Appendix S1

Specimen voucher numbers, identification, and geographic location of samples for this study.


STEP 1 Relationships between variables
Compute correlations between species diversity, SoK and PA size (S1.1)

STEP 2 Negative binomials models
Calculate negative binomial models to determine the effect of SoK on species diversity (S1.2). Use the standardized residuals as site-level relative species diversity for each taxon STEP 3 Taxon-specific data Combine taxon-specific data (SoK and relative species diversity) across all taxa into site-specific values in two ways: 1. Estimate the first principal component of a PCA to generate site-specific PC1 scores (S1.3 and S1.4) 2. Calculate site-level unweighted mean values across taxonomic groups to generate site-specific mean values

STEP 4 Model selection
Assess the differences between the PCA-generated values and the unweighted mean values to decide what to use when generating individual priority ranks (S1.5)

S1.1 Determining the relationship between variables used to build priority ranks 12
There was a tendency for the known species diversity to increase as the state of knowledge 13 (SoK) increased ( Fig. S1.2). Species diversity also showed a positive association with protected 14 area (PA) size for most taxa, though SoK was weakly related to PA size (Figs S1.3 and S1.4), 15 suggesting surveying effort by experts were evenly distributed among PAs of various sizes. 16  (right column) for endemic rodents, bats, endemic carnivora and lemurs (separated by rows). 28 29 30 S1.2 Negative binomial models 31 Several negative binomial models (one for each taxon) were performed to determine the effect 32 of SoK on species diversity by taking protected area (PA) size into account. Species diversity 33 was modelled as dependent on SoK (categorical variable) and log 10 -transformed PA size, as 34 well as their interaction. Since there were no significant interaction effects (p > 0.05) between 35 log 10 of PA size and SoK, the results below (R output) are shown only for the individual effects 36 of the predictor variables. SoK was a significant predictor (p < 0.05) of species diversity for all 37 taxa, while PA size was significant for most taxa (excluding amphibians and carnivorans). A 38 significant association with PA size likely reflects biological reality (i.e., more species in larger 39 areas), while a significant association with SoK likely reflects bias due to sampling effort. 40 Negative binomial models were fitted in R (R Development Core Team 2019) with the function 41 glm.nb from the MASS package (Venables & Ripley 2002

S1.3 Species diversity by state of knowledge (SoK) 136
In order to account for bias due to sampling effort (i.e., differences in SoK), we recalculated 137 negative binomial models for each taxon in which species diversity was dependent solely on 138 SoK. We then used the standardized residuals from those models to represent relative species 139 diversity ( Fig. S1.5). We performed principal components analysis (PCA) on the standardized residuals of 147 species diversity vs. SoK. The first two principal components captured 40.7% and 15.2% of 148 the variation, respectively. All taxa loaded in the same direction, though not to the same degree 149 as reptiles and bats showed particularly low loadings (Fig. S1.6). Sites that tended to be more 150 diverse than expected for its SoK for one taxon also tended to be more diverse for other taxa. 151

Endemic.Carnivora
Lemurs partitioning the variance among variables, which eliminated the need for a value judgement 162 about which taxa were more important at the cost of others. 163

State of knowledge (SoK) 165
The first and second principal components captured 61.4% and 12.7% of the variation, 166 respectively with all taxa loading in the same direction, though not to the same degree (Fig.  167   S1.7). The first principal component alone could be used to build a ranking based on SoK, but 168 since it is not justified to preference one taxon over the other, it may be more desirable to 169 weight all taxa equally. second accounted for 15.8%, with forest types loading in several different directions (Fig. 197 S1.9). This was probably because the forests varied in their makeup. 198 Since they all showed non-negative rates of loss, we simply looked at total forest loss 199 between the two time periods (1996-2006 and 2006-2016). The first principal component 200 captured 52.3% of the variation, with the remainder in the second principal component.

Relative species diversity (mean)
Standardized residual species diversity (mean)