per year,⁵² about 50 to 100 times slower than the rate from modern samples.
3. Measles. Clock studies built on recent samples had placed the split of measles from the cattle virus rinderpest in the 9th century AD. Ariane Düx and colleagues added a genome from a 1912 lung specimen and a model allowing for selection, and moved the split to "528 BCE." They explained: "Rapid short-term substitution rates captured by tip calibration can often not be applied over long evolutionary timescales, because of the effects of long-term purifying selection and substitution saturation."⁵³
4. Tuberculosis, a live dispute. Iñaki Comas and colleagues dated the tuberculosis complex by tying its tree to dates in human prehistory, such as the spread of modern humans out of Africa, which put its root tens of thousands of years back.⁵⁴ Kirsten Bos and colleagues, calibrating instead on three Peruvian skeletons about 1,000 years old, found a rate "more than tenfold faster" and a common ancestor "less than 6,000 years ago."⁵⁵ Here the tip-dated clock gives the young date, and the mainstream has not settled which is right. Bos's rate comes from a 1,000-year window, the kind of short span the other three cases show running fast.
5. Correcting for the pattern. Mahan Ghafari and colleagues built a model in which the fastest-changing sites saturate first, which reproduces the time-dependent pattern of Part 4. Applied to hepatitis C, it moved the common ancestor from "only 940" years to "423,000" years. For the coronaviruses related to SARS, it moved the root to "21,000" years, "nearly 30 times older than previous estimates."⁵⁶

Thousands of years, dated by radiocarbon:  mostly right.  Beth Shapiro and colleagues tested whether a clock can recover the age of a sample left out of the analysis.  On simulated data the true age fell inside the 95 % interval 95 % of the time, and on dengue virus 93 % of the time.  On 125 radiocarbon-dated bison, the true age fell outside the interval for 18 or 19 of them, "approximately 15% of the time."⁵⁷  The authors blamed the strict clock, and relaxing it recovered seven of the failures.  A young-earth reader will reject the bison ages, which run to 55,000 radiocarbon years.  That dispute belongs to Radiocarbon.

Millions of years, dated by geology:  the clock cannot test the event it is calibrated on.  The Isthmus of Panama split the Caribbean from the Pacific, and dozens of sea animals have sister species on either side.  Nancy Knowlton and Lee Weigt set their snapping-shrimp rate "with the assumption that the most similar pair diverged at approximately 3 Ma," and the other pairs then dated from 3 to about 18 million years.⁵⁸  So the clock could not check the closure date it was set on.  Haris Lessios found that of 115 such pairs, "34 are likely to have been separated at the final stages of Isthmus completion, 73 split earlier and 8 maintained post-closure genetic contact."⁵⁹  Aaron O'Dea and colleagues, with a fossil-calibrated rate, found even the youngest pairs at 3.21 to 4.1 million years, older than the closure they accept at 2.8.⁶⁰  On Bioko, an island cut off from Africa at the end of the last ice age, monkey SIV viruses differ from their mainland relatives.  Worobey and colleagues concluded that "SIV is ancient and at least 32,000 years old."⁶¹  Dated from recent samples alone, these viruses had come out as young as 1951, at rates up to 10⁻² changes per site per year.  The island split needs a rate of about 10⁻⁶, ten thousand times slower.⁴²

What the record shows

The clock passes where it is calibrated by the dates of the samples themselves and asked about the same span.  It fails, and fails toward youth, when a rate from a short span is stretched over a longer one, even across only a few centuries of written history.  The hepatitis B and measles results run directly against the young-earth reading of Part 4.  If the fast short-span rates were the true rates, ancient genomes would have confirmed the young dates.  They did the opposite.

