Table 1. Parameter estimates and performance statistics of models predicting biodiversity loss.

Countries
	Parameter estimates (standard errors)					Performance statistics
Model	Total number of plant and vertebrate species 2004	Human population size 1989	GDP PPP per capita 1989	GDP PPP per capita 1989 (quadratic)	Gini ratio of household income inequality 1989	P-value of inequality	ΔAICC	r 2
Power	0.81 (0.16)	0.10 (0.08)	−0.03 (1.58)	2.8×10−3 (0.09)	1.76 (0.34)	6.4×10−6	21.0	0.86
Linear	0.01 (2.3×10−3)	2.6×10−7 (1.2×10−7)	−2.1×10−3 (.01)	1.5×10−7 (6.6×10−7)	445.7 (208.8)	0.04	2.2	0.74
Negative binomial	6.1×10−5 (1.1×10−5)	9.7×10−10 (6.1×10−10)	2.9×10−5 (7.2×10−5)	−1.2×10−9 (3.3×10−9)	5.80 (1.05)	3.0×10−8	16.7	–

The dependent variable is the number of threatened plant and vertebrate species in 2004 (for countries) or of permanent resident bird species with statistically significant (P<0.10) declines in abundance from 1966 to 2005 (for US states). The sample size is 45 for both countries and states. For the power models, all variables were log-transformed before performing the regressions. Prior to log-transforming the numbers of declining permanent resident bird species in US states, however, we added 1 to each, since some US states had no declining species, and the logarithm of zero is undefined. For the linear and negative binomial models, no variables were log-transformed. See the Materials and methods section for more details, including an explanation of the quadratic term of GDP PPP/income per capita. Statistically significant parameter estimates are in bold (P<0.05). ΔAIC_C is the advantage, in terms of Akaike's Information Criterion corrected for small sample sizes, of the model shown, in comparison to the corresponding null model with the inequality term removed [27].