Microsoft Office Tutorials and References
In Depth Information
Now, let us consider in greater detail the use of experimental designs in studies
that are experimental and not observational . As I have stated, it is impossible to
consider all the possible designs, but there are three important designs worth con-
sidering due to their frequent use. Below I provide a brief description that explains
the major features of the three experimental designs:
Completely Randomized Design : This experimental design is structured in a
manner such that the treatments that are allocated to the experimental units
(subjects or observations) are assigned completely at random. For example, con-
sider 20 analysts (our experimental unit) from a population. The analysts will use
4 software products (treatments) for accomplishing a specific technical task. A
response measure, the time necessary to complete the task, will be recorded. Each
analyst is assigned a unique number from 1 to 20. The 20 numbers are written
on 20 identical pieces of paper and placed into a container marked subject . These
numbers will be used to allocate analysts to the various software products. Next,
a number from 1 to 4, representing the 4 products, is written on 4 pieces of paper
and repeated 5 times, resulting in 4 pieces of paper with the number 1, 4 pieces
with the number 2, etc. These 20 pieces of paper are placed in a container marked
treatment . Finally, we devise a process where we pick a single number out of each
container and record the number of the analyst (subject or experimental unit) and
the number of the software product (treatment) they will use. Thus, a couplet of an
analyst and software treatment is recorded; for example, we might find that ana-
lyst 14 and software product 3 form a couplet. After the selection of each couplet,
discard the selected pieces of paper (do not return to the containers) and repeat
the process until all pieces of paper are discarded. The result is a completely ran-
domized experimental design. The analysts are randomly assigned to a randomly
selected software product, thus the description—completely randomized design.
Randomized Complete Block Design : This design is one in which the exper-
imental subjects are grouped (blocked) according to some variable which the
experimenter wants to control. The variable could be intelligence, ethnicity, gen-
der, or any other characteristic deemed important. The subjects are put into
groups (blocks), with the same number of subjects in a group as the number of
treatments. Thus, if there are 4 treatments, then there will be 4 subjects in a block.
Next, the constituents of each block are then randomly assigned to different
treatment groups, one subject per treatment. For example, consider 20 randomly
selected analysts that have a recorded historical average time for completing a
software task. We decide to organize the analysts into blocks according to their
historical average times. The 4 lowest task averages are selected and placed into a
block, the next 4 lowest task averages are selected to form the next block, and the
process continues until 5 blocks are formed. Four pieces of paper with a unique
number (1, 2, 3, or 4) written on them are placed in a container. Each member of a
block randomly selects a single number from the container and discards the num-
ber. This number represents the treatment (software product) that the analyst will
receive. Note that the procedure accounts for the possible individual differences
Search JabSto ::




Custom Search