How can/should we compute confidence intervals for results from HB/MNL models, be they utilities, simulated preference shares, willingness-to-pay calculations, or what have you? One fundamental need is to work with the HB draws, not the posterior means. This talk will explore using lower-level draws vs. upper-level draws, and dig into whether and how sampling variance is or isn’t incorporated into the variance of the HB draws. Some computational issues will be included.
