Sources of extra zeros
Summarize Martin et al. (2005), and update with more recent thinking. Provide some veterinary examples?
We expect to measure zero in ecology. Whether that zero is in the context of a presence/absence measurement, or a count distribution, zeros are a natural and expected part of ecology. However, zeros sometimes form the majority of the observations, and we start to encounter problems building predictive models on that basis. Zeros arise from a number of mechanisms that vary according the scale (extent and resolution) of a study. So called ‘zero-inflated’ data can arise from the true process being studied. But it can also arise from the process used to collect the data, the observation process.
Martin et al (2005) categorized excess zeros into four sources, depending on whether they are true zeros (part of the biological process), or false zeros (arising from the observation process).
| Type of Zero | Definition |
| True Zero | Species does not occur at a site because of the ecological process, or effect, under study. |
| Species does not saturate it's entire suitable habitat, by chance. | |
| False Zero | Species occurs at a site, but is not present during the survey period. |
| Species occurs at a site and is present, but is cryptic and difficult to detect. |
True zeros [Table 1] arise when the species is absent from the unit of observation for ecological or epidemilogical reasons. When sampling a patch of habitat containing a population of hosts, a true zero means the prevalence is zero. This could arise because the habitat patch is empty of potential hosts, or the hosts present are unsuitable, or the parasite may not have dispersed to the patch yet.
False zeros arise from characteristics of the observation process. At ecological scales, one type of zero arises when the unit of observation is smaller than the home range of the species in question. The site or patch is used by the species, but at the particular time of the survey individuals of that species are in a different part of their home range, and impossible to detect. This type of false zero is not an issue for parasites, but it is a possible problem when surveying hosts. A second problem arises when the species is present during the survey, but goes undetected. For epidemiological surveys this arises if the method used to detect the parasite (or past infection of the parasite), has a non-zero probability of detecting the parasite in a particular sample. A method with low sensitivity has a high probability of a false negative. In some cases we have an estimate of the sensitivity of a method, but this is (to my knowledge) rarely used directly in species distribution modelling.
What are the questions and what kind of data are collected? Pooled sampling of mosquitos - minimum prevalence What drives detectability in epidemilogical samples and are they different ## Scale of questions
There are two important spatial scales to consider. First, what is the resolution of the ecological / epidemiological process? Is the ecology about a population within a discrete patch of habitat, or about the larger metapopulation operating across many patches of habitat.
infographic / toy examples