关于R包adespatial的beta.div.comp与betapart的beta.multi计算结果差异的技术问询

Hey there, great catch on this discrepancy! The different outputs from these two functions aren't due to code bugs—they come down to two fundamentally different ways of calculating and summarizing beta diversity under Baselga's decomposition framework. Let me break this down clearly:

1. Core Difference: Pairwise Averaging vs. Global Summary #

adespatial::beta.div.comp(coef="BJ") #

This function takes a "pairwise-first" approach:

1. It calculates the Jaccard dissimilarity (and its turnover/nestedness components) for every single pair of sites in your matrix.
2. It then averages all those pairwise values, and divides by 2 (since the distance matrix includes each pair twice, and sum(D)/(n*(n-1)) is equivalent to mean(D)/2).
3. The result you get is the average pairwise beta diversity across all site pairs, split into turnover and nestedness.

As you already verified, you can replicate its outputs directly with betapart's pairwise function:

pair_res <- beta.pair(A, index.family = "jaccard")
# Matches beta.div.comp$part$BDtotal
mean(pair_res$beta.jac) / 2
# Matches beta.div.comp$part$Repl
mean(pair_res$beta.jtu) / 2
# Matches beta.div.comp$part$Nes
mean(pair_res$beta.jne) / 2

betapart::beta.multi(index.family="jaccard") #

This function takes a "global-first" approach:

1. It first computes summary statistics across the entire community matrix—like total species richness in the region, sum of site-level richness, total shared species across all sites, and aggregated counts of species gains/losses.
2. It plugs these global stats directly into Baselga's formulas for beta diversity decomposition, producing a single value that represents beta diversity at the whole-community scale, not an average of pairwise differences.

2. What This Means For Your Analysis #

Neither approach is "wrong"—they answer different research questions:

• Use beta.div.comp if you want to describe the average degree of differentiation between individual site pairs (e.g., "On average, any two sites share X% of their species").
• Use beta.multi if you want to describe how much the entire community deviates from the regional species pool (e.g., "Across all sites, only Y% of the regional species are present in any single site").

3. Quick Validation of the Relationship #

To confirm this, run both calculations side-by-side with your test data:

# adespatial's average pairwise total beta
mean(beta.pair(A, index.family = "jaccard")$beta.jac) / 2
# Output: 0.3258676 (matches beta.div.comp$part$BDtotal)

# betapart's global total beta
beta.multi(A, index.family="jaccard")$beta.JAC
# Output: 0.9356033 (global community-scale beta)

The gap between these numbers is exactly the difference between averaging pairwise dissimilarities and calculating a single global dissimilarity metric.

内容的提问来源于stack exchange，提问作者Manuel Popp