Environmental Engineering Reference
In-Depth Information
biological communities. The model is predicated on the relative position of a wetland
on axes of groundwater and atmospheric water gradients for prediction of biological
expression. Predictions from the model and hypothesized relationships form
hypotheses for future studies.
1.7 Experiment, Treatment, Replication,
and Randomization
An
experiment
imposes a
treatment
on a group of elements or subjects (experimental
units) to measure a response and quantify an effect. A treatment is something that an
investigator imposes on experimental units in some manner. Treatments can be
applied at varying amounts or magnitudes that are usually referred to as levels. For
example, if one was interested in the effect of the ratio of emergent plant cover towater
on bird use of a wetland and applied treatments (i.e., cover to water ratios) of 0, 0.25,
0.50. 0.75. and 1.00, then there would be five levels in the treatment. Frequently in
wetland studies, treatments are categorical and selected from nature (i.e., not investi-
gator applied [e.g., watershed type, season, anthropogenic modifications, soil type]).
There are numerous approaches to experimentation, with differences among
them related to relative scientific rigor and application of randomization and repli-
cation (Skalski and Robson
1992
; Morrison et al.
2001
). An experiment can also be
referred to as manipulative experiments (Hurlbert
1984
), comparative experiments
(Kempthorne
1966
) or randomized experiments (Kirk
1982
). A
controlled experi-
ment
is the ideal type of design that essentially isolates the dependent and indepen-
dent variables of interest while controlling identified nuisance or confounding
variables. An experiment produces results and conclusions of greatest rigor and
least uncertainty of results. A true experiment requires random allocation of
treatments to experimental units and replication of experimental units (Fisher
1935
). All identified nuisance variables are controlled or accounted for in the design
or through randomization. Frequently, random allocation of treatments to experi-
mental units or strict randomization of selecting experimental units or samples is not
feasible. In such instances, the investigator must take precautions to ensure that
personal bias or confounding (nuisance variable influencing the treatment effect)
effects do not cause doubt in the representativeness of the results. Wetland
experiments can be conducted in both laboratory and field settings. Some wetlands
types (e.g., prairie potholes, playas, vernal pools) easily serve as replicates in a
controlled experiment by their very nature of being small, isolated wetlands with
discrete boundaries defined by soil, vegetation, or hydrology. For wetland types
(e.g., coastal marsh, peatlands, bottomland hardwoods) that exist over large areas
(e.g., 100s-1,000s ha) with no discrete demarcation, it is difficult to identify distinct
experimental units and special care is needed to artificially define experimental units
to conduct an experiment in these systems to ensure that results are representative.
