analysis (Aebischer et al. 1993) or log-linear modeling (Heisey 1985) modified to
include a randomization test (Green et al. 2000). Both methods analyze all the data
at once and yield a ranking of relative density of use. Log-linear modeling gives
values for relative density that are appropriately weighted for sample size. Both
methods also have the advantage that they regard the data for each study area, sur-
vey data, or individual bird (for tracking data) as statistically independent. This is
desirable because it is clearly unsatisfactory to regard multiple records of the same
animal on the same habitat patch as being independent. This defect was present in
some earlier widely used methods for testing the significance of habitat selection
such as that of Neu et al. (1974). Manly et al. (1993) provide a useful account of the
problems of measuring and testing for selection.
11.5.2 Relating numbers or densities of individuals or records of
tracked birds in spatial units to the habitat composition
of those units
If the data are counts of birds or records of tracked birds in areas such as transect
sections, circles around point counts, grid squares, fields, or woods that each
contains several habitats, then it may not be possible to attribute all the bird
records to a particular habitat. This might be because birds were detected by their
calls and not seen. A suitable analytical approach is then to carry out multiple
regression with the density of birds or records as the dependent variable and the
proportions of habitat types in each spatial unit as the independent variables.
Preferred habitats will tend to have statistically significant positive regression
coefficients. If the size of these sampled areas varies then this can be taken into
account in the analysis, for example, by converting the counts to densities (numbers
per unit area).
11.5.3 Comparison of habitat at places used by birds with that at
places that are representative of the study area or known
to be unused
This approach is useful when it is not feasible to map and measure habitats over
large areas. Instead habitat is recorded at a sample of small sites chosen at random
or on a regular grid to be representative of a much larger area available to the
birds. Habitat is also recorded at places where birds are seen. Data of this type
can be analyzed by multiple logistic regression with used or representative places
being scored as a binary dependant variable (1 or 0) and the habitat measures as
independent variables (Manly et al. 1993). As with the other analyses of selection,
it is important to think about the availability of the random or representative
places to the birds. For example, in the case of selection of foraging habitat by
262 |Habitat assessment