Statistics that count in the animal kingdom
A team of British statisticians is pioneering new ways of analysing information about how animals survive in the wild. The new techniques will give biologists a better understanding of the viability of populations of particular species. It will also help them to devise conservation programmes.
At the University of Kent at Canterbury, Professor Byron Morgan is developing statistical tools to improve the interpretation of the data on animal survival. Researchers can gather vast amounts of data over many years. But it can be difficult to make sense of the information and to relate it to what is actually happening to the population.
Data on the survival rates of animal populations not only yield information about the species itself but can also be an important indicator of environmental change. One way of obtaining this sort of information is to capture some of the population, mark them by placing a ring on the leg, for example, and then release them. The researcher then records sightings of marked animals along with details of the bodies of dead animals they find.
Professor Morgan’s team has devised a way of combining data from the recovery of dead animals with that from observations of surviving animals. This is significant because until now researchers have usually analysed the two types of information have independently. By integrating the data, experts in animal populations can make much more realistic conclusions.
The Canterbury team has also written software to help to choose the best statistical model for the type of data at hand. Data gathered from the field can be highly complex. Many factors can affect survival rates. With the new software, the operator feeds in the initial data, the software then prompts the researchers to indicate the most important parameters.
Using the data and the information from the operator, the system can help to select the best model population. Furthermore, the Kent team has developed a way of verifying if a particular statistical model is suitable for the kind of data that it is being presented with. The system allows the user to ‘interrogate’ any model to see if it is capable of analysing the data meaningfully.