UNDERSTANDING AND PREDICTING THE EFFECTS OF SPARSE DATA ON DEMOGRAPHIC ANALYSES

Abstract

Demographic models are an increasingly important tool in population biology. However, these models, especially stochastic matrix models, are based upon a multitude of parameters that must usually be estimated with only a few years of data and limited sample sizes within each year, calling into question the accuracy of the results of these models. We first discuss how these data limitations create sampling uncertainty and bias in the estimated parameters for a stochastic demography model. Next, we ask whether limited data can favor the construction of deterministic models that ignore variation and correlation of rates. With less than five years of data, the mean squared error of deterministic models will sometimes be smaller than that of stochastic models, favoring the use of simple models, even when their predictions are known to be biased. Finally, we introduce a procedure to estimate the sampling variation around population growth rate estimates made from demographic models that are based on specified sampling durations and intensities.