re: st: RE: RE: When to use ML or bootrapping

[Christopher Baum <kit.baum@bc.edu>]

<>
Amy said

For example, if ML can be used for linear models, why would you use ML instead of OLS?  And yes, I realize just how clueless that must sound.


You wouldn't.  MLE makes specific distributional assumptions. OLS point estimates are optimal even in the presence of non-iid errors, which is why we can use robust, HAC, cluster-robust VCE estimates to consistently analyse the precision of OLS estimates. That carries over to fancier linear techniques such as IV and IV-GMM. MLE estimates are asy efficient (in terms of reaching the Cramer-Rao lower bound of precision) if your distributional assumptions are appropriate--and thus the best that can be had---but not to be trusted if they are not. As OLS will also reach the CRLB with iid errors, and involves less computation, it is to be preferred in that circumstance. 

Kit

Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                              An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
   An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/