Study Design and Logistics - Wetland Techniques - page 29

Stratified random sampling can be used to increase sampling efficiency and
statistical estimation. The key for successful stratified sampling is that the basis for
stratification is correlated with the measured dependent variable. For example, if an
investigator is interested in the effect of watershed condition on water quality of a
wetland then sampling should be stratified using identified watershed conditions
(e.g., grassland, cultivation, forested) to ensure that each condition is properly
represented in the sample. By stratifying, one ensures that a single watershed
condition does not dominate the sample and consequently the final results.
Drawbacks to stratified sampling are (1) spatial and temporal scale of relevant
stratification variables can be difficult to determine; (2) increased complications for
analyses when homogeneous strata do not exist; and (3) sampling costs are
increased. Samples can be distributed among strata either by proportion of strata
size or an optimal allocation process. An example of this type of sampling would be
to stratify an area of coastal marsh by a salinity gradient (i.e., fresh, intermediate,
brackish, saline marsh), estimate the proportion of each strata (e.g., fresh ¼ 0.10,

intermediate 0.10), then determine sample

size within each strata by dividing the total number of samples to be taken
proportionally among the strata (e.g., if 250 total samples are needed to detect a
difference between treatment levels, then 25 would be taken in fresh and saline
marsh; 125 in brackish marsh; and 75 in intermediate marsh).
Strata can be defined within the study area (e.g., wetland and upland), study
period (e.g., seasons), and target population (e.g., small and large wetlands).
Strata cannot overlap, and elements cannot be available for selection in greater
than one stratum. For stratification to be useful, elements (experimental units)
should be more homogeneous within strata than among strata. If this is the case,
by stratifying, sampling standard error of the overall population mean should be
reduced to the standard error estimated by simple random sampling. Further,
estimates of dependent variables for each strata allows for comparisons among
strata, which are frequently of interest. However, it is critical to delineate strata
based on knowledge that the identified strata influence variables of interest. For
example, one would not test effects of herbicide treatments using strata of wetland
size, but rather stratification based on wetland hydrology, soil type, or vegetation
would be appropriate.
Systematic sampling represents an interesting approach that is rarely used in
wetland studies, but has a role in a variety of settings. Such a sampling approach is
possible when a population can be ranked in ascending or descending order of some
characteristic (e.g., wetland area, watershed area, salinity gradient). Here, one
would rank the population of interest relative to the characteristic and then sample
based on some rule (e.g., every 10th ranked object). In addition, systematic sam-
pling is often done on a spatial scale whereby a systematic grid of points or units is
established and those to be sampled are chosen by randomly selecting a starting
point and then establishing a rule to sample the remaining points or units in
reference to the starting point. For example, in a large coastal marsh where one is
interested in the distribution of a contaminant, use of an appropriately sized grid
overlaid on a map of the marsh provides unique sampling units. Upon randomly

¼ 0.30, brackish ¼ 0.50, and saline ¼