Environmental Engineering Reference
In-Depth Information
size of scientific interest. This determination is not a statistical decision but one that
must be made by the investigator. Such a decision is important in the context of
other constraints to sampling effort and should be made with a realistic expectation
that one would expect to find at the conclusion of the study.
Appropriate sample size can be computed explicitly via formulas or through
simulations. Use of previously collected data for the variables of interest in the
system being investigated can be used to explicitly calculate the necessary sample
size. Use of data from preliminary (i.e., pilot study - a preliminary period of
reduced data collection using the proposed study design) studies and literature
values can be used to estimate necessary sample size. Tragically, many published
and most unpublished studies with nonstatistically significant findings contain
statements apologizing for such findings and blaming it on the lack of a sufficient
sample size. Investigators should strive to avoid such situations to the extent
possible because concluding that results from a wetland study are essentially
meaningless due to insufficient sample size adds little to scientific process and
squanders precious resources and time.
Assessment of the sample size and statistical power to measure an effect is critical
to study design. Both of these aspects require an acceptable measure of precision.
Therefore, one needs to measure or estimate the level of variation associated with
each dependent variable to evaluate the ability of the proposed study design to
produce meaningful results. One can accomplish this either through use of values
in the literature or conducting a pilot study. There are numerous formulae and
approaches for sample size determination and determination of power; many of
which are available as calculators on a variety of websites or in statistical software
packages. There are a number of on-line and software sample size calculators (see
http://www.epibiostat.ucsf.edu/biostat/sampsize.html?iframe
¼
true&width
¼
100%
&height
100% for a comprehensive list of available programs). The investigator
needs to apply the formula appropriate for their particular study design. There are a
minimum of three categories of data that need to be determined or estimated for
most sample size formulae - effect size (i.e., the biological effect that one desires
to detect, usually represented as probability), a measure of variation related to the
dependent variable, and alpha level.
The initial step in determining a necessary sample size is to use the appropriate
sample size equation. There are equations for nearly every use of sampling scheme to
estimate a population parameter or detect a difference. In addition, there are variations
for many equations depending on whether the estimate of variation of the dependent
variable is from a pilot study, literature, or known population value (very rare in
wetland field studies). Examples of situations where sample size calculations are
available include (1) estimation of a population mean, (2) estimation of a population
proportion, (3) testing of hypotheses concerning a population mean, (4) testing of
hypotheses concerning a population proportion, (5) testing mean differences between
two or more populations, (6) testing difference in proportions between two or more
populations, (7) testing main and interactive effects in traditional experimental
designs, and (8) conducting human dimension survey studies.
¼
Search WWH ::




Custom Search