BGPT: hubbell regression

Explore by Goal

Quick Explanation Copied

Hubbell regression (what it does, in one picture)

“Hubbell regression” is a GLM-style link that connects environmental covariates to the Hubbell neutral theory diversity parameter (α) and then converts that α to observable diversity indices via an accumulation-curve–based mean function. In the GMTP arthropod application, actual evapotranspiration (AET) dominates (≈30% of variance explained in simple models), and human footprint has zone-dependent effects across tropical, dry, temperate/polar contexts.

Primary source:

Long Explanation

Hubbell regression — from environmental covariates to diversity via α-diversity

The core idea is to treat biodiversity as structured by a neutral-assembly diversity parameter (α), then regress that α against environmental covariates using a Hubbell accumulation-curve link.

1) Visual blueprint (pipeline)

2) What the model estimates (key numbers from the GMTP application)

In the GMTP arthropod study, diversity is modeled using DNA-barcoded COI BINs as species proxies (154,688 BINs across 2,415 samples), then used to fit the Hubbell-regression mean/accumulation structure with a shared accumulation-rate parameter σ.

Reported in the study extraction as M0 σ=0.569 and M6 σ=0.562.

3) Environmental drivers: AET (dominant) and Human Footprint (zone-dependent)

3A) AET effect on diversity

The study reports that per 10 mm/year increase in actual evapotranspiration (AET), diversity increases by about 12.4%, and that AET explains about ~30% of variance in simple models.

3B) Human Footprint (HFP) effects by climate zone

Reported human footprint effects differ by zone: decreases in tropical and dry zones, and increases in polar regions. Extracted coefficients (β) and p-values are zone-specific (e.g., tropical β≈−0.013 with p<0.0001; polar β≈+0.016 with p<0.038) and the study also reports missing-diversity shares that depend on zone.

Missing richness % values are taken from the extracted results for n=2000 by zone.

4) Model mechanics (what “regression” is doing)

4A) Accumulation-curve mean function and σ

The model uses a flexible link where σ controls accumulation growth (how expected richness scales with sampling effort n). The fitted σ then determines how covariate effects are expressed through the accumulation structure into the α parameter.

4B) Handling shared-bin dependence (Jaccard-adjusted uncertainty)

The study accounts for dependence between samples that share BINs by using Jaccard similarity and a Jaccard-adjusted sandwich estimator for robust standard errors, as part of the composite likelihood approach.

5) Converting α to common diversity measures

The framework is described as enabling conversion from α to diversity indices including Shannon, Simpson, and Hill diversities, using α as the central latent diversity parameter.

6) Skeptical interpretation: what could go wrong?

BINs as species proxies: only ~28% annotated to species; using BINs instead of fully resolved species can shift ecological meaning.
Sampling bias: sites are not uniform globally (e.g., potential concentration in protected areas), so covariate-diversity relationships may partly reflect sampling design.
Neutral framing + single σ assumption: the model assumes a common accumulation-rate structure; unmodeled niche processes may be partially hidden inside the neutral α parameter.
Generalization: the approach was built and tested in one arthropod barcoding/sampling context; transfer to other taxa or future dataset scales requires caution.

7) How to falsify the specific claims reported (from the same study’s criteria)

The study outlines falsification logic: claims would weaken if independent datasets show AET fails to predict α-diversity (or explains much less variance than reported), or if the zone-specific interaction patterns for human footprint do not reproduce.