Table 5.

CVA results using variable number of semi-landmark points used.

				Cross-validation assignment rate (%)
Data acquisition	Data processing	# of points used	# of PC axes	Observed	95% confidence interval
Curve tracing	Bending energy	20	18	67.4	67.4 – 93.5
Curve tracing	Bending energy	30	18	78.3	67.4 – 93.5
Curve tracing	Bending energy	41	10	84.8	73.9 – 95.7
Curve tracing	Bending energy	60	10	84.8	69.6 – 95.7
Curve tracing	Bending energy	82	9	87.0	69.6 – 95.7
Curve tracing	Bending energy	120	9	87.0	67.4 – 95.7
Curve tracing	Perpendicular projection	20	18	84.8	76.1 – 97.8
Curve tracing	Perpendicular projection	30	10	82.6	76.1 – 95.7
Curve tracing	Perpendicular projection	41	11	84.8	71.7 – 95.7
Curve tracing	Perpendicular projection	60	13	87.0	73.9 – 97.8
Curve tracing	Perpendicular projection	82	13	84.8	76.1 – 97.8
Curve tracing	Perpendicular projection	120	12	84.8	76.1 – 95.7
Curve tracing	Elliptical Fourier analysis	41	10	84.8	65.2 – 93.5
Curve tracing	Elliptical Fourier analysis	82	12	73.9	63.0 – 93.5
Fan	Bending energy	41	6	89.1	73.9 – 95.7
Fan	Bending energy	82	7	89.1	76.1 – 95.7
Fan	Perpendicular projection	41	9	87.0	73.9 – 97.8
Fan	Perpendicular projection	82	7	89.1	78.3 – 97.8

The rate of correct cross validation assignment based on the canonical variates analysis (CVA) was not highly dependent on the number of points used to represent the curve. The number of principal component (PC) axes used to optimize the cross-validation assignment rate varied with the data acquisition and processing methods and the number of points on the outline.