What the index asks
The Priority Species Index asks one question of a whole landscape: are the species that need
the most room doing better or worse than when counting began? It is built for species that are
rare, threatened and area-sensitive — the ones a landscape either keeps whole or loses.
How it is computed
1Each species against its own baseline
Each species' benchmark is its own first survey that fits the paper's criteria (SM Appendix
C), set to 100, and every later value is divided by it. Holding steady scores 100. Doubling
scores 200. The first surveys of different species need not fall in the same year.
2The geometric mean, never the arithmetic
An arithmetic mean lets one recovering species drown out four declining ones: a rise from 100
to 500 outweighs four falls from 100 to 50. The geometric mean treats a halving and a doubling
as equal and opposite, which is what a reader assumes an index does. Where species are not
surveyed in the same years, the manuscript states the rule in Methods (5): the index is
calculated every two to three years, and all estimates in that interval are averaged. Each
point is one geometric mean over every species value in the interval since the previous
point, so a species surveyed twice in it counts twice. SM Appendix C describes the rule
differently, as the most recent survey for each species in the interval; that question is
with the authors. Ndoki-Likouala's stated values sit slightly above the geometric mean of its
species values and agree, in our reproduction, with the mean of the simulation in step 3; the workbook does not label
which statistic it states.
3Uncertainty by simulation
SM Appendix C estimates uncertainty by Monte Carlo, adapting Soldaat et al. (2017): each
survey's standard error is taken from its reported 95% confidence interval, 10,000 values are
drawn for each species and year from a log-normal distribution, and the 95% interval runs from
the 2.5th to the 97.5th percentile of the simulated index. The manuscript calls the resulting
intervals optimistic and illustrative where variance had to be assumed. No interval on an
index is published in these records except where one is drawn: the inside- and
outside-protection series for Ndoki-Likouala and Rio Juruá. Nothing on this page is a
recomputed interval.
PSIi = exp[ (1/Si) · Σ ln(PRs,i) ]
PRs,i = (Ns,i / Ns,1) × 100
The formula is SM Appendix C's, where Si counts species. Under the
manuscript's Methods (5) the sum runs over every survey value in the interval, and Si
is the number of values.
Three things the index is not
Not a population count. It is a ratio to a baseline, so a landscape that
began badly and improved can outscore one that began well and held.
Not one method. Across these seven landscapes the underlying numbers are
population estimates, camera-trap detection rates, line-transect densities, occupancy values,
and in one case what experienced local fishers and hunters report having seen over twenty years.
Each landscape's record states its own method and a caveat on reading it. That wording is this
page's, drawn from the paper's figure captions and Appendix A; where the paper states a limit
itself, the record quotes it and says so.
Not comparable between landscapes. Two indices sit on the same axis here
because they are the same kind of quantity, not because 140 in one place means what 140 means in
another.
Where the boundaries come from
The frame the paper screens from
The formula, and what is drawn around the seven
What is withheld, and why
Provenance
Priority Species Index