Population viability analysis (PVA) is a widely used tool that applies demographic data in simulation frameworks to assess extinction risk for species or populations. It is used in diverse conservation applications, including evaluating management effectiveness, relative risk of threats, and potential changes to protective status (Beissinger & McCullough, 2002), and can be a critical tool for making decisions with imperfect knowledge of the system state, often on limited timelines (Meine et al., 2006).

Chaudhary and Oli (2020) recently developed a framework to appraise the quality of PVAs based on the presence of essential background, model, and analysis components. They evaluated 160 published PVAs and reported a decline in the quality of PVAs over time (1990−2017). We agree PVA studies should report unambiguous descriptions of their essential components (Table 1 in Chaudhary and Oli) and explicitly state the model's biological and statistical assumptions. The need for increased transparency in PVAs is evident. Morrison et al. (2016) reported that only 50% of PVAs published in peer-reviewed and gray literature were both reproducible and repeatable. Further, in an examination of 67 studies that used matrix population models (widely used in PVAs), Kendall et al. (2019) reported that models frequently contained misspecification errors. Given the rapid advancement of simulation techniques, updated guidance for PVA construction is warranted.

However, we believe the essential PVA components identified by Chaudhary and Oli contain a critical omission: the decision context in which the PVA was created and its usefulness in that context. Quality and utility are not mutually exclusive; however, some models that do not meet idealized quality standards might still be valuable because they are useful and represent the best available science for a given decision context (hereafter, decision-support models). The definition of quality for decision-support models should be different than models developed for the purpose of learning (hereafter, heuristic models) and should incorporate how useful the model was, despite information gaps. We further argue that assessment questions should be used prospectively to guide modeling projects, rather than for retrospective comparison of model quality.