Repository navigation
priorsense with high Pareto k #139
Description
Hi,
I have a model with high Pareto k, at least partially due to have random effects at the level of the log_lik values. What's the implication for power scaling here? When I do powerscale_plot_dens() I see the warnings for likelihood scaling, but not for prior scaling.
Cheers,
Matt
Activity
n-kall commentedon Sep 11, 2026
High pareto-k for likelihood power-scaling in priorsense means that it is difficult to reliably estimate the change in the posterior when the likelihood, in this case, is power-scaled by the given value of alpha. You can interpret the prior sensitivity regardless, but the likelihood sensitivity might be unreliable. Usually this means that the likelihood sensitivity is higher than it appears, so probably not too much of an issue. You can try with
auto_alpha_range = TRUEwhich will try to find values of alpha that keep the pareto-k lower than 0.7.