The importance of noise in fish population models

Abstract

This paper describes simple, realistic circumstances in which the noise component of a population model dramatically influences conclusions from the analysis. The first example, restricted to five data values, illustrates geometrically why different statistical assumptions can lead to fundamentally different results. Such diverse conclusions can also be a problem in realistic data analysis, as illustrated by a second example involving catch per unit effort (CPUE) data from a lingcod ( Ophiodon elongatus) stock. Using methods adapted from Kalman filter theory, I show that a high or low estimate of survival depends on whether or not the model allows for measurement error. Furthermore, additional data suggest that the more complex model with measurement error is more realistic. I conclude that earlier work on deterministic models may be somewhat misdirected, because statistics is treated almost as an afterthought.