Have an idea?

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

Calculating standard errors in CBC/HB

I'm relatively new to CBC/HB, but I've always been taught that standard errors were a good thing even in the world of Bayesian statistics. It doesn't look like the output of CBC/HB includes this statistic. Why doesn't it? And if I needed it, how do I calculate it with my raw utilities?
asked Aug 18, 2016 by JKincaid Bronze (1,035 points)

1 Answer

0 votes
The CBC/HB software outputs standard deviations of the draws for respondents for each utility level in the model (but only if you do not request "draws" as an advanced option).  That's the equivalent precision measure in the Bayesian sense at the individual level.  

For the standard deviation of the population means (the standard error from the frequentist world), the software doesn't provide this automatically, but you can compute it pretty easily from the output.  Open the _alpha.csv file that contains the population mean estimates at each iteration (the estimate of alpha at each iteration).  Ignore the first few thousand iterations prior to convergence.  Then, compute the standard deviation of the alpha draws after convergence for each attribute level.  This gives you the standard deviation of the population mean estimates for each parameter.
answered Aug 19, 2016 by Bryan Orme Platinum Sawtooth Software, Inc. (184,140 points)