← all lessons

Scaffold S14 — Rates of evolution across intervals

Five rounds. Each round picks a dataset and an interval length. The scaffold then computes the magnitude of evolutionary change over every sub-window of that length, and you predict the median |rate|. Rounds 1–3 use the same LTEE fitness series at three interval lengths; rounds 4–5 use Grant finch beak depth and a paleontological Hyopsodus body-size series. Watch the rates drop as the interval grows.

Locked — answer the pretest above first.

Running tally — median |rate| vs. interval length

What you just did has a name

Rounds 1–3 showed the same LTEE population, the same 50,000 generations of fitness data. You only changed the length of the sliding window over which you computed a rate. The median rate dropped as the window grew. Nothing changed about the biology.

This is Gingerich's 1983 result: rates of evolution decline with the interval over which they are measured. Part of the decline is a real biological fact (evolution proceeds in bursts separated by stasis, so longer windows average burst + stasis together). Part of it is a statistical artifact (measurement noise and short-term reversals average out over longer intervals). Without a null model you cannot separate the two.

The scale of the decline is enormous. LTEE at 500 generations gives rates orders of magnitude larger than Grant finches at 1 year, which are in turn orders of magnitude larger than Hyopsodus over 500,000 years. Published rate compilations (Gingerich 2019, Hendry & Kinnison 1999) show the same pattern across hundreds of studies. A single "rate of evolution" has no meaning without an interval. Always report the interval.