Study Design and Logistics - Wetland Techniques - page 13

habitats (e.g., isolated wetlands, islands), subdivided units within larger ecosystems
or habitats (e.g., management units of a contiguous system, pastures, watersheds),
or some measure of time.
Experimental units can be natural features (e.g., wetland, bird, plant) or
man-made (e.g., mesocosm, microcosm, greenhouse flat). Balcombe et al. ( 2005 )
tested hypotheses that invertebrate family richness, diversity, density, and biomass
were similar between mitigation and reference wetlands. As experimental units,
they selected 11 mitigation and four reference wetlands across three physiographic
regions of West Virginia. Maurer and Zedler ( 2002 ) tested hypotheses contributing
to the invasion of reed canary grass ( Phalaris arundinacea ) using a parent plant
transplanted into a cone-tainer and attached to aluminum troughs to measure tiller
growth over time in response to shade and nutrient treatments; each cone-tainer was
an experimental unit.
All study designs involve identification, measurements, or estimation of variables
considered to affect the hypothesis being tested. There are several classes of
variables to consider during study design. Basic to statistical models are independent
and dependent variables. Independent variables are those hypothesized (including
treatments) to contribute to variation in dependent or response variables , the
values of which depend on levels or types of independent variables. In most study
designs for wetlands, there is one dependent variable of interest; for example,
density of waterbirds, species richness of invertebrates, levels of nutrients in water
runoff, and soil moisture. However, there can be several associated independent
variables that may be categorical or continuous variables that are hypothesized to
influence the variance of the measured dependent variables; for example, wetland
type, wetland area, watershed condition, vertebrate sex and age, and time.
Analyses related to a single dependent variable are termed univariate , and there is
a long history of established methods to test hypotheses involving a single dependent
variable for both discrete (i.e., categorical [e.g., chi-square analyses] or factor
variables [e.g., analysis of variance]) and continuous (e.g., regression) independent
variables. However, simultaneous analyses of greater than one dependent variable are
often of interest and use of multivariate statistics (e.g., ordination, principal
components analysis, path analysis) has greatly increased during the past three
decades with advances in computing power necessary to conduct these analyses.
Regardless of the approach, the focus of an established study design is to quantify the
relationships among dependent and independent variables through some form of data
analysis. The goals of data analysis include evaluation of hypotheses, predicting or
forecasting an event or response, development of structure of future models, deter-
mination of important variables relative to variation of the dependent variable(s), and
detection or describing patterns and trends.
Each independent and dependent variable needs to be determined as a fixed or
random effect prior to determining the appropriate study design and analyses. A
fixed effect is a variable in which levels are not subject to random variation under
repetition of the experiment (e.g., wetland type, animal age and sex, levels of
nutrients applied, number of seedlings planted). A random effect is one where
repetition of the experiment will result in different levels within the analyses (e.g.,