2.2 Rule 2: use only relevant sites for a species

Some species at not present at every site. When a species is absent at a site, the counts are always zero. The numbers are not changing at that site, so locally there is no trend. If we would take such site into account, the overall trend will be biased towards zero.

Species that are rarely seen at a site pose a problem too. Imaging a species is only seen at the first winter with on average 10 individuals. The model tries to fit the first winter at this site as \(\log(10) = 2.303\). All other winters at this site should be \(\log(0) = -\infty\). This results is a strong negative local trend, which again biases the global trend.

Therefore, we require that a species is observed at a site during at least 3 different years before taking that site into account for this species. Table 2.2 is a very large table because it contains a row for each of the 107 species observed in the data. 40 species at so rare that there is not a single site at which they are observed during 3 different years.

Table 2.2: number of sites at which a species was observed split by the number of winters (columns) during which the species was observed at a site.