Biomedical Engineering Reference
In-Depth Information
ment data after making major revisions to any set of observations, devel-
oping measurement procedures is an iterative and time-consuming process.
(The illustrative examples presented here employed unrealistically few
objects and observations, and as such, they may have underrepresented the
work required to conduct a rigorous measurement study.) As a rule, an
investigator should borrow from other investigators and other studies
whenever possible, particularly if data exist to suggest that the scales have
good measurement properties; that is, they have been demonstrated to be
reliable and valid when used with objects (people or resources) similar to
those proposed for the investigator's own study. The Web site for this
volume (http://springeronline.com/0-387-25899-2) will provide links to com-
pendia of validated measurement instruments that address attributes of
interest in biomedical informatics.
In addition to computation of part-whole correlations, the data resulting
from measurement studies may be analyzed using one of the many statis-
tical techniques for grouping of observations. The most popular is
exploratory factor analysis and its many close relatives.
2,3
These techniques
suggest which of the observations are well behaved, in a way that can be
more precise and informative than inspecting part-whole correlations or
values of correlations between observations. The mechanics of these
methods are beyond the scope of this discussion. The reader is advised to
consult a local statistician or psychometrician, or to read one of the topics,
cited above, that addresses these techniques.
Self Test 6.2
1. Consider the following measurement result, with a reliability of 0.61.
What is your diagnosis of this result? What would you do to improve
it?
Results of six observations
Object
1
2
3
4
5
6
A
4
3
5
2
1
4
B
2
4
5
3
2
2
C
3
4
3
4
4
3
D
2
3
1
2
1
2
E
3
3
2
2
4
3
Part-whole correlation
0.49
0.37
0.32
0.37
0.32
0.49
2. Consider the following measurement result, for which the reliability is
0.72.
Search WWH ::
Custom Search