2.4. Data analysis
In preliminary statistical analyses, the day 0 (initial) samples re-
vealed strong between site differences in sediment properties and mac-
rofaunal community structure. These initial, site-specific differences in
community structure persisted throughout the experiment and resulted
in significant site × treatment and site × time interactions in subse-
quent uni- and multi-variate analyses (described below). Recognising
these initial differences, we chose to present separate analyses for
each site to better focus on the effects of treatment and time, and to bet-
ter illustrate differences in the response between sites.
Non-metric multi-dimensional scaling (nMDS) analysis, using
Bray–Curtis similarity matrices, was used on raw macrofaunal abun-
dance data to plot and compare benthic community structure among
treatments, as well as through time. We analysed these changes sta-
tistically using a repeated measures permutational multivariate
analysis of variance (PERMANOVA, Bray–Curtis similarity). Time
and treatment were treated as fixed factors (with 5 and 3 levels, re-
spectively), with plot (6 levels) treated as a random factor nested
within treatment (Anderson et al., 2008). PERMANOVA pair-wise
tests were used to determine where significant treatment and time
effects occurred, and SIMPER analysis determined the taxa that con-
tributed to the dissimilarity/similarity in community structure be-
tween treatments. The percent dissimilarity to standard deviation
ratio (Diss/SD) was also used to determine whether the taxa
identified in SIMPER analyses were good discriminating taxa (Diss/
SD N 1.3) (Clarke and Warwick 1994). Raw data were used for multi-
variate analyses as transformations did not alter the results. All mul-
tivariate analyses were performed using the PRIMER (with the
PERMANOVA A+ addition) statistical software program (Plymouth
Routines In Multivariate Ecological Research; Anderson et al., 2008;
Clarke and Gorley, 2006).
A repeated measures analysis (two-way, fixed factor, repeated mea-
sures analyses of variance (ANOVA)) was also used to test treatment ef-
fects on univariate variables (sediment properties, and macrofauna
taxonomic richness and abundance) through time. Newman–Keuls
post-hoc tests were used to determine differences between treatments
for each sample date. Raw data (including the initial day 0 samples,
which we analysed via t-tests) conformed to assumptions of homoge-
neity of variances and normality, therefore no transformations were
necessary. Univariate analyses were conducted using the STATISTICA
software package (Statsoft Inc.)