Expert assessments of hypotheses concerning the etiological agent(s) of Stony Coral Tissue Loss Disease collected during a rapid prototyping project

This dataset is from expert elicitation of a panel of 15 experts with knowledge of stony coral tissue loss disease (SCTLD) and its impacts on coral reefs. We gathered this group of 15 participants with diverse expertise who had previously studied SCTLD including at universities and various government agencies as microbiologists, pathologists, disease ecologists, population ecologists, and coral experts. Participants represented marine disease experts in Florida, Hawaii, South Carolina, and the US Virgin Islands. We then used a rapid prototyping approach (Runge and Converse, 2017) to elicit, structure, and evaluate existing knowledge regarding the etiology of SCTLD. Our approach began with eliciting hypotheses about the cause of SCTLD from the expert panel over the course of four meetings, conducted via videoconference between 8/13/2021 and 11/09/2021.Each expert was assigned a unique identification number ('identity') that was displayed with their responses in place of experts’ names to keep results anonymous. After the first meeting, we asked each expert to identify 2 – 6 hypotheses and associated predictions for the causative agent(s) of SCTLD. We consolidated the experts’ hypotheses and removed redundant ones, resulting in ten final hypotheses for the etiology of SCTLD.We considered two elicitation approaches that hereafter we refer to as method 1 (M1) and method 2 (M2).

M1 was intended to get an overall assessment of the state of knowledge across experts regarding the cause of SCTLD. For M1, we asked the experts to allocate 100 points across the 10 hypotheses based on the weight of evidence that they believe existed in support of each hypothesis. Experts were allowed to use their own knowledge and any sources of information available to them, but not to confer with each other regarding their scores. Following discussions and based on the input of the experts, we revised the definition of the hypotheses. We then asked the experts to revise their estimates, if needed, and used these revised estimates (Round 2 or R2 within the dataset) for the M1 analyses.

The second approach, M2, was developed to provide a framework for deriving belief weights for the hypotheses based on assessments of individual studies. We initially asked panel members to select four studies relevant to the etiology of SCTLD. From these, we selected the five studies that received the most votes from the experts including: Aeby et al., 2019; Kellogg and Evans, 2021; Landsberg et al., 2020; Ushijima et al., 2020; Work et al., 2021. For all studies, we provided background information and/or the associated publication, and authors associated with these studies either discussed the results directly or provided written comments about the studies to the expert panelists. Under the M2 approach, experts were asked to evaluate whether hypothesis h was supported or not by a given study s. The experts were asked to allocate 100 points between two options for each hypothesis and for each study: “yes” there is supportive evidence for hypothesis h, or “no” there is no support for hypothesis h according to study s. For example, “yes: 80; no: 20” (hereafter noted as “80/20”) for hypothesis h indicates that expert e considered that study s provided strong supportive evidence for hypothesis h (i.e., there was an 80% chance that the study supports hypothesis h and a 20% chance that it did not). If the study was irrelevant with regards to hypothesis h (i.e., the study could not by its design provide evidence for or against the hypothesis), the experts entered “Not Applicable” (“NA”).