Professional Documents
Culture Documents
Classification Bias in Discriminant Function Analyses Used To Evaluate Putatively Different Taxa
Classification Bias in Discriminant Function Analyses Used To Evaluate Putatively Different Taxa
245
246 JOURNAL OF MAMMALOGY Vol. 81, No. 1
inant function, it is likely to be an inferior males, 79 females) and 157 were C. l. thamnos
discriminator relative to a function derived (82 males, 75 females). Use of those observa-
from the entire data set (Johnson and Wich- tions allowed us to demonstrate with biological-
ern 1992). Another drawback to sample ly relevant data how bias in resubstitution clas-
sification can affect evaluation of morphological
splitting is that a relatively large number of
differentiation between putative taxa. Because
samples is required for this technique to be coyotes are sexually dimorphic (Lydeard and
informative. Kennedy 1988), males and females were ana-
Jackknife sampling, also known as Lach- lyzed separately.
enbruch’s holdout or the leave-one-out To contrast different results obtained from re-
Jackknifing reduces bias in classification er- data. Therefore, when reporting classifica-
ror rates; however, the level of classification tion results from DFA, the method of clas-
error that is sufficiently low enough to con- sification should be fully described. In most
sider putative taxa as distinct is arbitrary. cases, jackknife classification should be
Alternative approaches, such as identifying preferred over resubstitution.
phylogenetic subdivisions based on concor-
dant patterns in multiple characters (Ball ACKNOWLEDGMENTS
and Avise 1992), may provide a less arbi-
We thank C. Lydeard for providing the data
trary basis for assigning subspecific taxon-
used in this analysis. J. Abdalla, J. Akin, J. Ferr-
omy.
ence, C. Lance, D. Rogowski, C. Spencer, and J.
Although we focused on how resubsti- Waits provided many helpful suggestions during
tution-based DFA may overestimate levels preparation of the manuscript. R. Lance was
of differentiation between putative taxa, funded by Board of Regents Louisiana Enhance-
this problem is not restricted to taxonomic ment of Quality Science Fund Fellowship (1994–
analyses. If we had used ecological or sim- 1999)-GF-29, awarded through R. G. Jaeger.
ulated data for this analysis, we would have Analysis and manuscript preparation were sup-
found the same biases resulting from resub- ported by National Science Foundation grants
stitution that we found with morphological DEB-AC09-76SR00-819 and DEB-9123943.
February 2000 LANCE ET AL.—DISCRIMINANT FUNCTION BIAS 249