An island can make evolution visible in forms that look almost theatrical: giant rodents, miniature elephants and deer far smaller than their mainland relatives. The usual picture is a slow transformation unfolding over deep time. A broad comparison of mammal measurements found a different tempo at the beginning of that story.
Virginie Millien, then at McGill University’s Redpath Museum, reported that island mammals changed morphologically faster than mainland populations over shorter observation windows. The difference was most robust from 21 years to about 20,000 years, with island rates reaching as much as 3.1 times mainland rates. At longer timescales, the gap narrowed and eventually ceased to be statistically significant.
The result came from a peer-reviewed 2006 PLOS Biology analysis, not a single experiment in which scientists watched 88 species transform. Understanding what it established requires looking closely at what was measured, how the rates were calculated and what critics later challenged.
Eight hundred and twenty-six rates across mammal evolution
Millien surveyed 60 published studies containing measurements from living populations and the fossil record. From them, she calculated 826 evolutionary rates for 170 populations representing 88 species or evolutionary lineages in 14 mammalian orders.
The traits were linear dimensions of skulls, teeth, skeletons and external anatomy. The analysis did not include body mass, ratios describing shape or categorical characteristics. Areas, volumes and masses scale differently from lengths, so mixing them would have made the rates hard to compare.
The time intervals extended from 21 years to 12 million years. That range is central to the result. Twenty-one years was the shortest interval represented in the dataset. It does not mean all 88 species changed over 21 years, nor that every island population can be expected to produce a conspicuous difference within two decades.
The island advantage was strongest early
Evolutionary rates decreased as the interval over which they were calculated grew longer for both island and mainland populations. The island regression line nevertheless sat above the mainland line across much of the range. The separation was clearest at short and intermediate intervals.
Millien judged the faster-island conclusion most robust from the 21-year minimum to about 20,000 years. The difference between the two lines became statistically nonsignificant beyond about 45,000 years. Across the full analysis, island morphological rates were up to 3.1 times those calculated for mainland populations.
This pattern fits a fast-start model. A population newly isolated on an island may move quickly away from its ancestral form, then change more slowly after approaching a locally useful body plan. The study illustrated that idea with a hypothetical trajectory, though a later correction reversed two line labels in that figure’s caption. The correction did not change the dataset or statistical result.
Why an island changes the evolutionary equation
On many islands, large mammals tend to become smaller and small mammals tend to become larger. This tendency is often called the island rule. Reduced predation can make a tiny body less essential for escape, while fewer competing species can open ecological roles unavailable on the mainland. Limited food and space can push large animals in the opposite direction.
Those are broad tendencies, not a mechanical formula. Island area, distance from a mainland, climate, productivity, predators, competitors and the founding population can all alter the outcome. A 2021 global meta-analysis of terrestrial vertebrates found widespread island-rule effects in mammals, birds and reptiles, but also showed that their magnitude varied with climate, island size and isolation.
Millien’s later work added another piece: a 2011 analysis of 72 mammal populations on 44 islands found faster and larger changes on smaller islands. That follow-up covered 21 species and intervals from 21 to 47,000 years, with rodents again forming most of the sample.
How a “darwin” measures change
The 2006 paper expressed rates in units called darwins. For a given linear structure, the calculation takes the natural logarithm of its later measurement minus the logarithm of its earlier measurement, then divides by elapsed time in millions of years. The result describes proportional morphological change per unit time.
That method allows a tooth length and a limb-bone length to be compared as proportional changes. It also creates a difficulty: a rate measured over millions of years will usually look slower than one measured over decades, because a long interval averages rapid episodes together with stasis and possible reversals.
Millien addressed temporal scaling by regressing log rate against log time separately for island and mainland groups, then comparing the lines at matched time intervals. The paper also tested whether the heavy representation of rodents drove the result. Nearly 60 percent of species in the full dataset were rodents, and 74 percent of island taxa were rodents. Repeated subsampling with equal numbers of rodent and non-rodent records preserved the main short-interval island difference.
A published challenge exposed the main uncertainty
In 2007, Juan Antonio Pérez-Claros and Juan Carlos Aledo argued in a PLOS Biology comment that the distribution of observation windows could create a misleading pattern. The dataset had relatively few island points at the longest intervals and few mainland points at the shortest ones. When the critics restricted the comparison to the overlapping range of 2,400 to 21,000 years, they did not find a significant island-mainland difference.
Millien answered that rates in darwins do have known time-scaling problems, but argued that the original conclusion rested on the difference in elevation between the island and mainland regression lines, not merely on their downward slopes. In her published response, she said the critics’ restriction removed more than half the observations and almost all of the original range on a natural time scale.
The exchange does not reduce to one side discovering a simple error. It identifies the central trade-off in the evidence. The complete dataset is the broadest test, but its time intervals are unbalanced. A tightly overlapping subset is easier to compare, but much smaller and less able to test how the pattern changes with time. Later evidence has continued to support rapid and context-dependent island evolution, while the exact size of the rate advantage remains sensitive to data and method.
Rapid morphology is not automatically genetic adaptation
The paper compiled measurements rather than genomes. It therefore documented morphological change rates, not the particular genes, mutations or selective pressures responsible for every change. Environmental effects on growth, sampling differences and uncertain dates in the fossil record can complicate the inference.
Millien noted that it was not possible to calculate error ranges for all time estimates because the 60 source studies used different dating methods. Some mainland fossil datasets also combined specimens across localities, adding geographic variation to change through time. The author argued that there was no obvious reason for those issues to bias one group consistently upward, but they remain limitations of a literature synthesis.
Contemporary work can sometimes resolve those mechanisms more directly. A previous ScienceBlog report described how Chicago rodents developed smaller teeth and ears across roughly a century of urbanisation, combining museum specimens with genomic evidence. The island analysis covered a much wider evolutionary range but usually lacked that level of mechanism for individual cases.
Not every island transformation is fast
Some of the most dramatic island forms still took hundreds of thousands of years to emerge. The Sicilian dwarf elephant Elephas falconeri fell below one metre in height and about 100 kilograms, less than one percent of the mass of its mainland ancestor, over an estimated 200,000 to 400,000 years.
Large final differences do not necessarily imply high average rates. A modest change completed in decades can have a higher calculated rate than an enormous transformation spread across geological time. This is why the study compared tempo rather than simply ranking the strangest island animals.
The broader island story is also not limited to mammals. ScienceBlog’s account of the dodo and its flighted Nicobar pigeon relative describes how isolation, low predation and ecological opportunity can reshape a lineage. The direction and timescale differ, but the ecological reset is similar.
What the 21-year headline really means
The analysis supports a clear but bounded conclusion: mammalian morphology is capable of changing unusually quickly after island isolation, and the island-mainland rate difference is most evident over shorter timescales. It does not show that every trait changes, that every change is inherited, or that each of the 88 species completed a transformation within 21 years.
Nor did the study follow the same population from colonisation through a later slowdown. The declining island advantage was inferred by comparing many populations and lineages measured over different intervals. It is consistent with rapid adjustment followed by a return toward mainland rates, not a universal stopwatch for island evolution.
That distinction makes the result more useful, not less. Islands show that the bodies of mammals are not limited to imperceptible change over millions of years. Under a radically altered ecological regime, measurable shifts can occur within decades, while the deepest transformations still accumulate over thousands or hundreds of thousands of years.