Significant drivers of forest trends (0: no forest loss, 1: forest loss) modelled using general linear models with logit link functions.

Numbers in brackets give the mean and standard error of the coefficient; associated P values are given at *P<0.05,
**P<0.01,
***P<0.001).
We computed deforestation models for East Africa (Model 1) and for a subset of the study area (Model 2) because geographic data on the spatial location of towns and roads were available for the countries listed in Model 2 only [58]. ‘Protection Status’ is treated as categorical variable with the terms: Game Parks, Not Protected, National Parks, Nature Reserves, Other Protection). A subsequent Wald Chi-Squared test indicates that the overall effect of Protection Status is statistically significant (P<0.0001). Abbreviations: likelihood ratio (LR), McFadden’s pseudo R2 (Pseudo-R), Akaike Information Criterion (AIC) and significance of model (P).