Morphometric Data - An Invariant Approach to Statistical Analysis of Shapes - page 38

2.4.1 Measurement error studies
This section reports the measurement error associated with different
data collection procedures by describing the results of studies conduct-
ed in our laboratory (Corner and Lele et al., 1992; Richtsmeier and
Paik et al., 1995). By presenting two studies, we are emphasizing the
fact that different data collection devices may require varying types of
measurement error studies. The first study reports the measurement
error associated with the 3Space digitizer (see Section 2.3.1 ) . The sec-
ond study discusses the measurement error associated with the
landmark coordinates obtained from CT slice scan images (see Section
2.3.4 ).
a) Measurement error for collecting landmark coordinate data
using the Polhemus 3Space digitizer
Corner and colleagues (1992) studied the precision in landmark coor-
dinate data collected by the Polhemus 3Space digitizer using the
following data-gathering experiment (see Hildebolt and Vannier, 1988
for another measurement error study of this device). Two experienced
observers collected coordinate data from 11 landmarks located on a
single skull of a subadult male crab-eating macaque, Macaca fascicu-
laris . The skull was secured to the digitizer table and digitized twenty
times by each observer. Data were collected twice daily with approxi-
mately six hours separating the daily digitizations, until each
observer's sample of 20 digitizations was complete. The time gap was
introduced in order to ensure that the digitizations were independent
of each other, reducing any “memory effect.” Most importantly, the ori-
entation and location of the specimen was fixed. Lack of movement of
the object between digitizations is essential, as any movement of the
object introduces random rotation and translation into measurement
error calculations, making some of the variance parameters nonidenti-
fiable (see Chapter 3 for the details on this issue). Having controlled
for orientation and translation in this way, we can estimate the mea-
surement error around each landmark. Measurement error is given by
the sample variance for each landmark obtained from the 20 digitiza-
tions calculated using the coordinate locations for each axis. Since the
data collectors were experienced, it is assumed that measurement
error does not depend strongly on the data collector but depends most-
ly on the type of landmark and the measuring device.
Table 2.1 (Corner and Lele et al., 1992) provides the measurement