2.4. Statistical analyses
For both studies, data were averaged across depths within sites,
and log2 transformed to approximate normal distributions. Linear
models were used to examine the variation in bioindicator candidates
in response to water quality (Study 1) and to water quality
and regions (fixed) (Study 2). The presence of interactions between
water quality and regions was assessed in the model selection process,
but none were detected and hence only additive effects of
predictors were used. The results were presented as (1) tables to
document effect sizes, directions and their significance, and (2)
graphical displays of the form of the dependency of indicators on
water quality and region. The latter are called partial effects and
are conditional on the regions being in the model but held constant.
Redundancy analyses were used to investigate the joint responses
of all indicator candidates to the predictors (WQI in
Study 1, turbidity and chlorophyll, conditional on regions, in Study
2). Lastly, to calibrate the bioindicator system, data from Study 1
and 2 were combined, and the three water quality variables, turbidity,
chlorophyll and total suspended solids, were separately related
to the final 12 bioindicators from all sites using linear models.
All statistical analyses were done with the software R (R Development
Core Team, 2010).