Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: Test of independence of errors in logistic regression
From
Nick Cox <[email protected]>
To
[email protected]
Subject
Re: st: Test of independence of errors in logistic regression
Date
Thu, 3 Mar 2011 10:31:51 +0000
I agree with Maarten, naturally. But I think a little more can be
said, especially if we move away from the strange idea that model
assessment always requires formal tests.
Sometimes looking at residuals can throw light on implicit
independence assumptions. While this is standard for regression, it
appears less so for logistic regression. A bad reason is tribal or
personal habit, e.g. that logistic people are often "table people"
rather than "graph people". A good reason is that the usual kind of
plots are often difficult to decode. So, you may need some extra
technique for looking carefully at residuals.
For example, following the urging of Tony Lachenbruch, a specific
program -rbinplot- was added to my -modeldiag- package in 2010:
gr0009_1 from http://www.stata-journal.com/software/sj10-1
SJ10-1 gr0009_1. Update: Speaking Stata: Graphing model... / Update:
Speaking Stata: Graphing model diagnostics / by Nicholas J. Cox, Durham
University, UK / Support: [email protected] / After installation,
type help anovaplot, indexplot, / modeldiag, ofrtplot, ovfplot, qfrplot,
gr0009 from http://www.stata-journal.com/software/sj4-4
SJ4-4 gr0009. Graphing model diagnostics / Graphing model diagnostics /
by Nicholas J. Cox, University of Durham, U.K. / Support:
[email protected] / After installation, type help modeldiag
On Thu, Mar 3, 2011 at 10:11 AM, Maarten buis <[email protected]> wrote:
> --- On Thu, 3/3/11, Kouji Asakura wrote:
>> Is there a test available in Stata to test for independence
>> of errors?
>
> I do not think that is possible. Where would the information
> come from? Errors are not directly observed, but in logistic
> regression the dependent variable is also not directly
> observed. So now there is a double "moddeling layer" between
> the data and the variable you want to investigate. On top
> of that, a test of independence seems to me like a very
> non-specific and thus not very powerful test even if we
> directly observed the variable. All this together means that
> I do not think that such a test is possible.
*
* 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/