Is presenting ROPE's around fitted/predicted draws a valid approach for presentation of results?

[JLC]

While I have used frequentist mixed model for the last several years, I have only delved into Bayesian approaches over the last few. Bayesian analysis and reporting are still somewhat rare in the fields I work in, so I’m unsure of the validity of some of my reporting techniques.

I typically provide a regression summary table with mean/median coefficient/parameter estimates and an 89% CI.

In cases of logistic models (binomial, bernoulli, ordinal), I have been using the fantastic tidybayes package to generate fitted draws (response probabilities) and then with guidance from the research team designating ROPE boundaries as a way of presenting modelling results in an intuitive manner.

Is this approach valid?

For example:
This is from a bernoulli family model:

To produce this plot, I use add_fitted_draws to add predicted response probabilities and then use compare_levels to create the probability differences. Then I manually add the ROPE boundaries to the figure and accompany the figure with a table of percentages of the probability difference distributions inside the ROPE computed from in bayestestR.

[avehtari]

@JLC

In cases of logistic models (binomial, bernoulli, ordinal), I have been using the fantastic tidybayes package to generate fitted draws (response probabilities) and then with guidance from the research team designating ROPE boundaries as a way of presenting modelling results in an intuitive manner.

Is this approach valid?

Arbitrary ROPE boundaries can be a distraction and misleading, but if the ROPE boundaries are meaningful and provide reasonable interpretation for what difference size is interesting and what is not, then they are useful and valid.

It’s good that you still show the posterior density estimate as it contains more information than if you would only show some interval or probability of being inside the ROPE.