Environmental Sampling Design - Fundamentals of Environmental Sampling and Analysis - page 54

Simple random sampling is not applicable for heterogeneous populations. The
higher the degree of the heterogeneity, the lesser the random sampling approach
will adequately characterize true conditions. Furthermore, while the complete
randomization approach defines the statistical uncertainty, it could also lead to
several problems. First, sampling points by random chance may not be uniformly
dispersed in space and/or time particularly when sample size is small. This drawback
can be overcome by systematic sampling. Second, simple random sampling, by
ignoring prior site information or professional knowledge, often leads to more
samples, which is not as cost-effective as other sampling designs. Third, randomly
selected sampling points could be harder to locate precisely. Because of this
limitation in its implementation, simple random sampling is seldom recommended
for use in practice except for relatively uniform populations. The U.S. EPA guideline
does not recommend simple random sampling in flowing water bodies and states that
it is only practicable for sediment bed sampling in nonflowing (static) water bodies
(EPA, 1995).

3.3.3 Stratified Random Sampling

As shown in Figure 3.5b, stratified random sampling divides sampling population
into several nonoverlapping (mutually exclusive) strata and within each stratum a
simple random sampling is employed. Each stratum is relatively more homogeneous
than the population as a whole. The selection of strata, however, requires some prior
knowledge of the population to be sampled.
''Strata'' could be ''temporal'' or ''spatial.'' Temporal strata permit different
samples to be selected for specified time periods, for example, day vs. evening,
weekdays vs. weekend, four seasons of the year, and so forth. Spatial strata are more
common and come with many varieties. Spatial strata can be based on sampling
depth (stratified lakes, soil, or sediment cores), ages and sex of population (men,
women and children), topography, geographical regions, land types and uses, zones
of contamination, wind direction (downwind vs. upwind), political boundaries, and
so forth.
A major advantage of stratified random sampling is that the sample size can be
adjusted depending on the variations or the cost of sampling in various strata. Strata
expected to be more variable or less expensive should be sampled more intensively.
This provides greater precision and cost-saving than simple random sampling. It also
implies that stratified random sampling results in smaller standard deviation than the
simple random sampling, particularly if the strata are quite different from one
another. Another advantage is the additional information it provides regarding the
mean and standard deviation within each stratum, which may be of interest for a
particular project.
Within each stratum, the formulas used to calculate mean and standard deviation
are the same as those for simple random sampli ng (Eqs 2.12 and 3.1). Suppose we
have r strata (k ¼ 1, 2, r), the stratum mean (x k ) an d stratum standard deviation

(s k ) are then combined to estimate the population mean (x) and standard deviation (s)

...