Nonmetric multidimensional scaling
Nonmetric multidimensional scaling (NMDS) was used to examine the effects of seismic line density on songbird community composition. NMDS is an ordination method that does not require data to be normally distributed and is effective for datasets with multiple zeroes, as is the case for both my aspen and black spruce datasets (McCune and Grace 2002). The NMDS technique first calculates a dissimilarity matrix based on a distance measure (Bray-Curtis in this case), and the distances between all pairs of sampling units are then ranked. Next, it searches for the position of the sampling units on the ordination axes for which the distances between all pairs of sampling units is in rank-order agreement with their dissimilarities in species composition, as far as possible. This is the position which minimizes the stress of the configuration. In the resulting plot, points which are closer together represent sampling units that are more similar to each other, whereas points farther apart are less similar. In my NMDS plots, I colour-coded the points by treatment type in order to easily examine the effects of seismic line density on composition. This analysis was conducted using R (R Development Core Team 2008).
Multi-response permutation procedures
Multi-response permutation procedures (MRPP) is a non-parametric approach which tests whether there is a significant difference between two or more groups for multiple response variables combined (McCune and Grace 2002). I used MRPP to test whether songbird community composition differed significantly between seismic line density treatments (0 km/km2, >0-3 km/km2, >3-6 km/km2, and >6 km/km2 line density within 200 m for the aspen data; none, single seismic line, and double seismic lines for the black spruce data) . Because MRPP is non-parametric, it does not assume that data is normally distributed but does require that the chosen distance measure (Bray-Curtis in this case) adequately represents the variation in the data. An MRPP first calculates a distance matrix and determines the average distance between objects within each treatment. The MRPP statistic (delta), which is a linear combination of the within-treatment mean distances, is then calculated. This MRPP statistic is compared to a probability distribution generated under randomization to determine the probability of obtaining a value as small or smaller than the observed statistic. This analysis was conducted using PC-ORD (McCune and Mefford 2006).