Morphometric Data - An Invariant Approach to Statistical Analysis of Shapes - page 37

in locating landmarks; and 2) instrument error in identifying land-
mark coordinates. Only specific research designs can partition
measurement variability into these two components.
The study of repeatability, in contrast to precision, involves repeat-
ed measures taken on several specimens. Variability among repeated
measures on more than one specimen includes variability due to the
instrument, due to the observer, and due to biological differences
among specimens. Repeatability is the precision of a particular mea-
sure relative to the biological differences among specimens, and it is
measured as the ratio of the precision of a particular measure to the
biological differences among specimens (Kohn and Cheverud, 1992).
This section deals only with the study of precision of various meth-
ods of data collection. The purpose is to isolate the effects due to the
observer and due to the data collection instrument and not to confound
it with the inherent biological variability among specimens. Biological
variability is discussed in the next chapter.
Consider a single specimen fixed on a data collection device from
which we are measuring landmarks. Suppose a single observer collects
data from this specimen several times using a single instrument over
a period of several days. In an ideal world, all coordinate measures
would be identical to each other because the specimen is immobilized,
but they usually are not. The variability in the measurements is due in
part to instrument error and in part to error by the observer. Error
studies should look for systematic differences in precision along the
major axes.
Now suppose there is another observer who is also taking mea-
surements on the same immobilized specimen using the same device.
Though these measurements should be identical to the measurements
made by the first observer, they rarely are. Differences in the two data
sets stem from several sources. The second observer might be more or
less experienced with the digitizer, or (s)he may know more or less
about the specimen and the landmarks being collected. The differences
between the measurements made by two or more observers of the same
specimen constitute inter-observer error. Precision can be separately
calculated for each investigator to determine the contribution of inter-
observer error to the study. Excessive, inter-observer error can be
avoided by limiting data collection to a single observer.