Hi Joe,
The priors are not transformed, so they refer to the transformed variable (that is,
log(tariff), rather than tariff). This is because the normal priors on the observation
make more sense when the distribution is itself normal, which is usually after the log
transformation. Hope that helps.
Cheers,
matt.
~~~~~~~~~~~
Matthew Blackwell
PhD Candidate
Institute for Quantitative Social Science
Department of Government
Harvard University
url:
http://www.mattblackwell.org
On Thursday, September 8, 2011 at 3:04 PM, J. Dieleman wrote:
Hello,
I’m wondering if someone could provide a little bit of information about how
observation-level priors work for a variable that is also being transformed?
I’ll apply my question to the running example found in the AMELIA documentation (Honaker,
King, and Blackwell, 2011). Here, using the option logs=”tariff” seems appropriate as
discussed on pages 18 and 19. Imagine now that I also want to include observational-level
priors about tariff, as discussed on pages 24-26. Will my prior also be converted into
log-space? I assume without the transformation option, my prior is assumed to be normally
distributed, based on the confidence interval supplied by my five column priors matrix.
What is the distribution now that I’ve also specified the log transformation on the
tariff? To avoid this, should I simply not transform tariff?
Thanks for any help that anybody can offer.
Joe Dieleman
University of Washington
Institute for Health Metrics and Evaluation