Geoscience Reference
In-Depth Information
the more common habitat; this sampling could result in possible differences
between the average characteristics of the randomly chosen sites and the real
nests, suggesting that the species selected for nest sites in the less common
habitat type. This logic falters, though, for the reason highlighted in Mc-
Lellan's (1986) example of the buffet, in which a person chose equal portions
of beef and pork. In that example, the equal choice of the two types of food was
clear evidence of nonselection, or equal preference, regardless of availability.
Accordingly, in the hypothetical case of the 20 nests, one should question why,
if there was strong selection for one habitat type, did an equal number nest in
each type? The conclusion that this represents evidence of preference for the less
common type is ( just as in use-availability comparisons) based on the assump-
tion that if individuals had no preference, they would chose sites randomly,
which is typically thought to mean in proportion to availability. However, sup-
pose each individual, having no preference, just flipped a coin, so to speak, to
decide which of the two types of habitats to nest in each year. The result would
be as posed in the example, an equal number nesting in each type.
There are two other problems (really two aspects of the same problem)
with detecting selection using a site attribute design. James and McCulloch
(1990) showed that a species could be highly selective for a certain habitat
component, as indicated by a low variance among values of this component at
the sites it chose, but if the mean value for the component at selected sites
approximately equaled the mean in the available habitat (where the variance
was much higher than at selected sites), then selection might not be detected
(type II error). Conversely, Rexstad et al. (1988) tested multivariate statistical
procedures that are commonly used in site attribute studies and found that sig-
nificant results, which in a real study would be indicative of selection for cer-
tain habitat components, were often derived from a collection of meaningless
data (type I error). Taylor (1990) attributed Rexstad et al.'s spurious results to
ill-defined hypotheses and inappropriate use of statistical procedures; however,
Rexstad et al. (1990) showed that their study mimicked the majority of pub-
lished applications of these techniques. North and Reynolds (1996) expanded
on Rexstad et al.'s (1988, 1990) concerns that the statistical procedures used in
site attribute studies may yield misleading results because assumptions are
commonly violated.
INFERRING HABITAT QUALITY: FATAL FLAW #2
If habitat selection studies were simply an attempt to better understand behav-
ior and natural history, errors would merely be a setback in scientific inquiry.
However, these studies are typically designed to provide management guide-
