Both quantitative studies, including the experiment,were essentially
observational in the sense that prior presence of non-native species had
to be taken into account in design and therefore, treatments could not be
allocated at random to experimental units. Observational studies are
more susceptible to biases than experiments because treatment
allocation may have been influenced by response variables (Eberhardt
and Thomas, 1991). In our case, impacts on native fish communities
could have been underestimated if non-native species had been
released preferentially in wetlands with above-average abundance or
diversity of native species. There is no indication that this would have
been the case. The paired design successfully reduced environmental
variation and increased the statistical power of the impact study, as is
evident for example from the noticeable correlation species richness
values between impacted and control sites (Fig. 1). Comparison of
wetland characteristics between impacted and control sites indicated
that mean total phosphorus concentration and fishing intensity were
slightly, but not significantly higher in impacted sites. Regression
analysis of relationships between impact indicators, TP and fishing
intensity showed that observed differences in the latter are unlikely to
confound the impacts estimated from the paired comparisons.