Use of state-space population dynamics models in hypothesis testing: advantages over simple log-linear regressions for modeling survival, illustrated with application to longfin smelt (Spirinchus thaleichthys)

Factors impacting the survival of individuals between two life stages have traditionally been evaluated using log-linear regression of the ratio of abundance estimates for the two stages. These analyses require simplifying assumptions that may impact the results of hypothesis tests and subsequent conclusions about the factors impacting survival. Modern statistical methods can reduce the dependence of analyses on these simplifying assumptions. State-space models and the related concept of random effects allow the modeling of both process and observation error. Nonlinear models and associated estimation techniques allow for flexibility in the system model, including density dependence, and in error structure. Population dynamics models link information from one stage to the next and over multiple time periods and automatically accommodate missing observations. We investigate the impact of observation error, density dependence, population dynamics, and data for multiple stages on hypothesis testing using data for longfin smelt in the San Francisco Bay-Delta.