what we know about the ecology of birds, it will often make sense to stratify
our sample by obvious factors, such as habitat and altitude. Where surveys
rely on local observers, it might also make sense to stratify by their availability.
Stratification by observer density might seem odd at first sight, but it provides an
efficient way of maximizing the use of skilled volunteers when their distribution
is uneven, as it often is. Stratification is strongly recommended because it can
improve both precision and accuracy, and it ensures proper habitat coverage.
Thankfully, there are simple rules that help us choose the most appropriate
strata—and it turns out that, even when our prior assumptions about strata prove
to be wrong, there is no detrimental effect.
In those situations where we have little information about the habitats used by
a species, it makes sense to sample in proportion to the area of the different
habitats. For example, if 80% of the area is forest and 20% farmed land, then
80% and 20% of our samples should be in forest and farms, respectively. When
we know more about species density in different habitats there are some simple
rules designed to improve precision. For example, Sutherland (2000) suggests
that sampling should be proportional to the likely proportion of the species
occurring in a habitat—so if preliminary information suggests 60% of a
Sampling strategies| 31(a) (b)
Habitat 1Habitat 2Fig. 2.7Next, imagine our study area comprises two distinct habitats of roughly
equal area, within which our chosen study species lives but at quite different
densities. (a) A random sample across the whole area is quite likely to result in an
uneven split of survey squares between the two habitats. If 70% of the squares happen
to fall in one habitat then the population estimate for the whole area based on the
10 squares would inevitably be dominated, or biased, by that habitat. (b) The solution
to this problem is to stratify so, for example, half the samples fall in each habitat—the
data are then analyzed by strata and the results combined to give an unbiased estimate
of population size (see text for further information on sampling within strata).