Blockquote
No doubt the modeling here was done in a frequentist framework (if any), but it seems to me that a Bayesian view can be adopted here, in which we ask what priors would have yielded the curves as drawn, under a reasonable Bayesian model. Here ‘reasonable’ would impose a certain amount of smoothness; this could become the parameter of a family of priors. Presumably the tails (and even the negative probabilities plotted) offer useful information about the prior outside the range of the data.
Robert Matthews adapted what he calls “Bayesian Analysis of Credibility” from IJ Good and his “method of imaginary results.” I posted links to key papers by Matthews in this thread. In some more recent work with Leonhart Held, they adapt their method to meta-analysis.
I prefer calculating both skeptic and advocate priors regardless of “significance”, giving a probability interval, leading to Robust Bayesian analysis and imprecise probability theory.