“In what state is my state-space? An evaluation of model selection and diagnostic tools for state-space stock assessment models.”
Saltwater Inc. is seeking a postdoctoral scientist to develop diagnostic tools for evaluating state-space stock assessment models. Although state-space assessment models have become increasingly popular in recent years, there are open questions regarding whether traditionally-used stock assessment model diagnostics (such as retrospective error, model residuals, etc.) are reliable indicators of model performance in state-space models. The research project will use an established operating model to generate simulated stock assessment datasets and test a variety of commonly-used and newly-developed diagnostic metrics. The goal will be to determine which diagnostics perform best at identifying the model with the lowest estimation error. The postdoctoral scientist will have the opportunity to develop their own diagnostic tests and apply the methods to real datasets, which will inform the 2023 State-space Assessment Methods Research Track. (View NOAA’s Research Track page.)
State-space models have the ability to include many unobserved processes (e.g., survival, catch misreporting, recruitment stanzas) which can greatly increase model fit at the cost of only 1-2 additional parameters. Consequently, commonly-used diagnostic tools (e.g., model residuals) and model selection criteria (e.g., AIC, likelihood ratio tests) have been shown to be unreliable because they do not take into account the high degree of flexibility of the unobserved processes (i.e., the random effects). Thus, common diagnostic tools and model selection techniques that are routinely applied to statistical catch-at-age models may be inappropriate for state-space models. Alternative diagnostics have been proposed elsewhere, primarily based on evaluating short-term prediction error, however these methods have not yet been evaluated in a stock assessment context. The objective of the study is to address the need for general guidelines on appropriate diagnostic tools and model selection techniques for state-space stock assessment models.
This is 2-year research opportunity; second year of funds contingent on federal budget.
- PhD in quantitative fisheries, statistics, applied mathematics, marine fisheries ecology, theoretical ecology, or a related field.
- Strong quantitative skills
- Experience in quantitative modeling, stock assessment, population dynamics, statistics, and computer programming (R, Template Model Builder, AD Model Builder)
The successful candidate will be motivated and capable of working independently and collaboratively. The successful candidate will be expected to give oral presentations at a range of scientific fora, as well as publish in peer reviewed written literature.
The postdoc will be hosted by the NOAA Northeast Fisheries Science Center at the Woods Hole Laboratory in Woods Hole, MA (currently 100% remote due to COVID-19).
How to Apply
E-mail a cover letter describing your interest in the position, a CV, and the names and contact information of two references to Stacey Hansen (stacey.hansen [at] saltwaterinc.com). Inquiries regarding the position are also welcome. Review of candidates will continue until the position is filled.