bayesplain.core.intervals.monte_carlo_se

bayesplain.core.intervals.monte_carlo_se(draws)[source]

Compute the standard error of a posterior mean from i.i.d. draws.

Because the draws are independent – these are exact samples from a closed-form distribution, not a Markov chain – this is simply sd / sqrt(n), with no effective-sample-size correction needed. That absence is itself the point: there is no autocorrelation to discount.

Parameters:
draws : array_like

Posterior draws.

Returns:

Monte Carlo standard error of the mean.

Return type:

float