Extended Data Table 1 Robustness of results when accounting for spatial autocorrelation

From: Large reductions in tropical bird abundance attributable to heat extreme intensification

  1. The results of models with either the full (1–4) or reduced set (5–8) of climate variables, using different specifications to assess the uncertainty of parameter estimates. Estimates are shown when the covariance matrix is adjusted for clustering at the population level (1, 4); when adjusted as to assume full autocorrelation of all variables between populations located up to a distance of 1000 km from one another, using the specification of Conley (2, 5); and when assuming arbitrary levels of autocorrelation between all populations using the specification of Driscoll-Kraay (3, 6). All covariance matrix adjustments also assume and thereby correct for heteroskedasticity and temporal autocorrelation. Across all choices the impacts of heat extremes and their interaction with latitude remain strongly robust. Estimated coefficients are shown for each dependent variable and indicate the effect on the abundance growth rate (changes in the logarithm of abundance) per unit change of the independent variable. The standard errors of these coefficient estimates are shown below in parentheses and the significance of each estimate is indicated following a two-sided t-test with the null hypothesis that the coefficient parameter is zero.
Back to article page