Results from analyses of the effects of log properties and environmental conditions on longitudinal ground contact and our two measures of percentage cover by soil contact and ground vegetation, using 4-m branchless logs.

We used a set of 720 logs distributed over 122 experimental blocks in 30 sites. Results are from a full GLM model, a best fit model based on AICs, a hierarchical partitioning analysis (independent contribution), and a Pearson's correlation. The random factors “locality”, “site(locality)” and “block(site(locality))” were included in the full model and best model approaches, but do not show up in the table as a REML approach was used. Means from all logs within sites were used in the hierarchical partitioning analysis, because it cannot deal with random effects. Hence, this analysis only treats among-site (N = 30) variability in the response variables. The factors are sorted according to their F-value in the full model. Percentage data on cover were arcsine transformed before analysis. Longitudinal ground contact (0–7 points with contact) was not transformed. Degrees of freedom for predictors were 1 in all cases.
aChosen using the Akaike Information Criterion [40].
bFor tree species only qualitative relationships are presented, because this factor is binary and categorical.
c10 logs did not have ground contact at any of the sampling points and could therefore not be used here.