|
|
Upon thoughtful consideration of the sample survey approach, several questions
may come to mind. This section answers several commonly asked questions
concerning survey sampling. These questions are addressed in fairly general
terms. As noted in the introduction, additional technical details are in a
series of methods manuals. (Back to Frequently Asked
Questions
)
Survey Sampling
Why is it so important to select sites randomly?
The way we select the sample (i.e., choose the units from which to collect data)
is crucial for obtaining accurate estimates of population parameters. We
clearly would not get a good estimate of the percentage of all students at a
university who participate in intramural sports if we polled students at the
entrance to the gymnasium. This preferential sample would, most likely, include
a much higher proportion of athletes than the general population of students.
Similarly in a stream study, preferential sampling occurs if the sample includes
only sites downstream of sewage outfalls in a watershed where sewage outfalls
affect only a small percentage of total stream length. This kind of sampling
program may provide useful information about conditions downstream of sewage
outfalls, but it will not produce estimates that accurately represent the
condition of the whole watershed. An illustrative example depicting the
difference between assessments that rely on preferential (non-probability)
sampling and probability sampling is shown for
streams.
Preferential selection can be avoided by taking random samples. Simple random
sampling ensures that no particular portion of the sampling frame (i.e., groups
of students or kinds of river reaches) is favored. Within streams, the chance
of selecting a sampling unit that has degraded ecological conditions would be
proportional to the number of sampling units within the target population that
have degraded conditions. For example, if 30% of the target population has
degraded conditions, then on average 30% of the (randomly selected) units in
the sample will exhibit degraded conditions. This property of random sampling
allows estimates (based only on the sample) to be used to draw conclusions
about the target population as a whole. Go to Top
What if more than one target population or
sub-populations are of interest? Can a survey design address
this?
One goal of a sample survey may be to compare two parts of a target population
(these could be described as sub-populations), or a part of the population to
the whole population. For instance, an opinion poll might be used to determine
if a higher percentage of the people living in Rhode Island is likely to vote
Democratic than in the nation as a whole. Given its small size, Rhode Island
probably would receive very little attention in a national poll if samples are
allocated by a simple random process. One way to achieve a sample of people in
Rhode Island that is sufficient to make this comparison is to increase sampling
effort for the nation as a whole until enough people from Rhode Island are
included in the randomly selected national sample. This option is not very
cost-effective because it requires considerable, unnecessary sampling effort in
other areas to achieve a desired sample size in one small area.
Another, preferable, alternative would be to divide the entire target population
into two subpopulations,. Voters in the United States could be divided into (1)
those living in Rhode Island, and (2) those living elsewhere. A simple random
sample of desired size could then be selected from each of these groups.
Stratified sampling could be used in a stream survey to enhance sampling effort
in a watershed of special interest so that its condition could be compared with
that of a larger area. In a study area with 1000 kilometers of streams, for
example, an area of special interest may contain 200 kilometers of streams. If
budget constraints limit the size of the total sample to 60 sampling units, 30
could be randomly selected from the special interest area, and 30 from the rest
of the sampling frame. If simple random sampling is used, the area of special
interest, which represents 20% of the area, will contain only about 12 of the
60 selected sampling units. A sample of 12 would be insufficient to estimate
the condition of the special interest area reliably. Go
to Top
Many States are moving toward a rotating basin design
for conducting their waterbody assessments and water resource planning.
How can sample surveys be incorporated into this approach?
One aspect of a rotating basin design is to describe the condition of the basin
as a whole. Unless the basin will be censused, a sample survey can be used to
characterize the overall status of the basin. A combination of a sample survey
and targeted monitoring as outlined above can be used to produce an overall
description of the basin. In addition, the rotating basin design could be
embedded in a state wide sample survey by intensifying the sampling on the
particular basin or set of basins for the year(s) that basin or set of basins
was under study. A routine ongoing statewide survey could be conducted at some
baseline level of intensity, along with intensified sampling in the targeted
basins. In this way, a state could track statewide progress as well as progress
in individual basins as they are revisited, and determine the condition of the
basin relative to the condition of the state as a whole. Go
to Top
States are required to identify or list waterbodies that
are impaired, for example, by the 303(d) listing process. How useful are
sample surveys for identifying those waterbodies?
Because survey sampling is intended to characterize the status of the resource
as a whole, it is generally not useful for enumerating a list of a specific
type, for example, a list of all the impaired waterbodies. However, an
important aspect of sample surveys is that they can provide a check on the
completeness of the list. Suppose that a state agency had submitted a list of
impaired waterbodies it had gathered from ongoing monitoring programs, and from
reports from other agencies. This list would presumably be a census of the
impaired waterbodies. The state agency could then compare the amount of the
resource impaired against the magnitude of the total resource to derive a
percent of the resource impaired. The state agency could check this proportion
by conducting a sample survey and classifying the sample sites into
impaired/not impaired. It could then check the proportion impaired based on the
sample survey with the proportion calculated from the inventory. Consistency
between the two estimates would indicate that the census of the impaired sites
is reasonably complete (within the uncertainty of the sample survey);
inconsistency between the two estimates would provide information on how good
the impairment census was (again within the confidence limits of the sample
survey). Go to Top