Bayesian Phylogenetics: Priors, Sampling and Judging Convergence · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

What a Bayesian tree run really tells you

Science · Genetics & Evolution · ages 22-24
Name ______________________   Date ____________

  1. What does the tree prior mainly shape in a dated analysis?

    • The file format of the alignment
    • Node ages through expectations about speciation and extinction
    • The number of bootstrap replicates to run

  2. What does a posterior probability of 0.95 on a clade mean?

    • The clade appears in 95 percent of the posterior trees
    • The clade was resampled in 95 percent of bootstrap replicates
    • The clade covers 95 percent of the sequence length

  3. A posterior probability and a bootstrap proportion answer the same question about a clade.

    Circle one:   True   False

  4. Your chain shows a low effective sample size and a trace that drifts upward. What do you do?

    • Quote the posteriors with extra decimals
    • Lengthen the run and check independent runs before quoting anything
    • Delete the drifting part and publish the rest silently

  5. Your dates swing widely when you try reasonable alternative priors. What should you conclude?

    • The chain has converged and the dates are firm
    • The sequences are long enough to ignore priors
    • The priors are carrying the conclusion, not the data

  6. Two independent runs give very different posteriors for the same clade. What does that mean?

    • The runs disagree, so the chain has not converged
    • The runs prove the clade is absent from the tree
    • The runs show the prior was uniform

  7. You move from a strict clock to a relaxed clock prior. What changes in your model?

    • Rates may now vary freely across branches instead of staying shared
    • Node ages become fixed and immune to the tree prior
    • Bootstrap replicates replace the posterior distribution

  8. A colleague quotes 0.99 from one short chain with no convergence checks as proof of a clade. What is the flaw?

    • High posterior values never need any checks
    • Only bootstrap values above 0.99 count as proof
    • An unconverged sampler can report strong support that means nothing yet
LightMySky · lightmysky.comW1-mt_7jWhYh1n5X-s1

Answer key

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

What a Bayesian tree run really tells you W1-mt_7jWhYh1n5X-s1

  1. Node ages through expectations about speciation and extinction · The tree prior encodes speciation and extinction expectations, which matter most for node ages.
  2. The clade appears in 95 percent of the posterior trees · It is a share of your posterior distribution, computed under your data, model, and priors.
  3. False · One states belief given the data and model, and the other measures stability under resampling.
  4. Lengthen the run and check independent runs before quoting anything · A drifting trace with low effective size means the sampler has not converged yet.
  5. The priors are carrying the conclusion, not the data · A big swing under reasonable priors means the data are too thin to carry the dates.
  6. The runs disagree, so the chain has not converged · Trustworthy runs must agree on clade posteriors, so disagreement means no convergence.
  7. Rates may now vary freely across branches instead of staying shared · The clock prior controls how strictly rates are shared, from strict to fully relaxed.
  8. An unconverged sampler can report strong support that means nothing yet · Without convergence checks, a high number is sampler noise dressed as support.
Worksheet · LightMySky