Do forecasters misunderstand logarithmic versus natural scale elicitation?

Investigate whether human forecasters submitting probabilistic predictions on the Metaculus platform during the 2022 mpox forecasting questions were susceptible to misunderstanding the difference between providing forecasts on a base-10 logarithmic scale versus a natural (linear) scale.

Background

The study found that several forecasting questions were elicited on a base-10 logarithmic scale, and evidence suggested that such elicitation produced right-skewed predictive distributions with large uncertainty intervals when transformed back to the natural scale. The authors hypothesize that this could have influenced forecast behavior and performance but did not directly test whether participants misunderstood the implications of specifying predictions on a logarithmic scale.

Because no survey or randomized elicitation experiment was conducted, the authors could not ascertain whether observed distributional properties and performance differences arose from genuine forecaster intent or from confusion between logarithmic and natural scale inputs. They suggest that a randomized design posing identical questions on both scales could resolve this uncertainty.