Environmental Engineering Reference
In-Depth Information
analysis of variance. If only one experimental unit is assigned per treatment, then no
statistical inference beyond the experimental units sampled is possible because
experimental error cannot be estimated. Much too frequently in wetland science, a
study is conducted once with weak evidence for conclusions that results in uncer-
tain knowledge; repeating the study would either strengthen evidence for
conclusions or show that the initial conclusions are not supported allowing for the
development of alternative hypotheses.
Another form of replication is the practice of repeating a study to strengthen
conclusions. Results based on studying a fewwetlands in a limited area or constrained
environmental conditions (e.g., only wet or dry years) could be confirmed by a similar
study or succession of studies conducted over larger temporal and spatial scales to
determine if conclusions hold under more general conditions. For example, Luo
et al. (
1997
) concluded that unsustainable accumulation of upland sediment was filling
playa wetlands and represented the greatest impact to this unique wetland system.
Their results were based on data from 40 playas (20 cropland watersheds and
20 grassland watersheds) in a limited spatial distribution. However, since the initial
study, a number of subsequent studies have confirmed that the original conclusion is
valid at larger spatial scales and under a variety of environmental conditions with
steadily increasing evidence of negative impacts of sediment accumulation on playa
wetlands (e.g., Tsai et al.
2007
,
2010
; Johnson et al.
2011
,
2012
; Smith et al.
2011
;
Burgess and Skagen
2012
; O'Connell et al.
2012
).
The concept of pseudoreplication is frequently confused with true replication
(Hurlbert
1984
). Replication is based on the number of experimental units
whereas pseudoreplication usually refers to multiple measurements from a single
experimental unit that are treated as independent experimental units during
analyses. For example, if one was interested in biomass production between
grazed and ungrazed wetlands (i.e., grazed/ungrazed are treatments) but only
applied each treatment to a single wetland of each treatment and clipped and
weighed aboveground biomass in 20 plots/wetland, then analyses using plots as
experimental units would be pseudoreplicated. Furthermore, because the
treatments were applied to the entire wetland, the plots are samples and not
individual experimental units randomly assigned to a grazing treatment. Because
there is only one experimental unit per treatment, estimation of experimental error
is impossible because the variation among samples within each experimental unit
would be considered sampling error (i.e., variation among samples of a given
experimental unit; see Sources of Error below). Pseudoreplication can easily be
avoided when it is understood what the unit is to which the randomization rule
applies when assigning treatments to experimental units. The results from studies
that include multiple samples from single experimental units should not be
considered invalid because it is often difficult or impossible to replicate certain
experimental units (e.g., oil spills in a coastal marsh) in applied ecological
research (Wester
1992
). However, it is critical to realize that inference of results
can be strictly extended only to the experimental unit(s) sampled. In most
ecological fields, use of pseudoreplication is considered a fatal flaw for studies,
but many times this approach is appropriate in wetland studies as there may exist
