8/

A clear introduction on intrinsic #uncertainty in #ComputationalScience modelling (and epistemic humility) by Richard de Neufville (MIT #OpenCourseWare, Risk and Decision Analysis):

- Flaw of Averages: why central #OneDimensionalMetrics in non-trivial systems may mislead
https://ocw.mit.edu/courses/ids-333-risk-and-decision-analysis-fall-2021/resources/unit-2-flaw-o-f-averages-video-5/

- unexpected surprises in the real world
https://ocw.mit.edu/courses/ids-333-risk-and-decision-analysis-fall-2021/resources/unit-2-forecast-always-wrong-video-1/

- Porcupine Graphs: how "often experts don't learn", think "they know better" and miss evident mental biases
https://ocw.mit.edu/courses/ids-333-risk-and-decision-analysis-fall-2021/resources/unit-2-porcupine-graphic-video-3/

3/

A partial illustration on #HiddenUncertainty influencing attempts to compress #complexity into #OneDimensionalMetrics may be "a large-scale crowdsourced research effort involving 73 teams" which found "that analyzing the same hypothesis with the same data can lead to substantial differences in statistical estimates and substantive conclusions" [2]

"Instead of convergence, teams’ results varied greatly, ranging from large negative to large positive effects"

See Fig 1
https://www.pnas.org/doi/10.1073/pnas.2203150119#fig01