Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Stata: Command for F-Test Iterations in Stata 9


From   n j cox <n.j.cox@durham.ac.uk>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Stata: Command for F-Test Iterations in Stata 9
Date   Tue, 31 Oct 2006 21:12:25 +0000

Perhaps you should quiz your "professor", not us, on
precisely what he or she has in mind and what information or
impression is behind this claim.

What you seek here is the statistical
Holy Grail, to some (and so at best elusive and at worst
non-existent). What is on offer in this terrain is
also Snake Oil, to many.

I don't think there is any command that does exactly
what you seem to want, for a variety of reasons:

0. What is "best"? Even if you focus on some single
numerical criterion for selecting best, there are many
other possible criteria with equal or greater merit,
in at least some experts' judgement.

1. Automated search avoids scientific and practical
issues of deciding what predictor choice makes most
substantive sense. There is a long history of trying
this with various stepwise approaches -- which _are_ available
as -stepwise- in Stata, from before 9 -- but there's lots
of negative opinion about those. I don't think StataCorp would be
adding -stepwise- to the repertoire now if, somehow,
it wasn't currently there.

2. Automated search doesn't usually include various
important steps, such as deciding that some initial models
and their diagnostics might indicate that we should,
for example, be adding interaction terms, transforming
predictors, or working with non-identity link functions,
let alone moving to a different model or a different
estimation method.

2. Specifically, a shotgun approach employing _multiple_
mutually dependent significance tests has as an important
side-effect that it shoots to pieces any standard
interpretation of P-values, at least to statistical
conservatives.

Nick
n.j.cox@durham.ac.uk

--------------------------------------------------------------------------------
Vaughn Elliott II

I was told by a professor that Stata 9 has the ability to iterate out a
series of F-tests automatically, and tell you which form of your model
is the best.

For example, you had a model with seven independent variables, and after
manually computing all these F-tests, you discover that one combination
of 5 of those variables is the best.

The command I referring to supposedly has Stata 9 determine which is the
optimal model, with Stata telling you which independent variables to
pick for the optimal condition.

I've searched what's new in Stata 9 on the Stata Homepage, and I've
searched the help documentation thoroughly, but I can find no mention of
such a command.

Is anyone aware of such a command ?

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index