Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Standard errors of posterior distribution point estimates

Dear all,

I have estimated my model using the CBC/HB program v5.5.3.
While I have managed to compute the standard errors from my individual-level estimates, I am actually mainly interested in reporting the posterior distribution of the parameters (i.e. mean, variance and covariances). In this regard, I was wondering whether the output provided by the software allows to compute the standard errors for the  posterior distribution of the population parameters.

Would the standard deviation computed across the iterations used for estimation after convergence is assumed to be achieved (as reported in the _alpha.csv and _covariances.csv files) be a correct measure for the standard errors?

Thanks in advance and kind regards
asked Jul 28, 2016 by lotika (455 points)

1 Answer

+1 vote
In Frequentist stats, the standard error is the standard deviation of the  population means.  It's how much on average we expect the mean for a parameter to vary if we were to repeat the sampling 1000s of times (drawing new sample 1000s of times and estimating the parameter of interest).  

The standard deviation of the alpha draws (after convergence) is the equivalent idea in Bayesian stats.  The alpha draw are successive estimates of the population means for a given parameter (e.g. part-worth utility).
answered Jul 28, 2016 by Bryan Orme Platinum Sawtooth Software, Inc. (131,390 points)