Study Design and Logistics - Wetland Techniques - page 25

prescribed strategy. Numerous textbooks are available to assist in designing
sampling strategies beyond what is described in this chapter (e.g., Cochran 1977 ;
Scheaffer et al. 1979 ; Thompson 1992 ). Points, plots, transects, and marking
captured animals are among the techniques used to sample experimental units.
The goal of sampling is to achieve an unbiased (closeness of observed values to
true value) and precise (close proximity of repeated measurements of the quantity)
estimate of a population parameter value (Fig. 1.2 ). Ideally, sample measurements
for an estimator should have a narrow range of variation (i.e., precision) centered on
the population value (i.e., unbiased), which represents an accurate estimate. For
example, an objective of sampling must be to produce a sample mean population
parameter with low variance around the sample mean. Groups of sample
measurements that are centered on the population value but yet have a wide
range are considered unbiased, but the presence of a high variance will decrease
the reliability of detecting treatment effects. Biased samples generate a mean or
other statistic that is not representative of the population parameter, but can have a
narrow (precise) or wide range of values. A sample that is both unbiased and
imprecise yields little information relative to the target population. Unfortunately,
it is rarely possible to determine if one has an unbiased and precise sample because
rarely are population means and variances known. Finally, wetland systems are
exceptionally complex such that strict adherence to sampling schemes may be
difficult, even for laboratory studies. However, it is critical that protocols associated
with sampling designs be followed as explicitly as possible. Inference from samples
to a target population is conditional on the protocol for selection of study sites and
subsequent sampling. Thus, information from any sampling design is subject to
interpretation based on the context in which the samples were collected.
Sampling protocols can be categorized as (1) haphazard sampling, (2) judgment
sampling, (3) search sampling, and (4) probability sampling (Gilbert 1987 ). There
are many variations of the sampling process within these categories (Gilbert 1987 ;
Gilbert and Simpson 1992 ) that are beyond the scope of this chapter. The sampling
designs described below are not meant to be inclusive of all possible approaches,
but rather a description of those that would be commonly used in wetland studies.
However, as a caveat, a minimal goal for reliable inference of results is some form
of probability sampling where all potential experimental units have the same
probability of being selected as a sample. This strategy produces unbiased estimates
of the population mean, variance, and other attributes. Frequently, the phrase
sampling frame is used to describe a list of all members of a target population
(i.e., elements) that potentially can be sampled (Jessen 1978 ). In field studies, the
sampling frame is usually spatially (study area) or temporally (study period)
defined. In laboratory and human dimension studies, the sampling frame is usually
defined by a list of all potential elements that could be selected for study. Finally, it
is highly recommended that any proposed sampling design be reviewed by a
statistician or quantitative biologist to ensure that all possible contingencies have
been addressed and the proposed sampling strategy will allow for the desired
inference of results.