Our estimates of BII for a country, region or the world are derived from statistical modelling of the whole data set.
To calculate the uncertainty of these estimates, we fitted the same statistical models ten times, leaving out the data from one major biome (broad natural habitat type) each time, and used these to project BII.
This technique, known as cross-validation, yielded ten estimates and we report the lowest and highest of these as our uncertainty margins at the end of each decade.
The range of estimates can be quite wide, partly because two of the biomes in particular - temperate broadleaf and mixed forests, and tropical and subtropical moist broadleaf forests - are particularly well represented in the data.
Cross-validation commonly leaves out a random subset of the data each time, but that approach underestimates the true uncertainty in parameter estimates.
In the future, we plan to also cross-validate by splitting the data up into major taxonomic groups.
Indicator: The Biodiversity Intactness Index, accessed through the Biodiversity Trends Explorer
Data set: Available through the NHM Data Portal
Modelling framework: Available through GitHub
Related Museum project: PREDICTS
Project and research leads: Professor Andy Purvis and Dr Adriana De Palma
Data last updated: October 2021