s-news
[Top] [All Lists]

FW: Help interpreting results of bias tests in lmRob

To: <s-news@lists.biostat.wustl.edu>
Subject: FW: Help interpreting results of bias tests in lmRob
From: "J. Philip Miller" <JPhilipMiller@WUstl.edu>
Date: Thu, 18 May 2006 15:57:28 -0500
Organization: WUMS Biostatistics
Thread-index: AcZ6u8xfktnyu+l5QnqzjugKIWCt/wAAc31g
I am sorry, I mistakenly rejected this posting.

-phil


-----Original Message-----
From: Schwarz,Paul [mailto:PSchwarz@gcrinsight.com] 
Sent: Thursday, May 18, 2006 3:44 PM
To: s-news@lists.biostat.wustl.edu
Subject: Help interpreting results of bias tests in lmRob

S-News readers,
 
I'm a bit confused about how to interpret (and then what to do) when the
results of the bias tests, which are part of the lmRob() summary output,
are significant. Clearly when the tests are not significant, then the
interpretation is that there is no significant bias in the parameter
estimates of the least squares solution relative to the robust solution
estimates (at least this is what the manual says). So, when the tests
are significant, obviously the converse must be true, but how do I use
the information? Is there an estimate of the bias? Should I make some
sort of correction? For example, I got the following message for one of
the models that I estimated using lmRob():

> test.lmRob( m2 )

Test for Bias:
            Statistics  P-value 
 M-estimate       29.4 0.044166
LS-estimate       52.8 0.000028
    Significant test at level 10%. The bias is high, and inference based
on final estimates is not recommended. Use initial estimates as
        exploratory tools. in: print(object)


How do I interpret these results? How can I compare the initial
estimates vs. the final estimates? I'm not very familiar with these
robust methods, but it looks like they could be quite useful.  

Thank you for your time and patience,

-Paul Schwarz


<Prev in Thread] Current Thread [Next in Thread>
  • FW: Help interpreting results of bias tests in lmRob, J. Philip Miller <=