[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: Factor variables and mim

From (Yulia Marchenko, StataCorp LP)
Subject   RE: st: Factor variables and mim
Date   Wed, 12 Aug 2009 14:46:44 -0500

Tony Lachenbruch <> had a follow-up question
for the discussion of -mi estimate-'s -cmdok- option:

> Does this imply that I can run a stepwise regression or LARS by using cmdok?

Stepwise estimation as performed by the -stepwise- command presently cannot be
used with -mi estimate-.  -stepwise- is a prefix command and it is not allowed
with -mi estimate-.  That is, typing

   . stepwise : mi estimate : ...

will result in error.  More generally, -mi estimate- cannot be used with any
prefix commands other than -xi-, but we do plan on expanding the list of
allowed prefix commands as part of future development.

The user-written command -lars- cannot be used with -mi estimate- even if the
-cmdok- option is used because of two technical issues:

  1.  -lars- does not save the estimated variance-covariance matrix;
  2.  -lars- is an r-class command and not an e-class command

In general, the second issue can be resolved by writing an e-class wrapper
which calls an r-class command and resaves the appropriate results in -e()-;
see for the list
of required -e()- results.  The first issue, however, is not easily resolved.

I haven't seen much discussion in the literature about applications of the
Lasso method, which is implemented in -lars-, and, more generally,
variable-selection methods to multiply-imputed data.  This is something that
we may look into in the future.

*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index