[18865] in s-news-athena
[S] Error in the predict.nls() and nonlinear least square prediction
daemon@ATHENA.MIT.EDU (Xin_Tian@bd.com)
Thu Aug 5 10:00:56 1999
From: Xin_Tian@bd.com
To: s-news@wubios.wustl.edu
Message-Id: <482567C4.004C449F.00@bd.com>
Date: Thu, 5 Aug 1999 09:50:30 -0400
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Hi, Splus users,
I want to use predict.nls( ) to get the standard errors and predicted
intervals
for the future values in nonlinear regression. I think that this function gives
me the result
from the linear approximation. But I found in this function
it use
pred$se.fit <- drop((((grad %*% fit.summary$cov) * fit.summary$sigma^2 * grad)
%*% rep.int(1,
length(coeff)))^0.5)
Should we use the std.err = s(1+f'*cov*f)^.5 ?
predplus1$se.fit <- drop(((((grad %*% fit.summary$cov)*grad) %*%rep.int(1,
length(coeff)))+1)^0.5)*fit.summary$sigma
And I also found that in the menu
Statistics--Regression--Nonlinear--Predict diaglog box,
the confidence level is not (1-alpha) but is (1-2*alpha), so if you enter .95,
it will give you 90% CI.
Is that ture ?
I use Splus4.5 or 2000 professional for windows.
Any help will be greatly appreciated.
Xin Tian
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
