How to estimate uncertainty interval (instead of confidence interval)

[Sonnen Blume]

Hi,

I just came across an interesting article that reported 95% Uncertainty Interval using Monte Carlo method, and wondering if this is available in Stata.

Here is the reported method:
Estimates of uncertainty were modelled using a Monte Carlo-like approach where 1000 estimates of the drinking status, mean, and SD of the alcohol consumption estimates and relative risks were randomly simulated based on their respective uncertainty distributions (appendix p 5). The 2·5th and 97·5th percentiles were taken from the 1000 modelled PAF estimates to construct the 95% uncertainty intervals (UIs). All analyses were carried out using R (version 3.6.1).

[Maarten Buis]

Sounds similar to parametric Bootstrap. The advantage of such simulation approaches is that they are very flexible, and thus are applicable to a wide range of situations. The disadvantage of such simulation approaches is that they are very flexible, and thus often require custom code. It is not that bad, simulate can take care of much of the work.

One thing that worries me is that 1,000 replications seems way to low for creating such intervals: The lower bound is based on only the 0.025*1000=25th smallest estimate, and thus sensitive to outliers. That is ok for a quick first look, but for publications I would be more comfortable with something like 20,000 replications. That way the lower bound would be based on the 0.025*20000=500th smallest estimate.