Environmental Engineering Reference
In-Depth Information
locations are selected based on the known occurrence of variables of interest (e.g.,
certain plant species, known nesting locations of birds) rather than an informed
opinion that the variables of interest would be found at a certain location.
The strongest inference comes from data collected using a form of probability
sampling. Probability sampling is when all elements within a defined population
have an equal probability of being selected to be sampled and that probability is
known. There are a number of probabilistic sampling schemes including simple
random sampling, stratified random sampling, and systematic sampling (see below,
Fig. 1.3 ). By selecting experimental units at random, statistical properties are
unbiased and represent the target population. Furthermore, such samples allow
for evaluation of the magnitude of treatment effect size. These sampling strategies
can range from rather straight forward to increasingly complex depending on
restrictions (e.g., subsets of experimental units and nuisance variables that may
need to be addressed). In most wetland studies, elements or experimental units are
selected without replacement as each element appears only once in a sample (Levy
and Lemeshow 1991 ). Sampling with replacement is when elements are returned to
the target population following measurement and have the potential to be sampled
again. Sampling without replacement increases precision of the results (Caughley
1977 ). There are several types of probability sampling strategies.
One type of probability sampling is simple random sampling , which occurs
when each element of a sampling frame or target population has an equal probabil-
ity of being selected. The elements or experimental units are not subdivided or
stratified in any manner. Random selection of each element is independent of all
other elements. Morrison et al. ( 2001 ) outlined five basic steps to achieve a simple
random sample. First, the investigator assumes that the target population consists of
a finite number of elements (i.e., experimental units) available to be selected. All
selected elements can be located, accessed, and the variable(s) of interest can be
measured without error. The elements must occur throughout the sampling frame
and cannot overlap in any manner. Elements do not need to be identical, but as
differences among elements increase in magnitude or subsets occur, then a more
complex design may be necessary to avoid biasing a sample with overrepresenta-
tion by certain element types. All elements are normally sampled (i.e., consist with
all other elements) without replacement.
Use of simple random sampling can be problematic if the target population is
comprised of groups or subsets of similar elements. In wetland studies, this occurs
when elements are clumped and patchy, such that a relatively small sample size
(typical for field studies) may result in an overrepresentation of certain groups or
elements with distinctive characteristics that can skew results to properties of
subgroups rather than the entire target population. Dividing the elements of a
target population into independent subsets or groups (i.e., strata) and then apply-
ing a random sampling approach within each stratum can increase the likelihood
that results are representative of the target population in addition to increasing
knowledge for elements of distinct strata that could be missing using a simple
random sample.
Search WWH ::




Custom Search