Statistical Genomics Foundry
Paper
domain/phylogenetic-comparative-methods

Beaulieu & O’Meara 2016 — HiSSE (Hidden State Speciation and Extinction)

1. Citation

Beaulieu JM, O’Meara BC. 2016. “Detecting Hidden Diversification Shifts in Models of Trait-Dependent Speciation and Extinction.” Systematic Biology 65(4):583–601. DOI: 10.1093/sysbio/syw022. Published version: https://academic.oup.com/sysbio/article/65/4/583/1753616 (All Rights Reserved, © The Author(s) 2016, Oxford University Press). Preprint: bioRxiv 10.1101/016386 (4 versions, 2015-03-14 → 2015-12-09), license CC-BY-NC-ND, category Evolutionary Biology, later published as syw022. Access date: 2026-07-03.

2. Access note — READ THIS FIRST (dominant paywall boundary)

I could reliably read ONLY the abstract. Every full-text route was blocked or paywalled on the access date:

  • OUP article HTML and OUP article-pdf (syw022.pdf) → HTTP 403 / paywall (ARR).
  • bioRxiv full HTML and full PDF (all 4 versions) → Cloudflare interstitial (“Just a moment…”).
  • Jina reader proxy → 401 (auth required); Europe PMC → “Subscription required”; ResearchGate → “Request PDF”; Semantic Scholar confirms an OA GREEN PDF exists (CC-BY-NC-ND) but it is the same Cloudflare-blocked bioRxiv URL.
  • The full abstract WAS recoverable verbatim from Europe PMC core (matches the published abstract).

Consequence: the entire body — Methods, model equations, the observed/hidden state parameterization, the specific character-independent-diversification (CID) model variants, simulation Type-I/false-positive rates, any ancestral-state-reconstruction discussion, and the R software/function names — is NOT in the text I read. Statements below are drawn from the abstract only. A separate WebFetch “extraction” of the OUP page returned model-structure specifics (e.g., a “0A/0B/1A/1B” scheme, “CID-2”/“CID-4” labels); those were NOT verifiable from readable text and appear model-confabulated, so they are deliberately excluded from this note rather than passed through as paper facts. To recover the body, obtain the OUP PDF or an un-blocked preprint copy.

3. Thesis (1 sentence)

State-dependent speciation/extinction (SSE) models can mistake complex, unmeasured, co-distributed drivers of diversification for an effect of the observed trait; HiSSE fixes this by adding, to each observed state, a linked hidden state with its own diversification and transition dynamics, and can also be configured as a character-independent diversification model for use as a null.

4. Problem & context

SSE models are used to test whether evolving a novel trait raised speciation and/or lowered extinction rates. Per the abstract, it remains unclear whether such models identify real diversification drivers whether they are instead simply flagging more complex patterns that involve many unmeasured, co-distributed factors. HiSSE is framed as answering a leading criticism of BiSSE (Binary State Speciation and Extinction) methods: that unmeasured factors, not the scored character, may drive the rate differences BiSSE attributes to the trait. (The full statement of this false-association critique lives in the inaccessible body.)

5. Method / approach — (abstract-level only; body inaccessible)

  • HiSSE model. Extends SSE models by assuming that, linked to each observed state in the model are ‘hidden’ states whose diversification dynamics and transition rates may differ from those of the observed states on their own. So each observed character state carries an associated unobserved (hidden) state that can have its own speciation/extinction and transition rates. (Exact rate parameterization and the state/transition matrix are in the inaccessible body.)
  • Character-independent diversification (CID) use. The abstract states the model can serve as character-independent diversification models that permit a complex diversification process decoupled from the evolution of a character — i.e., rate heterogeneity driven by the hidden state alone, decoupled from the observed trait. (Specific CID variants/parameter counts are in the inaccessible body.)
  • Validation. Assessed via rigorous simulation tests and application to empirical data. (Simulation design, sample sizes, and quantitative error rates are in the inaccessible body.)
  • Software. Not named in the abstract. (The R package/function names are in the inaccessible body / package docs; per project convention any package-doc value should be tagged [tool-doc], not attributed to this paper.)

6. Key claims / findings (all abstract-sourced)

  • HiSSE specifically accounts for unmeasured factors that could affect the diversification rates estimated for the states of any observed trait, addressing a major BiSSE criticism.
  • Hidden states can have diversification and transition dynamics distinct from the observed states alone.
  • The model can be run as a character-independent diversification model (complex diversification independent of the character’s evolution).
  • Under simulation and on empirical data, HiSSE performs reasonably well and can at least pick up net diversification-rate differences between observed and hidden states, and detect when diversification-rate differences fail to track the observed states.
  • The authors position HiSSE as giving a more nuanced picture of trait-dependent diversification.
  • No quantitative thresholds, Type-I rates, or effect sizes appear in the abstract; those were not readable.

7. Load-bearing statements — own-words (CC BY-NC-ND is own-words-only per license-policy.yml). Summarized from the openly displayed abstract; body inaccessible.

