ELSEVIER Biological Conservation 73 (1995) 93-100 © 1995 Elsevier Science Limited Printed in Great Britain. All rights reserved 0006-3207/95/$09.50+.00 0006-3207(95)00046-1 SENSITIVITY ANALYSIS FOR MODELS OF POPULATION VIABILITY Michael A. McCarthy* Forestry Section, University of Melbourne, Parkville, Victoria 3052, Australia Mark A. Burgman ,Forestry Section, University of Melbourne, Creswick, Victoria 3363, Australia & Scott Ferson Applied Biomathematics, 100 North Country Road, Setauket, NY 11733, USA (Received 4 November 1993; accepted 8 March 1994) Abstract A method of sensitivity analysis for population viability models is presented that uses logistic regression to evalu- ate the importance of model parameters that influence the risks of extinction. This approach is used to evaluate the importance of fecundity parameters and the initial number of non-breeding birds in a stochastic stage-struc- tured model of helmeted honeyeater Lichenostomus melanops cassidix population dynamics. The regression analysis indicates which model parameters have the greatest impact on the ,,isk of population decline. The results demonstrate that a simple expression containing the parameters of the model can encapsulate predictions of risk. This technique is proposed as an efficient alternative method of sensitivity analysis for population viability models. Of four fecundity parameters, the mean fecun- dity of intact pairs had the greatest influence on the risks faced by the helmeted honeyeater population. Mean fecundity of split pairs and the sex ratio of offspring were also important parameters. Over the range of parameters considered in this paper, environmental variation in fecundity and the initial number of non-breeding birds had little influence on the risks of decline. The importance of interactions between parameters was analysed. Keywords: sensitivity zLnalysis, population viability, extinction, logistic regression. INTRODUCTION Sensitivity analysis is an important component of mod- elling. It provides practical information for model builders and users by highlighting the parameters that have the greatest influence on the results of the model. Sensitivity analysis can highlight model parameters that ought to be most accurately measured so as to max- * Corresponding author. 93 imise the precision of the model, give a general indica- tion of the reliability of the model predictions, and highlight parameters and interactions that have the largest influence on the population to help determine effective management strategies. Models of natural populations may become complex, particularly when individuals are modelled, and under- standing the relative importance of different parameters and interactions between parameters may become com- putationally difficult. However, sensitivity analysis is employed only occasionally in population viability analysis (PVA), despite its benefits and the importance of using it when assessing management options (Poss- ingham et al., 1993). Usually, sensitivity analysis is measured by varying a parameter by a small amount around its estimated value. The resulting change in the state variable provides an index of sensitivity of the model to that parameter (e.g. Burgman et al., 1993). Applications of PVA to biological conservation would be improved with the use of sensitivity analysis. The sensitivity and elasticity of the deterministic growth rate of matrix models can be determined ana- lytically by eigen analysis (Caswell, 1978; de Kroon et al., 1986). However, extension of these techniques to models of population viability is not appropriate for at least three reasons: (1) population viability models tend to be nonlinear so obtaining analytical solutions is often complex if not impossible; (2) in PVA the result of in- terest is the risk of population decline rather than the deterministic growth rate (Burgman et al., 1993); and (3) these analytical methods cannot determine if there are any significant interactions between parameters. There is no single, universally accepted procedure for the sensitivity analysis of stochastic models, but Swartzman and Kaluzny (1987) recommend character- istics that should be evident in such a method. The method should be clearly defined, interactions between parameters should be distinguishable from single