1
Statistical consideration for strategic sampling – Part III
Following the discussions on random sampling methods after simple random
sampling and stratified random sampling, we will now look into the other
techniques.
3. Systematic sampling
As the term ‘systematic’ dictates, we have to select the first test sample at
random and then pick up the subsequent samples at a fixed interval, such as
every fifth item or at a fixed time during the streaming of items on a
conveyor belt upon production.
For example, consider 1000 sequentially numbered items in a batch of a
product of which 20 need to be selected for testing. What we have to do is
to select the first at random from items marked 1-50, which every 50th item
is chosen. If item number 9 was selected at random as the first test sample,
the subsequent samples chosen would be 59, 109, 159, …, 959.
The main advantage for systematic sampling is due to its simplicity to
implement over a period of time or a stream of material. In this type of
sampling, the sample values are spread more evenly across the population,
thus, many systematic samples are highly representative of the population
from which they were selected. It also reduces the occurrence of long or
very short intervals that may occur in simple random sampling.
However, systematic sampling can only be considered equivalent to random
sampling if the variation in the material is reasonably independent of the
item numbering or location. An extreme case can be that all items selected at
intervals turn out to be of good quality whilst there might be bad ones in
between! Thus, the precision of the test result obtained is mainly influenced
by the ‘gap’ between the sampling points, and no valid estimate of sampling
error can be calculated from a single sample taken.
Systematic samples are treated as if random, so the statistical treatment is as
for simple random sampling as discussed in the Part I article previously.
4. Cluster sampling
Cluster sampling can be used to overcome time and cost constraints when a
relatively small number of samples are required from a large and widely
dispersed population which would be costly to examine in full using simple
random sampling.
