Distance and Likelihood Trees, and What Branch Support Means · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

Reading trees without fooling yourself

Science · Genetics & Evolution · ages 20-21
Name ______________________   Date ____________

  1. How do distance methods and likelihood methods differ?

    • Distance minimizes total steps while likelihood counts differences
    • Distance compresses sequences into pairwise gaps, while likelihood picks the most probable tree
    • Distance reads support values while likelihood draws branch lengths

  2. Why does counting raw differences underestimate true change?

    • Some sites were hit twice and some changes reversed, hiding change
    • Models always add extra change that never happened
    • Resampling removes differences before anyone counts them

  3. What does a support value actually report?

    • The age of the branch in millions of years
    • How often resampling recovers that split
    • The chance the branch is true

  4. Why do good methods downweight rapidly evolving sites?

    • Fast sites are likely saturated, so their similarity often misleads
    • Fast sites never change, so they carry no information
    • Slow sites are too rare to matter for the tree

  5. A published tree shows 42 percent on the branch your claim needs. What should you do?

    • Rest the claim on it, since published trees are always safe
    • Redraw the tree by hand until the value looks higher
    • Treat the split as undecided and narrow the claim to strong nodes

  6. A node has poor support. Which fix targets the cause most directly?

    • Adding informative characters and species to break up long branches
    • Deleting the weak branch from the figure before publishing
    • Counting raw differences instead of using any model

  7. A branch with 95 percent support is 95 percent likely to be true.

    Circle one:   True   False

  8. Raw similarity groups two fast evolving species together, but a model based tree separates them. Why trust the model tree here?

    • Models always agree with raw similarity, so this cannot happen
    • Parsimony used more steps, which is always more accurate
    • Convergent and reversed changes create false similarity that only model correction discounts
LightMySky · lightmysky.comW1-mt_mEO9PUEdh6-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

Reading trees without fooling yourself W1-mt_mEO9PUEdh6-s1

  1. Distance compresses sequences into pairwise gaps, while likelihood picks the most probable tree · Distance shrinks data into gaps then clusters. Likelihood scores each tree by how probably it made the data.
  2. Some sites were hit twice and some changes reversed, hiding change · Double hits and reversals hide change from raw counts. Correction exists to reveal them.
  3. How often resampling recovers that split · Support is about repeatability of the result. It never certifies truth.
  4. Fast sites are likely saturated, so their similarity often misleads · Rapid sites get hit repeatedly until their signal saturates. Downweighting stops false similarity from voting.
  5. Treat the split as undecided and narrow the claim to strong nodes · Weak support cannot carry a bold conclusion. Honest science shrinks the claim to fit strong nodes.
  6. Adding informative characters and species to break up long branches · Weak nodes come from thin evidence, so more characters, denser sampling, or a better fitting model is the cure.
  7. False · Even high support reports repeatability, not truth. Strong nodes still depend on data and model behind them.
  8. Convergent and reversed changes create false similarity that only model correction discounts · Fast lineages converge by chance, faking closeness. Model correction discounts exactly that fakery.
Worksheet · LightMySky