Part 6: Clocks against the first fossils

For the great radiations, clock dates come out older than the oldest fossils.  Dos Reis, Donoghue and Yang summarize the history:  as sequence data grew in the late 1990s, "the clock estimates were much older than the dates suggested by the fossil record, sometimes twice as old."¹⁹  On the mainstream's own timescale, crown animals date to 833 to 650 million years against accepted animal fossils "younger than about 575 Ma."²¹⁴  Crown placental mammals date to 83.3 to 77.6 million years, down from 107 in 2003, against Protungulatum at 64.85.⁶²⁶³⁶⁴  The root of modern birds dates to about 100 million years against Asteriornis at 66.8 to 66.7,⁶⁵⁶⁶ and crown flowering plants to 217 (182 to 257) against Valanginian pollen at about 135.⁶⁷⁶⁸

The gaps have narrowed for placentals and birds, and the experts dispute their size in print.  Budd and Mann concluded in 2024 "that in neither case does the molecular clock analysis provide any evidence for an origin of the clade deeper than what is suggested by the fossil record."¹⁴  The clock's authors replied that the explosive models they favor "have a posterior probability of ∼0."¹⁵  Some of the agreement that has appeared was built in:  one leading bird study wrote the absence of Late Cretaceous crown birds into its root prior, which "compresses the age of the neornithine MRCA by 30 Ma."³⁸  The comparison group by group, how the gaps moved as methods changed, the four mainstream explanations tested, and the young-earth readings are in the child article Do Clock Dates Agree with the First Fossils?.

The young-earth reading

Tomkins and Bergman list as a standing failure that "divergence dates commonly disagree with paleontology despite being calibrated by it."²⁹  That is true, and the mainstream sources above say so.  For the great radiations the disagreement runs one way, with the clock the older.  It has known causes in how calibrations are drawn and how rates change, and has narrowed where calibrations improved.  It shows that deep clock dates are imprecise and biased old.  It does not bear on the age of the earth, because both the clock dates and the fossil dates sit on the same radiometric timescale.  A young-earth reader who rejects that timescale rejects both sets of dates at once, and the gap between them then measures nothing.

Part 7: The young-earth clocks

Young-earth geneticists do not reject molecular clocks.  They use them, with mutation rates measured in families or laboratory lines, inside created kinds, and over about 6,000 years.  Between kinds they use no clock at all.

Jeanson's mitochondrial clock, in his words

Nathaniel Jeanson, a biologist with a Harvard doctorate who worked at the Institute for Creation Research and now at Answers in Genesis, built the clock in two long papers.  In 2013 he applied measured mitochondrial mutation rates for humans, a roundworm, a fruit fly and a water flea to the diversity within their genera, and wrote:  "Clearly, the young-earth model came close to the real value while the evolutionary model could not predict reality."⁶⁹  His data, he wrote, "challenge the millions-of-years timescale."⁶⁹  In 2015 he extended it across many animal families, with figures titled "Molecular Clock Confirmed" for the roundworm, the fly and the water flea.¹⁷  He named one assumption:  "the only assumption that the mitochondrial DNA clock made was that the rates of DNA change had been constant through time, a uniformitarian assumption that, in this instance, fit the YE timescale."¹⁷  For the nuclear genome he takes the opposite view.  In Replacing Darwin he summarizes:  "mtDNA differences are due to a constant rate of mutation over 6,000 years, whereas the majority — if not vast majority — of nuclear DNA differences are due to preexisting DNA differences that were placed in the ancestors of modern species 6,000 years ago."⁷⁰

He also stated how the clock could fail.  "By measuring the mutation rates in additional species within a family; making genetic diversity predictions from these rates under the assumption that rates have been constant; and then comparing the predictions to actual diversity, the utility of the constant rate assumption could be evaluated."¹⁷  He excluded ancient DNA:  "it seems unlikely that reliable sequences will ever be obtained from samples older than just a few years or decades, unless remarkable care is taken to preserve them."¹⁷

Jeanson's clock, tested

The fly clock works only on mutations that do not last.  Jeanson's fruit-fly rate comes from Haag-Liautard's laboratory lines.  Converted to whole genomes per year, it gave him a 95 % range of 0.0049 to 0.116 mutations, against 1,168 average differences among six Drosophila species.⁶⁹  Over 6,000 years that range predicts 59 to 1,390 differences, which brackets the observed 1,168.  But 23 of the 24 mutations those lines produced in genes changed a protein, and the measuring authors say such mutations "contribute little to between-species divergence."⁸  At most 5 of the 28 point mutations, the one synonymous change and the four outside genes, are of the kinds that can build up between species.  Using only those, the 6,000-year prediction falls to 11 to 248 differences, 4.7 to 110 times short of the 1,168 observed.¹⁶  The count of 5 is small and the correction is rough.  Its direction is not in doubt, because the excluded mutations are the ones selection removes.

