A novel Bayesian approach to assessing impacts of rain forest logging

• Published in: Ecological applications 1996-11, Vol.6 (4), p.1104-1123
• Main Authors: Crome, F. H. J., Thomas, M. R., Moore, L. A.
• Format: Article
• Language: English
• Subjects: Applied ecology; APROVECHAMIENTO DE LA MADERA; Aves; Bayesian analysis; Birds; BOSQUE TROPICAL HUMEDO; Forest canopy; Forest ecology; FORET TROPICALE HUMIDE; LOGGING; LOGGING EFFECTS; Mammalia; Mammals; Microhabitats; Rats; RECOLTE DU BOIS; Species; TROPICAL RAIN FORESTS
• ISSN: 1051-0761; 1939-5582
• DOI: 10.2307/2269595

We present a novel application of Bayesian procedures to assess the impacts of logging rain forest on birds and small mammals in tropical Queensland, Australia, using data from a 1983-1986 BACIP (measures made before and after on control and impact sites) study. The procedure was compared with the usual approach to the analysis of BACIP designs following the methods of Stewart-Oaten et al., which are based on classical Neymann-Pearson significance testing. Significance tests were performed at the 0.05 and 0.1 levels, power being calculated for a 25% reduction in species capture rates. For the Bayesian analysis, we elicited one noninformative and three informative prior distributions representing polarized beliefs about the effect size of logging: a strong negative effect, little or no effect, and an effect related to the amount of canopy loss. Effect size was estimated by determining the extent to which there was agreement between the posteriors for eight propositions concerning effect size. These propositions ranged from a negative effect of >25% reduction being very likely to a large positive effect of $>25% increase being very likely. Of 76 bird species recorded and nine species of mammals captured, there were sufficient data for analysis of 21 bird species, five ecological groups of birds, and five mammal species. The bulk of the classical tests were of low power; of the 99 species/microhabitat combinations tested, only 10 were significant at the 0.05 level (12 at the 0.1 level). The "standard" classical analysis allowed few conclusions other than the data were uninformative. The Bayesian procedure was more informative. For 68 species/microhabitat combinations, there was consensus among the posterior distributions for one or more of the propositions about effect size. The Bayesian analysis indicated that, over the entire study area, negative effects were not likely to be greater than the degree of canopy opening. However, in the microhabitats that received most of the damage, negative effects were likely for far more species.