Confidence intervals for demographic projections based on products of random matrices (Q1070963)

The theory of demographic projections in age-structured populations with randomly fluctuating vital rates has progressed considerably during the last decade, mainly through the work of \textit{J. E. Cohen} [see e.g. Bull. Am. math. Soc., new Ser. 1, 275-295 (1979; Zbl 0401.60065)] and \textit{S. D. Tuljapurkar} [see e.g. J. Math. Biol. 19, 335-350 (1984; Zbl 0552.92014) and ibid. 14, 221-230 (1982; Zbl 0494.92015)]. It uses the products of stationary ergodic sequences of random matrices. This paper takes another important step. It obtains strong laws of large numbers and central limit results for logarithms of age-class sizes or linear combinations of them, namely the a.s. convergence to a constant growth rate and the asymptotic normality of the growth rate. However, these results are obtained under more general conditions than previously. The generalization is not trifling, even from the point of view of applications. Large-sample asymptotic estimators of the parameters are derived and, from these, confidence intervals on population size and growth rates are, for the first time, obtained. The finite-sample estimators are studied numerically. An application to a population of striped bass in the Potomac river illustrates the results as well as the hugeness of the confidence invervals of prediction when the available data is scarce, which is usually the case.