Environmental Engineering Reference
In-Depth Information
time [day, month, year] or environmental condition, wetland area, water depth),
unless the same experimental units are used. Statistical inference only can be
applied to the target population under the actual (fixed) treatment levels that are
within the range of random variables being tested. When a study includes both fixed
and random variables, it is considered a mixture of effects and requires some
additional consideration during analyses (i.e., mixed models). It is important to
define variables as fixed or random when describing the methods used in the study.
When more than one independent variable is being assessed in a study, the
interaction between effects is of great interest. A significant interaction indicates
that the magnitude of differences between levels of one effect depends on the level
of the other effect. Many times the interaction between effects is more interesting
than individual main effects in explaining data observations and results, albeit this
is frequently considered more cumbersome to explain than results for simple main
effects. However, an investigator must use their knowledge of the system to ensure
that significant interactions have biological meaning and are not a spurious result.
Spurious (an apparent relationship between noncausal events or variables) results
typically result from the presence of a confounding or nuisance variable. At times,
further investigation of interactions is necessary to develop confidence that the
interaction is meaningful.
In addition to the proper identification of dependent and independent variables,
there are many other types of variables that can impact results and should be
considered during development of a study design. It is important to categorize all
variables that contribute to the variation of dependent variables into those that are of
interest and related to the hypotheses being tested and those that are
nuisance
variables, which are assumed to be of little interest but may affect study results.
Extraneous or nuisance variables can have disproportionate impacts on results from
a study unless accounted for in the study design. Indeed, the failure to control for
nuisance variables frequently results in spurious conclusions. Through study
design, nuisance variables can be controlled to account for the potential bias
associated with the variable. Examples of methods of controlling nuisance variables
in study designs using analysis of variance include grouping experimental subjects
into
blocks
(use of some common characteristic to group homogeneous subunits of
the sampled population) or use of a
covariate
(a random variable of little interest
associated with but varies among experimental units) if the variable is categorical
or continuous, respectively.
There are a variety of approaches to account for nuisance variables in a study
design. For example, in a study of avian response to prescribed fire in wetlands,
Brennen et al. (
2005
) acknowledged that migration timing (changing avian densities
over time) and wetland size (species-area relationship) could influence results, yet
these variables were not of primary interest in assessing the effect of spring burning
of wetlands. Furthermore, the investigators recognized that conducting a wetland
study over a large geographic region could be influenced by varying environmental
conditions (e.g., differing precipitation patterns) across the target population of
wetland. Therefore, because the primary interest was in avian response to a burning
treatment, they paired adjacent burned and unburned (control) wetlands across the