"Measured" rates differ a hundredfold depending on what is counted.  Jeanson contrasts his clock, "derived from empirical measurements of mutation rates," with clocks calibrated on fossils.¹⁷  But the same measurement gives very different rates under different counting rules.  In the Adélie penguins, counting every heteroplasmy gave 54.9 changes per site per million years and modelling their survival gave 0.55, a factor of 100.⁷  In humans, the counted rates in Part 4 run from about 0.0045 to 0.74 changes per genome per generation, a factor of about 165.⁴⁸  Choosing among them is not a measurement.  It is the step at which the answer is decided.

A test close to his own has been run, and it favors the slow rate.  In 2015 Jeanson wrote that "no further mitochondrial DNA mutation rates have been measured in additional animal species."¹⁷  The penguin study, published in 2008, is one he did not use.  It compared a family-counted rate not with the diversity of a family, as his test would, but with the rate from dated bones on the same 344 letters.  Once heteroplasmy was modelled the two were equal within error, which leaves no excess of the kind a 6,000-year clock needs.⁷

The exclusion of ancient DNA removes the strongest test.  Ancient DNA is the first of Jeanson's own two proposed tests.¹⁷  It is no longer "a few years or decades."  Hepatitis B genomes have been read from a medieval skeleton dated by radiocarbon to 1020–1116 AD, inside the historical period that both sides accept.¹³  Part 5 shows what those genomes did to clock dates built from modern samples:  they made them older.

The nuclear genome of the same flies needs a second mechanism.  The same laboratory lines give a nuclear rate of 5.8 × 10⁻⁹ per site per generation, and D. melanogaster and D. simulans differ at about 12 % of nuclear synonymous sites.⁸  Mutation alone would need about 10 million generations on each branch.  In 6,000 years that is about 1,700 generations a year, a new generation every five hours.¹⁶  On his model the nuclear differences between two sibling fly species in one kind were therefore mostly created, while their mitochondrial differences arose by mutation.  That is consistent with what he wrote for humans and chimpanzees.  It means the mitochondrial clock is confirmed in a genome where its rate is inflated by short-lived mutations, while the genome with the lower, steadier rate must be set aside as created.

"Within an order of magnitude" is a weak standard.  Jeanson's 2013 predictions fell inside the observed values for three genera and "within an order of magnitude" for the fourth.⁶⁹  A clock that is right to a factor of ten cannot tell 600 years from 60,000.

Young-earth geneticists disagree about the clock

Carter, reviewing Jeanson's 2022 book Traced, wrote that "The molecular clock concept (even if only applied to the local tree topography) is a simplifying assumption in Jeanson's model.  This does not make it wrong, per se, but it does qualify his conclusions."¹⁸  Jeanson replied in 2025 that "patriarchal drive," the extra mutations of very old fathers after the Flood, means "the date for the Y chromosome ancestor of Native Americans given in Traced could easily be inflated."⁷¹  Wood argued from ancient DNA for "an accelerated mutation rate near the time of the Flood."⁴⁷  Jeanson's clock assumes a constant rate;  Wood and Carter hold that the rate was not constant.  These are not small differences, because a varying rate is exactly what makes a clock unusable for dating.

Between kinds: created, not dated

For differences between created kinds, young-earth geneticists do not use a clock.  Carter writes that dating the great-ape lineages amounts to "uncovering the differences God built into their initial genomes."⁹  Jeanson writes of the human-chimp differences that "A creationist would respond by saying that these differences were created."⁷⁰  On that position the deep clock dates of Part 6 are neither right nor wrong.  They are readings of designed differences as if they were mutations.  The test of that position is not a clock.  It is whether the differences between kinds look like mutations, in their spectrum, their sex-chromosome pattern and their placement, which is worked for humans and chimpanzees in The Human-Chimp Split and the Mutation Rate.

