Table 3.

Bootstrap estimates of the cross-validation rates of CVA assignments : Variable number of PC axes method

			Cross-validation assignment rate (%)
Data acquisition	Data processing	# of PC axes	Observed	95% confidence interval (derived from bootstrap)
Curve tracing	Bending energy	9	87.0	69.6 – 95.7
Fan	Bending energy	7	89.1	76.1 – 95.7
Curve tracing	Perpendicular projection	13	84.8	76.1 – 97.8
Fan	Perpendicular projection	7	89.1	78.3 – 97.8
Curve tracing	Elliptical Fourier analysis	12	73.9	63.0 – 93.5
Curve tracing	Eigenshape analysis	24	69.6	67.4 – 95.7

Each method of outline processing shown used 82 points around the periphery of the feather. Rates of cross-validation assignment based on canonical variates analysis (CVA) were similar for all methods, given the overlapping 95% confidence intervals. The number of principal component (PC) axes used to optimize the cross-validation assignment rate varied slightly over the different methods.