A Simulation Study of Animal Population Estimation Using the Capture-Recapture Method

Abstract

Capture-recapture methods are widely used to gather data for the estimation of various parameters of animal populations, where these data consist of a record of the captures and recaptures of all the individual animals seen in samples taken over a period of time. Unfortunately, problems are not over when the data have been obtained since a suitable method of analysis must then be used. The choice of a method is often not a simple matter since there are a number of alternatives, requiring different assumptions and giving different results. The choice is relatively easy if it is possible to decide precisely what assumptions can be made with regard to the sampling situation and behaviour of the animals (see, for example, Southwood 1966; Parr, Gaskell & George 1968 or, for a more detailed discussion, Cormack 1968), but this is often not the case. Difficulties arise when it is not known whether an assumption holds, or when an assumption is known to almost hold. It is hoped that the present paper will help the experimenter to choose between three particular methods of analysis, each of which can be used when the population under study is changing due to ingress ('births') and egress ('deaths'). We consider a form of the capture-recapture experiment where five samples are taken, with an equal interval of time between each, but it is felt that the relevance of the results is not restricted to this particular situation. The methods of analysis that are considered are given by Fisher & Ford (1947), Jolly (1965) and Manly & Parr (1968). If we consider the assumptions: (a) that sampling is random; (b) that survival rates and probabilities of capture are unaffected by the marking of animals; (c) that survival rates are independent of the age of animals; and (d) that survival rates are approximately constant during the capture-recapture experiment, then all of these assumptions are required for the method of Fisher & Ford, the first three for the method of Jolly and the first two for the method of Manly & Parr. The results reported here illustrate the behaviour of the three methods when all four assumptions hold, and also their behaviour when only the first two assumptions hold. The technique of computer simulation was used to obtain the results. A considerable number of artificial animal populations were generated on the K.D.F.9 computer at the University of Salford and random samples taken from these so that population parameters could be estimated. In this way the three methods of data analysis have been examined empirically over a wide range of conditions, with true parameter values being known exactly. This technique would seem to be the only way to perform such an extensive examination of the behaviour of the methods.