Objections and responses

1. "Clocks are calibrated on fossil dates, so they assume the deep time they are cited to prove."  For fossil-calibrated deep dates this is correct, and the mainstream says so, as Part 2 shows.  Such a date is not independent evidence for an old earth.  But the clock can be calibrated without fossils, on families, sampled viruses and dated samples.  Those routes give nuclear-DNA dates as old as the fossil route or older, and Y-chromosome dates of the same size.  Removing the fossils removes the circularity, not the long timescale.

2. "Family rates are many times faster than fossil rates.  Use them, and mitochondrial Eve lived 6,000 years ago."  The large gap exists only for mitochondrial DNA, and only when heteroplasmies or other variants that rarely last are counted, as Part 4 shows.  In penguins, the one species where both rates were measured on the same stretch of DNA, the gap closed once the counting was modelled.  On the nuclear genome the family rate is slower than the fossil rate.  The human mitochondrial case is argued in full in Mitochondrial Eve and the Young-Earth Timeline.

3. "Clock dates disagree with the fossils by tens of millions of years, so the method fails."  They do disagree, by amounts the field measures and argues over in print, and for the great radiations the clock date is the older.  It is not always so.  A 2021 bird study dated the split of ducks from fowl to 62.5 million years, about 4 million years younger than Asteriornis, which its describers place "close to the last common ancestor" of the two groups.⁶⁶⁷²  The disagreement shows that deep clock dates are imprecise and lean old.  It cannot show that the earth is young, because both the clock dates and the fossil dates sit on the same radiometric timescale.

4. "Two enzymes date the same split to 3,990 and 224 million years.  Clocks are worthless."  Ayala's comparison shows that single genes can mislead, which Zuckerkandl and Pauling warned in 1965.¹  Modern dates average hundreds of genes, which removes that kind of error but not the calibration error, so the dos Reis and Yang floor of about ±16 % remains.  "Worthless" overstates it.  "Imprecise at depth" is fair.

5. "Viral clocks hit the right month for SARS-CoV-2.  That proves clocks work."  It proves they work inside the span the samples cover.  The same viral literature shows rates measured over years fail when stretched over centuries, by one to three orders of magnitude, in the case of hepatitis B against medieval genomes.  Success at shallow depth says little about deep nodes, which rest on calibrations.

6. "The clock beat the fossils on humans and apes, so it is an independent check."  In 1967 Vincent Sarich and Allan Wilson put the human-African ape split at 5 million years, when fossils such as Ramapithecus were taken to show over 14.  "The controversy was settled once it was recognised that the fossils are more closely related to the orang-utan than to the African apes."¹⁹  The clock was right about which fossils belonged where.  Its date still came from an assumed 30-million-year split between apes and New World monkeys,¹⁹ so it checked the fossils' placement, not the timescale.

7. "Rates vary, so they may have been far faster in the past, and a young earth fits."  Measured rates vary by up to about twentyfold across birds and fortyfold per generation across vertebrates.  Fitting a node the mainstream dates to 66 million years into 6,000 years would need rates about 11,000 times faster on every branch.  No young-earth model proposes that for differences between kinds;  they call those differences created.  Inside kinds, young-earth writers disagree with each other about whether rates were constant, as Part 7 shows.

8. "Ancient DNA is degraded and cannot be trusted."  Jeanson excludes it.¹⁷  Other young-earth scientists do not:  Wood used it to argue for faster mutation near the Flood.⁴⁷  The medieval hepatitis B genome at Petersberg came with the chemical damage typical of old DNA and with viral protein recovered from the same teeth.¹³  A test that excludes the one kind of evidence that could check it is not a test.

9. "Time-dependent rates are an artefact, as Emerson and Hickerson argue."  If they are right, short-span rates should apply at depth, and the young-earth reading gains nothing, because the hepatitis B and measles results rest on dated genomes, not on a model of time-dependency.  If Ho and colleagues are right, the long-term rate is lower than the short-term rate for reasons that make deep dates older.  Neither side of that dispute supports a young timescale.

