Contents lists available at ScienceDirect Biological Conservation journal homepage: www.elsevier.com/locate/biocon Review Evaluating the assumptions of population projection models used for conservation Julia E. Earl ⁎ School of Biological Sciences, Louisiana Tech University, Ruston, LA 71272, USA ARTICLE INFO Keywords: Carrying capacity Matrix models Model assumptions Model evaluation Sensitivity analysis Simulations ABSTRACT Population projection models, such as matrix and integral projection models, are used increasingly to understand potential effects of anthropogenic stressors and inform conservation actions. However, vital rate and life history information needed to create robust population models is often missing or incomplete, making assumptions about parameters and population processes necessary. Understanding how assumptions affect results is critical, particularly if the study will be used to guide policy or management actions. I review published amphibian population projection models to determine whether model output is evaluated with population data, what as- sumptions are made, and whether sensitivity analyses are performed. I found that only 21% of published models were evaluated with population data, and most models (67%) were explored with sensitivity analyses. I then simulated the effects of four assumptions and varying population carrying capacities on model output and sensitivity results using existing matrix population models from three amphibian species with different life histories: Anaxyrus boreas, Lithobates sylvaticus, and Ambystoma maculatum. Simulations showed that changes in model output and sensitivity analyses under different assumptions depended more on the species examined than the assumption implemented. There were changes in which parameters model output was most sensitive to under all assumptions examined for all species, suggesting caution when using results if there is great uncertainty about model assumptions. Models and their parameterization should be regularly updated with new information to ensure conservation biologists are using the most robust information on potential outcomes of threats and conservation actions. 1. Introduction Population projection models, including matrix models and integral projection models, are important tools in conservation biology, because they are a low-cost way to predict future population dynamics and determine the population-level implications of anthropogenic and nat- ural stressors and conservation actions (Morris and Doak, 2002). Po- pulation projection models have been identified as key sources of in- formation for decision makers determining whether or not to list species as endangered (McGowan et al., 2017; Earl et al., 2018) and whether key recovery criteria have been met for delisting (McGowan et al., 2014). Models are a key step in management decision-making and have been used to assess the effects of most major threats to bio- diversity, including emerging pathogens (Canessa et al., 2019b - this issue-b), habitat destruction and fragmentation (e.g., Harper et al., 2015), pollution (e.g., Willson and Hopkins, 2013), and climate change (e.g., Cayuela et al., 2016). Models are thought to be most reliable in this context as a comparative tool rather than explicit predictions of the future. Further, population models can also evaluate the likely effec- tiveness of different management strategies, allowing researchers to explore a range of scenarios and compare their effectiveness and cost, facilitating the identification of cost-effective actions that enhance the probability of success (Canessa et al., 2014). Strategies that have been evaluated include population supplementation from captive popula- tions (Kissel et al., 2014), reintroductions (Gerber et al., 2018), and invasive species control methods (e.g., Govindarajulu et al., 2005) and can be used to inform management decisions through structured deci- sion making (Converse and Grant, 2019 - this issue) and adaptive management (Canessa et al., 2019a - this issue-a). All models contain assumptions that help researchers focus on key concepts and research questions but that can alter model performance. Additional complexity may better reflect biological reality, but para- meterizing complex models for species at risk is often difficult due to lack of data (Morris and Doak, 2002; Earl et al., 2018). When relying on models to make decisions about conservation actions, it is critical to understand which assumptions alter inference and therefore https://doi.org/10.1016/j.biocon.2019.06.034 Received 1 November 2018; Received in revised form 11 February 2019; Accepted 27 June 2019 ⁎ Louisiana Tech University, School of Biological Sciences, Box 3179, Ruston, LA 71272, USA. E-mail address: jearl@latech.edu. Biological Conservation 237 (2019) 145–154 0006-3207/ © 2019 Elsevier Ltd. All rights reserved. T