Correction to: Nature https://doi.org/10.1038/s41586-023-06410-z Published online 23 August 2023

In the version of the article initially published, the Extended Data tables were inadvertently truncated. They have now been moved to Supplementary Information, where they can be viewed in full.

Two of the originally published Extended Data (now Supplementary) tables in our article contained values that had not been updated from previous runs of the analytical models. We have updated the coefficient values in Extended Data (now Supplementary) Table 1 and the associated propensity score matching results in Extended Data (now Supplementary) Table 3. Most of these changes were very slight. The corrections did not alter the significance (or lack thereof) of any of the covariates except that bird species richness was now significantly related to ‘distance to protected area [PA]’ while mammal functional richness was not. We have updated the main text of the paper accordingly, although our original statement is still true that “For sampling locations outside PAs, distance to the nearest reserve was significantly associated with only one of the six diversity metrics…”. This was a relatively minor point in our paper that was not reported in the Abstract or depicted in the figures, and on which we did not base our inference.

We have also corrected two statements in the Methods. In the first we removed reference to a statistical interaction between the human development index and accessibility. We had initially tested this but removed it from the models on learning that interactions are not interpretable or recommended in propensity score matching. In the second, we removed reference to a covariate for forest understory density. We had initially examined this but removed it because the effect was negligible and it was tangential to the hypotheses we were testing.

The corrections to these tables and Methods do not alter our main results, major inference, Abstract or figures.