10. "Fossil-only models also assume deep time."  They do.  They are cited here only to show that, on the mainstream's own terms, a group should be older than its oldest fossil, so a gap between clock and first fossil is expected.  They do not bear on the age of the earth either.

Issues and unanswered questions

1. The causes of time-dependent rates have not been separated. Purifying selection, ancestral variation, saturation and calibration error are each supported, and nobody has measured their shares. A study with dated samples across several depths in one lineage, partitioned by site class, could do it.
2. The Bioko case is not reconciled. The island's separation is put at about 10,000 years, yet Worobey's abstract gives SIV "at least 32,000 years."⁶¹ The main paper was not obtained, so the step between the two figures is not checked here.
3. Tuberculosis is unsettled inside the mainstream. A tip-dated clock from 1,000-year-old skeletons gives under 6,000 years; calibration on human prehistory gives tens of thousands.⁵⁴⁵⁵ This is the one known case where a short-span clock gives a young date that the field has not overturned. Whether it is right, or is the time-dependent pattern again, is open.
4. Jeanson's fly correction rests on 5 mutations of 28. A direct count of synonymous and protein-changing differences among the six Drosophila mitochondrial genomes he used would replace that estimate with a measurement. It has not been done here.
5. No young-earth source has run Jeanson's own test on the rates measured since 2008. The penguin pedigree and later direct mitochondrial rates in other animals are available. A young-earth analysis of them would be the natural next step for that side.
6. Fossil-only origin estimates were not found for animals or birds. They exist for placental mammals and flowering plants, and those two disagree with each other's critics.
7. Some primary papers were not obtained: Thorne, Kishino and Painter 1998 on autocorrelated clocks; Gillespie 1989 on lineage effects; Sarich and Wilson 1967, read here only through dos Reis and colleagues; the full texts of Wray and colleagues 1996, Cooper and Penny 1997, Mühlemann and colleagues 2020 on Viking-age smallpox, and Kocher and colleagues 2021 on hepatitis B.
8. Gillespie's "two of five" is ambiguous. His own test applied to his own printed values makes ribonuclease significant at the 5 % level too. Whether he used a stricter threshold or unrounded values is not stated.
9. Direct comparisons of family rates with fossil-calibrated dates outside humans are few. Bergeron's correlation across 68 species is real but loose, adjusted r² of 0.23. Butterflies and apes give dates within about a factor of two in opposite directions. More species would show whether that factor is typical.

Verdict

As a method, the molecular clock is a ratio with an imported time.  The sequences measure rate times time.  The time comes from a calibration, and for deep nodes the calibration is a fossil dated by radiometric methods, which carries nearly all of the uncertainty.  Clocks are irregular between genes and lineages, by factors of two to forty.  Their rates depend on the span over which they are measured, and their dates for the great radiations run older than the first fossils by tens of millions of years on the mainstream's own timescale.  Where the clock can be graded against a known answer, it passes when calibrated by the dates of its own samples and asked about the same span:  SARS-CoV-2, HIV, the Benghazi outbreak, the bacterial experiment.  It fails, toward youth, when a short-span rate is stretched further, even across a few centuries of written history, as hepatitis B and measles show.  A deep clock date is a reasonable estimate on the mainstream timescale, with an honest uncertainty of at least ±16 % and a known lean toward the old side.  It is not independent evidence for that timescale.

On the young-earth side, the circularity charge is correct and does not help.  Fossil-calibrated clocks do borrow deep time.  But the routes that borrow nothing, families and dated samples, give nuclear and Y-chromosome dates as long as the fossil route or longer.  The claimed hundredfold-and-more excess of family rates holds only for mitochondrial DNA, and only when variants that rarely last are counted.  Jeanson's mitochondrial clock fits 6,000 years in fruit flies only with a rate dominated by mutations that do not persist.  It excludes the ancient DNA that tests it, and needs the same flies' nuclear differences to be created.  Young-earth geneticists themselves disagree on whether the clock was constant.  Their stronger position is the one Carter and Jeanson already take:  that differences between created kinds were made, not mutated.  That position does not rest on any clock, and it has to be tested by whether those differences look like mutations.

Sources