  • On the false-association problem: it remains unclear whether trait-dependent diversification models are detecting genuine drivers of diversification or merely flagging more complex patterns driven by many unmeasured, co-distributed factors.
  • On the fix / BiSSE criticism: HiSSE extends the popular state-dependent speciation-and-extinction models to explicitly account for unmeasured factors that could affect the diversification rates attributed to an observed trait’s states — addressing a major criticism of BiSSE (Binary State Speciation and Extinction).
  • On the hidden state: the model assumes each observed state has associated “hidden” states with potentially distinct diversification dynamics and transition rates from the observed states alone.
  • On character-independent use / null framing: the framework can serve as a character-independent diversification model, allowing a complex diversification process that is independent of the focal character’s evolution.
  • On performance: HiSSE performs reasonably well — at minimum detecting net diversification-rate differences between observed and hidden states, and detecting when rate differences do not correlate with the observed states.

8. Stated scope, assumptions, limitations (as far as readable)

The abstract itself hedges performance (performing reasonably well, able at least to detect the differences) and says the authors discuss the outstanding problems with state-dependent speciation-and-extinction models in general — signaling the paper does not claim to fully resolve SSE-model problems. The specific stated caveats live in the inaccessible Discussion.

9. Failure modes / invalidity patterns (referee-relevant)

From the abstract, the named failure mode HiSSE targets: an observed-state SSE model (e.g., BiSSE) will attribute diversification-rate heterogeneity to the scored character even when the true driver is an unmeasured, co-distributed factor — a false association / spurious trait-dependence. HiSSE’s character-independent configuration is offered as the check (detect when diversification-rate differences do not track the observed states). Concrete detectors, diagnostics, model-comparison criteria (e.g., AIC deltas), and any quantified false-positive rates are in the inaccessible body — not captured here. [summarizer-inferred] note: the abstract implies the character-independent model is meant as a null for comparison against the trait-dependent model, but the explicit “use it as the null before claiming trait-dependence” recommendation wording is in the inaccessible body.

10. What the source does NOT address (within readable text)

  • Any statement about ancestral-state / marginal reconstruction accuracy — see §12; the abstract addresses diversification-rate inference only.
  • Quantitative Type-I error / false-positive rates and simulation conditions.
  • Model equations, exact parameterization, CID variant definitions.
  • Software package/function names.

11. Open questions / ambiguities (from limited access)

  • Exact structure and parameter count of HiSSE and of each character-independent (null) variant.
  • The precise recommended workflow and model-comparison criterion for declaring trait-dependent diversification.
  • Whether/where the body discusses reconstruction of ancestral states as opposed to rates.
  • Reported false-positive rates and the tree sizes / regimes where HiSSE is reliable vs. underpowered.

12. Guidance answers

Answering each guidance question against readable text (abstract only); body was inaccessible (see §2).

  1. Exact problem when observed-state SSE (BiSSE) attributes all rate heterogeneity to the focal character — its own framing of false-association risk. Partially answered. Abstract framing: models may be simply pointing to more complex patterns that involve many unmeasured, co-distributed factors rather than true drivers; HiSSE specifically accounts for unmeasured factors, addressing at least one major criticism of BiSSE. The detailed statement of the false-association mechanism is in the inaccessible body.

  2. What a hidden state is; HiSSE structure; why it’s a valid BiSSE null. Partially answered. A hidden state is an unobserved state linked to each observed state with diversification dynamics and transition rates that may differ from those of the observed states on their own. The precise structure (rate matrix, transitions) enabling it as a null is in the inaccessible body; the abstract only asserts the character-independent configuration exists.

  3. Explicit recommendation to use HiSSE / a character-independent (hidden-state-only) model as the null before claiming trait-dependent diversification — quote it. NOT fully answerable from readable text. The abstract describes the character-independent capability (how their model can serve as character-independent diversification models that permit a complex diversification process independent of the evolution of a character) but the explicit “use it as the null model” recommendation sentence is in the inaccessible body and could not be quoted. Flagged: the must-quote recommendation is behind the paywall.

  4. Does the paper claim anything about ANCESTRAL STATE / marginal reconstruction accuracy (vs. diversification inference only)? From readable text: NO — the abstract makes no claim about ancestral-state or marginal reconstruction accuracy; it addresses diversification-rate inference only (detecting net diversification differences and whether they correlate with observed states). Whether the body contains an ASR-accuracy claim is unknown (inaccessible). This is load-bearing for the skill: on the text I could verify, this citation supports the diversification-inference claim, NOT an “ancestral-state silently wrong” claim. The ASR angle is unconfirmed and must be checked against the body before relying on this cite for reconstruction accuracy.

  5. Simulation Type-I / false-positive rates and conditions. NOT in readable text. Abstract says only that validation used rigorous simulation tests and that HiSSE performs reasonably well; no numbers. Rates/conditions are in the inaccessible body.

  6. Software: R package and functions. NOT in the abstract. (Known separately to be the hisse R package, but that is package-doc / external knowledge — tag [tool-doc], do not attribute to this paper; not stated in the readable text.)